refactor(core): use shared tool transform lifecycle

Centralize control dispatch and first-Step preparation in advanceToStep. Keep input delivery outside logical-Step retries and preserve queue ordering, Location handoff, context refresh, and durable settlement.

.changeset/shell-scanner-compatibility.md

@@ -0,0 +1,10 @@
+---
+"@opencode-ai/core": patch
+---
+
+Make the experimental portable shell scanner authoritative, with no Tree-sitter
+fallback. Scan common Bash and PowerShell control flow, heredocs, functions,
+expressions, quoting, and substitutions natively. Preserve existing redirect and
+declaration permission matching, and make PowerShell saved approvals cover the
+original command spelling. Parser failures remain visible without changing the
+permission engine. The default Tree-sitter path is unchanged.

bun.lock

@@ -125,7 +125,7 @@
         "@effect/platform-node": "catalog:",
         "@opencode-ai/client": "workspace:*",
         "@opencode-ai/plugin": "workspace:*",
-        "@opencode-ai/pty": "0.1.9",
+        "@opencode-ai/pty": "0.1.11",
         "@opencode-ai/schema": "workspace:*",
         "@opencode-ai/server": "workspace:*",
         "@opencode-ai/tui": "workspace:*",
@@ -364,7 +364,7 @@
         "@opencode-ai/ai": "workspace:*",
         "@opencode-ai/codemode": "workspace:*",
         "@opencode-ai/plugin": "workspace:*",
-        "@opencode-ai/pty": "0.1.9",
+        "@opencode-ai/pty": "0.1.11",
         "@opencode-ai/schema": "workspace:*",
         "@opencode-ai/util": "workspace:*",
         "@parcel/watcher": "2.5.1",
@@ -553,6 +553,20 @@
         "@typescript/native-preview": "catalog:",
       },
     },
+    "packages/latex": {
+      "name": "@opencode-ai/latex",
+      "version": "0.0.0",
+      "dependencies": {
+        "@opencode-ai/plugin": "workspace:*",
+        "@opentui/core": "catalog:",
+        "string-width": "catalog:",
+      },
+      "devDependencies": {
+        "@tsconfig/bun": "catalog:",
+        "@types/bun": "catalog:",
+        "@typescript/native-preview": "catalog:",
+      },
+    },
     "packages/merman": {
       "name": "@opencode-ai/merman",
       "version": "0.0.0",
@@ -731,6 +745,7 @@
       },
       "devDependencies": {
         "@happy-dom/global-registrator": "20.0.11",
+        "@playwright/test": "catalog:",
         "@tsconfig/node22": "catalog:",
         "@types/bun": "catalog:",
         "@types/luxon": "catalog:",
@@ -836,6 +851,7 @@
         "@opencode-ai/client": "workspace:*",
         "@opencode-ai/session-ui": "workspace:*",
         "@opencode-ai/ui": "workspace:*",
+        "@playwright/test": "catalog:",
         "@solidjs/meta": "catalog:",
         "@storybook/addon-a11y": "10.4.4",
         "@storybook/addon-docs": "10.4.4",
@@ -877,6 +893,7 @@
       "dependencies": {
         "@opencode-ai/client": "workspace:*",
         "@opencode-ai/core": "workspace:*",
+        "@opencode-ai/latex": "workspace:*",
         "@opencode-ai/merman": "workspace:*",
         "@opencode-ai/plugin": "workspace:*",
         "@opencode-ai/schema": "workspace:*",
@@ -2148,6 +2165,8 @@
 
     "@opencode-ai/httpapi-codegen": ["@opencode-ai/httpapi-codegen@workspace:packages/httpapi-codegen"],
 
+    "@opencode-ai/latex": ["@opencode-ai/latex@workspace:packages/latex"],
+
     "@opencode-ai/merman": ["@opencode-ai/merman@workspace:packages/merman"],
 
     "@opencode-ai/plugin": ["@opencode-ai/plugin@workspace:packages/plugin"],
@@ -2156,19 +2175,19 @@
 
     "@opencode-ai/protocol": ["@opencode-ai/protocol@workspace:packages/protocol"],
 
-    "@opencode-ai/pty": ["@opencode-ai/pty@0.1.9", "", { "optionalDependencies": { "@opencode-ai/pty-darwin-arm64": "0.1.9", "@opencode-ai/pty-darwin-x64": "0.1.9", "@opencode-ai/pty-linux-arm64-gnu": "0.1.9", "@opencode-ai/pty-linux-arm64-musl": "0.1.9", "@opencode-ai/pty-linux-x64-gnu": "0.1.9", "@opencode-ai/pty-linux-x64-musl": "0.1.9" }, "bin": { "opencode-pty": "bin/opencode-pty.js" } }, "sha512-9WysQgX9J3RXfZy/t/8MGqf1IGLeckyHQsT/eVnKiBnp+GnzOXtZePryj08CN/3GT2BuP5tWqMRMH0JzCMsrgg=="],
+    "@opencode-ai/pty": ["@opencode-ai/pty@0.1.11", "", { "optionalDependencies": { "@opencode-ai/pty-darwin-arm64": "0.1.11", "@opencode-ai/pty-darwin-x64": "0.1.11", "@opencode-ai/pty-linux-arm64-gnu": "0.1.11", "@opencode-ai/pty-linux-arm64-musl": "0.1.11", "@opencode-ai/pty-linux-x64-gnu": "0.1.11", "@opencode-ai/pty-linux-x64-musl": "0.1.11" }, "bin": { "opencode-pty": "bin/opencode-pty.js" } }, "sha512-Q4p0XXZWbc8FnpEJaaLqVbCdodxR9lVzaQjMH18KvjX/4m6tYfuspz03mvkN9MdmtDJ2GOZS7QgGQ7Q+RE9aWw=="],
 
-    "@opencode-ai/pty-darwin-arm64": ["@opencode-ai/pty-darwin-arm64@0.1.9", "", { "os": "darwin", "cpu": "arm64" }, "sha512-a2OZGutBdVGDO+X4t37L2K8wD1phTcLJBcNLj8j3LqCmaHbJmQyNYYFo6i8loZCTjjhb+8Wq0wMxSLny4tqdwA=="],
+    "@opencode-ai/pty-darwin-arm64": ["@opencode-ai/pty-darwin-arm64@0.1.11", "", { "os": "darwin", "cpu": "arm64" }, "sha512-Hz59ImecqeBdLQ40TknPDc9k4xWjQPaTgZ7cXzF3vclAvZiFYSM1rRdbBF6rOaOB0DBp0OFYsiaPP1ykszLsfw=="],
 
-    "@opencode-ai/pty-darwin-x64": ["@opencode-ai/pty-darwin-x64@0.1.9", "", { "os": "darwin", "cpu": "x64" }, "sha512-+vyRLwzNMzP/JFtYEIkMHxRC9Lkd7sUeuAqI5medtdgZmvTwS3NdkHbq6zQQG3rTQ9asObPKilvcpQDf+LYHRQ=="],
+    "@opencode-ai/pty-darwin-x64": ["@opencode-ai/pty-darwin-x64@0.1.11", "", { "os": "darwin", "cpu": "x64" }, "sha512-TPpA+FZ08BXtTcOeqe0FEJitqLld6Nl46UizcSmQCTRM22xOKPar6OoxHGYtqdFCaFk43a+hk/0GWpV7mcEQEg=="],
 
-    "@opencode-ai/pty-linux-arm64-gnu": ["@opencode-ai/pty-linux-arm64-gnu@0.1.9", "", { "os": "linux", "cpu": "arm64" }, "sha512-vjNKhCsw6mI6w+9bITCyCmb5nS+XiWA8xMGn02GOKE9AahcVX9JIbjUF1zAmPXZ1QD7nxWoanh4FvnqukbYPHw=="],
+    "@opencode-ai/pty-linux-arm64-gnu": ["@opencode-ai/pty-linux-arm64-gnu@0.1.11", "", { "os": "linux", "cpu": "arm64" }, "sha512-PTU9Ss5a5pApw6IeVvjjbFPuui2oKMoTQ/nY+K1+idLpgMeQHXk2URJbQqetqJxH4OLL5eSutDeWvpkweW0tTw=="],
 
-    "@opencode-ai/pty-linux-arm64-musl": ["@opencode-ai/pty-linux-arm64-musl@0.1.9", "", { "os": "linux", "cpu": "arm64" }, "sha512-xpS0N6/uEiJPabv6Ib9BpOlfyZdcUES7sMVa4bCrgcxy6y4bnQMeGF/Ju4u7LBABD/rjBKZM0XhPMnvZ6qoX1w=="],
+    "@opencode-ai/pty-linux-arm64-musl": ["@opencode-ai/pty-linux-arm64-musl@0.1.11", "", { "os": "linux", "cpu": "arm64" }, "sha512-NFZ2LLfEaO6858cYtEwfQKda/HnCrPR0WflnWoDllHmdY12umeHkeE8DnZQK48tziXxvEACspItXmleiAMAg3g=="],
 
-    "@opencode-ai/pty-linux-x64-gnu": ["@opencode-ai/pty-linux-x64-gnu@0.1.9", "", { "os": "linux", "cpu": "x64" }, "sha512-N1Dx8rOLkpJd2DSetZZW9dPnpL3mxbRQagk/7K7TbS7M1rDnLazhq8/T9vulLET8KTwu6lmWcZYuSDfPLNWXnA=="],
+    "@opencode-ai/pty-linux-x64-gnu": ["@opencode-ai/pty-linux-x64-gnu@0.1.11", "", { "os": "linux", "cpu": "x64" }, "sha512-2Wbko2tFkgTmY6ceB+QMA3E+omaRd5IBBFjJoikMxkc7n75TXYQVz09tCC0pPW+flGvApFQ2YcIkSotdLwit+A=="],
 
-    "@opencode-ai/pty-linux-x64-musl": ["@opencode-ai/pty-linux-x64-musl@0.1.9", "", { "os": "linux", "cpu": "x64" }, "sha512-osb203LrlwXpQKQABrCfD5QRVK9Ajiq5atulC/VuMye6KkErcetDnjQvobjhaEjgnSEGPCH4U9sM84okGqip5Q=="],
+    "@opencode-ai/pty-linux-x64-musl": ["@opencode-ai/pty-linux-x64-musl@0.1.11", "", { "os": "linux", "cpu": "x64" }, "sha512-PF7vbOsSOVbRSo11pOOmJq/Vp34Ww7Xoo8rUeMSAoG1tJSIp2SXFHCRKEGH9H4KxGUfpRHQlDBV1HLz8onnyJQ=="],
 
     "@opencode-ai/schema": ["@opencode-ai/schema@workspace:packages/schema"],
 

nix/hashes.json

@@ -1,8 +1,8 @@
 {
   "nodeModules": {
-    "x86_64-linux": "sha256-QWLIdvu985FH5I9cZJOAuoeFeXU+4Jx9RzBB9RPoeeQ=",
-    "aarch64-linux": "sha256-SSzGD5hMj2vFvyw+dUPR9g/ZH6qhs0ZyZ/DnltZt3N8=",
-    "aarch64-darwin": "sha256-CeFUxiV+e8pKho+YcSclC3soQBogoxNMxwyIMztAExU=",
-    "x86_64-darwin": "sha256-FYwcACzU72y0+KtOpFfU7ndak8vMasqMgd5NLS6+XtY="
+    "x86_64-linux": "sha256-NV1PD2fCgWEKsr9kR0pV9jgkC400dzoF7/DnI/fY5yI=",
+    "aarch64-linux": "sha256-TDTdwE0mcHLrrKPDwPPBk3qIDl/PXJrLX6Zbwp7EH3I=",
+    "aarch64-darwin": "sha256-6MEoiV1UKAWgC7C6PR4USCP/LLZXROfBfPg6sb2VVWg=",
+    "x86_64-darwin": "sha256-8JV6YVZFq1BC++zpARxBWhQ+wuNJrWgTZJ6jfQhDybs="
   }
 }

packages/ai/src/protocols/anthropic-messages.ts

@@ -406,6 +406,7 @@ const AnthropicEvent = Schema.Struct({
 type AnthropicEvent = Schema.Schema.Type<typeof AnthropicEvent>
 
 interface ParserState {
+  readonly providerMetadataKey: string
   readonly tools: ToolStream.State<number>
   readonly reasoningSignatures: Readonly<Record<number, string>>
   readonly usage?: Usage
@@ -440,18 +441,18 @@ const cacheControl = (breakpoints: Cache.Breakpoints, cache: CacheHint | undefin
   return Cache.ttlBucket(cache.ttlSeconds) === "1h" ? EPHEMERAL_1H : EPHEMERAL_5M
 }
 
-const anthropicMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ anthropic: metadata })
+const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })
 
-const signatureFromMetadata = (metadata: ProviderMetadata | undefined): string | undefined => {
-  const anthropic = metadata?.anthropic
-  if (!ProviderShared.isRecord(anthropic)) return undefined
-  return typeof anthropic.signature === "string" ? anthropic.signature : undefined
+const signatureFromMetadata = (metadata: ProviderMetadata | undefined, key: string): string | undefined => {
+  const provider = metadata?.[key]
+  if (!ProviderShared.isRecord(provider)) return undefined
+  return typeof provider.signature === "string" ? provider.signature : undefined
 }
 
-const redactedDataFromMetadata = (metadata: ProviderMetadata | undefined): string | undefined => {
-  const anthropic = metadata?.anthropic
-  if (!ProviderShared.isRecord(anthropic)) return undefined
-  return typeof anthropic.redactedData === "string" ? anthropic.redactedData : undefined
+const redactedDataFromMetadata = (metadata: ProviderMetadata | undefined, key: string): string | undefined => {
+  const provider = metadata?.[key]
+  if (!ProviderShared.isRecord(provider)) return undefined
+  return typeof provider.redactedData === "string" ? provider.redactedData : undefined
 }
 
 const lowerTool = (breakpoints: Cache.Breakpoints, tool: ToolDefinition, inputSchema: JsonSchema): AnthropicTool => ({
@@ -511,13 +512,16 @@ const serverToolResultType = (name: string): AnthropicServerToolResultType | und
   return undefined
 }
 
-const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult")(function* (part: ToolResultPart) {
+const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult")(function* (
+  part: ToolResultPart,
+  providerMetadataKey: string,
+) {
   const wireType = serverToolResultType(part.name)
   if (!wireType)
     return yield* invalid(`Anthropic Messages does not know how to round-trip server tool result for ${part.name}`)
   // Prefer the provider-owned replay payload; fall back to the result value for
   // histories constructed directly from provider events.
-  const payload = part.providerMetadata?.anthropic?.["result"] ?? part.result.value
+  const payload = part.providerMetadata?.[providerMetadataKey]?.["result"] ?? part.result.value
   return {
     type: wireType,
     tool_use_id: scrubToolCallID(part.id),
@@ -804,6 +808,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
   breakpoints: Cache.Breakpoints,
 ) {
   const messages: AnthropicMessage[] = []
+  const providerMetadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)
 
   for (const [index, message] of request.messages.entries()) {
     if (message.role === "system") {
@@ -849,8 +854,8 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
         if (part.type === "reasoning") {
           // A signature marks visible thinking; only signature-less parts carrying
           // redactedData round-trip as opaque redacted_thinking blocks.
-          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata)
-          const redactedData = redactedDataFromMetadata(part.providerMetadata)
+          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata, providerMetadataKey)
+          const redactedData = redactedDataFromMetadata(part.providerMetadata, providerMetadataKey)
           if (signature === undefined && redactedData !== undefined) {
             content.push({ type: "redacted_thinking", data: redactedData })
             continue
@@ -879,7 +884,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
           continue
         }
         if (part.type === "tool-result" && part.providerExecuted) {
-          content.push(yield* lowerServerToolResult(part))
+          content.push(yield* lowerServerToolResult(part, providerMetadataKey))
           continue
         }
         return yield* invalid(
@@ -1069,7 +1074,7 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
 // inclusive `inputTokens` the rest of the contract expects. Extended
 // thinking tokens are included in `output_tokens`; newer responses also
 // expose that subset through `output_tokens_details.thinking_tokens`.
-const mapUsage = (usage: AnthropicUsage | undefined): Usage | undefined => {
+const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string): Usage | undefined => {
   if (!usage) return undefined
   const nonCached = usage.input_tokens ?? undefined
   const cacheRead = usage.cache_read_input_tokens ?? undefined
@@ -1083,7 +1088,7 @@ const mapUsage = (usage: AnthropicUsage | undefined): Usage | undefined => {
     cacheWriteInputTokens: cacheWrite,
     reasoningTokens: usage.output_tokens_details?.thinking_tokens,
     totalTokens: ProviderShared.totalTokens(inputTokens, usage.output_tokens, undefined),
-    providerMetadata: { anthropic: usage },
+    providerMetadata: { [providerMetadataKey]: usage },
   })
 }
 
@@ -1092,7 +1097,7 @@ const mapUsage = (usage: AnthropicUsage | undefined): Usage | undefined => {
 // field prefers `right` when defined, falls back to `left`. `inputTokens` is
 // recomputed from the merged breakdown so the inclusive total stays
 // consistent with `nonCached + cacheRead + cacheWrite`.
-const mergeUsage = (left: Usage | undefined, right: Usage | undefined) => {
+const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerMetadataKey: string) => {
   if (!left) return right
   if (!right) return left
   const nonCachedInputTokens = right.nonCachedInputTokens ?? left.nonCachedInputTokens
@@ -1110,7 +1115,9 @@ const mergeUsage = (left: Usage | undefined, right: Usage | undefined) => {
     reasoningTokens,
     totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
     providerMetadata: {
-      anthropic: mergeJsonRecords(left.providerMetadata?.["anthropic"], right.providerMetadata?.["anthropic"]) ?? {},
+      [providerMetadataKey]:
+        mergeJsonRecords(left.providerMetadata?.[providerMetadataKey], right.providerMetadata?.[providerMetadataKey]) ??
+        {},
     },
   })
 }
@@ -1128,7 +1135,7 @@ const SERVER_TOOL_RESULT_NAMES: Record<AnthropicServerToolResultType, string> =
 
 const isServerToolResultType = (type: string): type is AnthropicServerToolResultType => type in SERVER_TOOL_RESULT_NAMES
 
-const serverToolResultEvent = (block: AnthropicStreamBlock): LLMEvent | undefined => {
+const serverToolResultEvent = (block: AnthropicStreamBlock, providerMetadataKey: string): LLMEvent | undefined => {
   if (!block.type || !isServerToolResultType(block.type)) return undefined
   const errorPayload =
     typeof block.content === "object" && block.content !== null && "type" in block.content
@@ -1142,7 +1149,7 @@ const serverToolResultEvent = (block: AnthropicStreamBlock): LLMEvent | undefine
     providerExecuted: true,
     // The complete payload is irreducible provider replay state: subsequent
     // stateless requests must round-trip the typed result block verbatim.
-    providerMetadata: anthropicMetadata({ blockType: block.type, result: block.content }),
+    providerMetadata: providerMetadata(providerMetadataKey, { blockType: block.type, result: block.content }),
   })
 }
 
@@ -1151,8 +1158,8 @@ type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
 const NO_EVENTS: StepResult["1"] = []
 
 const onMessageStart = (state: ParserState, event: AnthropicEvent): StepResult => {
-  const usage = mapUsage(event.message?.usage)
-  return [usage ? { ...state, usage: mergeUsage(state.usage, usage) } : state, NO_EVENTS]
+  const usage = mapUsage(event.message?.usage, state.providerMetadataKey)
+  return [usage ? { ...state, usage: mergeUsage(state.usage, usage, state.providerMetadataKey) } : state, NO_EVENTS]
 }
 
 const onContentBlockStart = (
@@ -1204,14 +1211,16 @@ const onContentBlockStart = (
   if (block.type === "thinking" && block.thinking !== undefined) {
     const events: LLMEvent[] = []
     const id = `reasoning-${event.index ?? 0}`
-    const providerMetadata =
-      block.signature === undefined ? undefined : anthropicMetadata({ signature: block.signature })
-    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, providerMetadata)
+    const metadata =
+      block.signature === undefined
+        ? undefined
+        : providerMetadata(state.providerMetadataKey, { signature: block.signature })
+    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, metadata)
     return [
       {
         ...state,
         lifecycle: block.thinking
-          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, providerMetadata)
+          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, metadata)
           : lifecycle,
         reasoningSignatures:
           event.index === undefined || block.signature === undefined
@@ -1234,14 +1243,14 @@ const onContentBlockStart = (
           state.lifecycle,
           events,
           `reasoning-${event.index ?? 0}`,
-          anthropicMetadata({ redactedData: block.data }),
+          providerMetadata(state.providerMetadataKey, { redactedData: block.data }),
         ),
       },
       events,
     ]
   }
 
-  const result = serverToolResultEvent(block)
+  const result = serverToolResultEvent(block, state.providerMetadataKey)
   if (!result) return [state, NO_EVENTS]
   const events: LLMEvent[] = []
   return [{ ...state, lifecycle: Lifecycle.stepStart(state.lifecycle, events) }, [...events, result]]
@@ -1321,7 +1330,7 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
         Lifecycle.textEnd(state.lifecycle, events, `text-${event.index}`),
         events,
         `reasoning-${event.index}`,
-        signature === undefined ? undefined : anthropicMetadata({ signature }),
+        signature === undefined ? undefined : providerMetadata(state.providerMetadataKey, { signature }),
       )
   events.push(...resultEvents)
   const reasoningSignatures = { ...state.reasoningSignatures }
@@ -1333,7 +1342,7 @@ const onMessageDelta = (
   state: ParserState,
   event: AnthropicEvent & { readonly delta?: AnthropicStreamDelta },
 ): StepResult => {
-  const usage = mergeUsage(state.usage, mapUsage(event.usage))
+  const usage = mergeUsage(state.usage, mapUsage(event.usage, state.providerMetadataKey), state.providerMetadataKey)
   return [
     {
       ...state,
@@ -1346,7 +1355,7 @@ const onMessageDelta = (
         providerMetadata:
           event.delta?.stop_sequence === null || event.delta?.stop_sequence === undefined
             ? undefined
-            : anthropicMetadata({ stopSequence: event.delta.stop_sequence }),
+            : providerMetadata(state.providerMetadataKey, { stopSequence: event.delta.stop_sequence }),
       },
     },
     NO_EVENTS,
@@ -1472,7 +1481,8 @@ export const protocol = Protocol.make({
   },
   stream: {
     event: Protocol.jsonEvent(AnthropicEvent),
-    initial: () => ({
+    initial: (request) => ({
+      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
       tools: ToolStream.empty<number>(),
       reasoningSignatures: {},
       lifecycle: Lifecycle.initial(),

packages/ai/src/protocols/bedrock-converse.ts

@@ -258,19 +258,21 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
     tool: (name) => ({ tool: { name } }) as const,
   })
 
-const bedrockMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ bedrock: metadata })
+const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })
 
-const reasoningSignature = (part: ReasoningPart) => {
-  const bedrock = part.providerMetadata?.bedrock
+const reasoningSignature = (part: ReasoningPart, providerMetadataKey: string) => {
+  const metadata = part.providerMetadata?.[providerMetadataKey]
   return (
     part.encrypted ??
-    (ProviderShared.isRecord(bedrock) && typeof bedrock.signature === "string" ? bedrock.signature : undefined)
+    (ProviderShared.isRecord(metadata) && typeof metadata.signature === "string" ? metadata.signature : undefined)
   )
 }
 
-const reasoningRedactedData = (part: ReasoningPart) => {
-  const bedrock = part.providerMetadata?.bedrock
-  return ProviderShared.isRecord(bedrock) && typeof bedrock.redactedData === "string" ? bedrock.redactedData : undefined
+const reasoningRedactedData = (part: ReasoningPart, providerMetadataKey: string) => {
+  const metadata = part.providerMetadata?.[providerMetadataKey]
+  return ProviderShared.isRecord(metadata) && typeof metadata.redactedData === "string"
+    ? metadata.redactedData
+    : undefined
 }
 
 const lowerToolCall = (part: ToolCallPart): BedrockToolUseBlock => ({
@@ -318,6 +320,7 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
   breakpoints: BedrockCache.Breakpoints,
 ) {
   const messages: BedrockMessage[] = []
+  const providerMetadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)
 
   for (const message of request.messages) {
     if (message.role === "system") {
@@ -365,8 +368,8 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
           continue
         }
         if (part.type === "reasoning") {
-          const signature = reasoningSignature(part)
-          const redactedData = reasoningRedactedData(part)
+          const signature = reasoningSignature(part, providerMetadataKey)
+          const redactedData = reasoningRedactedData(part, providerMetadataKey)
           if (signature === undefined && redactedData !== undefined) {
             content.push({ reasoningContent: { redactedContent: redactedData } })
             continue
@@ -466,7 +469,7 @@ const mapFinishReason = (reason: string): FinishReason => {
 
 // AWS reports inputTokens separately from cache reads and writes.
 // Bedrock does not break reasoning out of outputTokens for current models.
-const mapUsage = (usage: BedrockUsageSchema | undefined): Usage | undefined => {
+const mapUsage = (usage: BedrockUsageSchema | undefined, providerMetadataKey: string): Usage | undefined => {
   if (!usage) return undefined
   const inputTokens = ProviderShared.sumTokens(
     usage.inputTokens,
@@ -480,11 +483,12 @@ const mapUsage = (usage: BedrockUsageSchema | undefined): Usage | undefined => {
     cacheReadInputTokens: usage.cacheReadInputTokens,
     cacheWriteInputTokens: usage.cacheWriteInputTokens,
     totalTokens: ProviderShared.totalTokens(inputTokens, usage.outputTokens, usage.totalTokens),
-    providerMetadata: { bedrock: usage },
+    providerMetadata: { [providerMetadataKey]: usage },
   })
 }
 
 interface ParserState {
+  readonly providerMetadataKey: string
   readonly tools: ToolStream.State<number>
   // Bedrock splits the finish into `messageStop` (carries `stopReason`) and
   // `metadata` (carries usage). Hold the terminal event in state so `onHalt`
@@ -541,20 +545,14 @@ const step = (state: ParserState, event: BedrockEvent) =>
       const reasoning = event.contentBlockDelta.delta.reasoningContent
       const events: LLMEvent[] = []
       const redactedData = reasoning.redactedContent ?? reasoning.data
-      const providerMetadata = reasoning.signature
-        ? bedrockMetadata({ signature: reasoning.signature })
+      const metadata = reasoning.signature
+        ? providerMetadata(state.providerMetadataKey, { signature: reasoning.signature })
         : redactedData !== undefined
-          ? bedrockMetadata({ redactedData })
+          ? providerMetadata(state.providerMetadataKey, { redactedData })
           : undefined
       const lifecycle =
-        reasoning.text !== undefined || providerMetadata !== undefined
-          ? Lifecycle.reasoningDelta(
-              state.lifecycle,
-              events,
-              `reasoning-${index}`,
-              reasoning.text ?? "",
-              providerMetadata,
-            )
+        reasoning.text !== undefined || metadata !== undefined
+          ? Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${index}`, reasoning.text ?? "", metadata)
           : state.lifecycle
       return [
         {
@@ -596,7 +594,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
             events,
             `reasoning-${index}`,
             state.reasoningSignatures[index]
-              ? bedrockMetadata({ signature: state.reasoningSignatures[index] })
+              ? providerMetadata(state.providerMetadataKey, { signature: state.reasoningSignatures[index] })
               : undefined,
           )
       events.push(...resultEvents)
@@ -633,7 +631,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
     }
 
     if (event.metadata) {
-      const usage = mapUsage(event.metadata.usage) ?? state.pendingFinish?.usage
+      const usage = mapUsage(event.metadata.usage, state.providerMetadataKey) ?? state.pendingFinish?.usage
       return [
         {
           ...state,
@@ -698,7 +696,8 @@ export const protocol = Protocol.make({
   },
   stream: {
     event: BedrockEvent,
-    initial: () => ({
+    initial: (request) => ({
+      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
       tools: ToolStream.empty<number>(),
       pendingFinish: undefined,
       hasToolCalls: false,

packages/ai/src/protocols/gemini.ts

@@ -229,6 +229,7 @@ type GeminiEvent = Schema.Schema.Type<typeof GeminiEvent>
 
 interface ParserState {
   readonly route: string
+  readonly providerMetadataKey: string
   readonly finishReason?: string
   readonly hasToolCalls: boolean
   readonly promptFeedback?: GeminiPromptFeedback
@@ -285,22 +286,23 @@ const lowerUserPart = Effect.fn("Gemini.lowerUserPart")(function* (part: TextPar
   return { inlineData: { mimeType: media.mime, data: media.base64 } }
 })
 
-const googleMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ google: metadata })
+const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })
 
-const thoughtSignature = (providerMetadata: ProviderMetadata | undefined) => {
-  const google = providerMetadata?.google
-  return ProviderShared.isRecord(google) && typeof google.thoughtSignature === "string"
-    ? google.thoughtSignature
+const thoughtSignature = (metadata: ProviderMetadata | undefined, key: string) => {
+  const value = metadata?.[key]
+  return ProviderShared.isRecord(value) && typeof value.thoughtSignature === "string"
+    ? value.thoughtSignature
     : undefined
 }
 
-const lowerToolCall = (part: ToolCallPart, omitIds: boolean) => ({
+const lowerToolCall = (part: ToolCallPart, omitIds: boolean, metadataKey: string) => ({
   functionCall: { ...(omitIds ? {} : { id: part.id }), name: part.name, args: part.input },
-  thoughtSignature: thoughtSignature(part.providerMetadata),
+  thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey),
 })
 
 const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMRequest) {
   const contents: GeminiContent[] = []
+  const metadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)
   const omitCallIds = omitsFunctionCallIds(request.model.id)
   const legacyToolMedia = routesLegacyToolMedia(request.model.id)
   let pendingMedia: GeminiInlineDataPart[] | undefined
@@ -342,15 +344,19 @@ const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMR
         if (!ProviderShared.supportsContent(part, ["text", "reasoning", "tool-call"]))
           return yield* ProviderShared.unsupportedContent("Gemini", "assistant", ["text", "reasoning", "tool-call"])
         if (part.type === "text") {
-          parts.push({ text: part.text, thoughtSignature: thoughtSignature(part.providerMetadata) })
+          parts.push({ text: part.text, thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey) })
           continue
         }
         if (part.type === "reasoning") {
-          parts.push({ text: part.text, thought: true, thoughtSignature: thoughtSignature(part.providerMetadata) })
+          parts.push({
+            text: part.text,
+            thought: true,
+            thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey),
+          })
           continue
         }
         if (part.type === "tool-call") {
-          const lowered = lowerToolCall(part, omitCallIds)
+          const lowered = lowerToolCall(part, omitCallIds, metadataKey)
           const signature = lowered.thoughtSignature
           parts.push({
             ...lowered,
@@ -498,7 +504,7 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
 // `cachedContentTokenCount` subset. `candidatesTokenCount` is *exclusive*
 // of `thoughtsTokenCount` — visible-only, not a total — so we sum the two
 // to produce the inclusive `outputTokens` the rest of the contract expects.
-const mapUsage = (usage: GeminiUsage | undefined) => {
+const mapUsage = (usage: GeminiUsage | undefined, metadataKey: string) => {
   if (!usage) return undefined
   // Explicit provider nulls decode as `null`; normalize to `undefined` so the
   // token arithmetic below treats them like absent counts.
@@ -519,7 +525,7 @@ const mapUsage = (usage: GeminiUsage | undefined) => {
     cacheReadInputTokens: cached,
     reasoningTokens: thoughts,
     totalTokens: ProviderShared.totalTokens(promptTokens, outputTokens, usage.totalTokenCount ?? undefined),
-    providerMetadata: { google: usage },
+    providerMetadata: providerMetadata(metadataKey, usage),
   })
 }
 
@@ -567,14 +573,14 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
       lifecycle,
       events,
       "reasoning-0",
-      googleMetadata({ thoughtSignature: state.reasoningSignature }),
+      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.reasoningSignature }),
     )
   if (state.textSignature !== undefined)
     lifecycle = Lifecycle.textEnd(
       lifecycle,
       events,
       "text-0",
-      googleMetadata({ thoughtSignature: state.textSignature }),
+      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.textSignature }),
     )
   Lifecycle.finish(lifecycle, events, {
     reason: {
@@ -584,7 +590,9 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
     },
     usage: state.usage,
     providerMetadata:
-      state.promptFeedback === undefined ? undefined : googleMetadata({ promptFeedback: state.promptFeedback }),
+      state.promptFeedback === undefined
+        ? undefined
+        : providerMetadata(state.providerMetadataKey, { promptFeedback: state.promptFeedback }),
   })
   return events
 }
@@ -593,7 +601,9 @@ const step = (state: ParserState, event: GeminiEvent) => {
   const nextState = {
     ...state,
     promptFeedback: event.promptFeedback ?? state.promptFeedback,
-    usage: event.usageMetadata ? (mapUsage(event.usageMetadata) ?? state.usage) : state.usage,
+    usage: event.usageMetadata
+      ? (mapUsage(event.usageMetadata, state.providerMetadataKey) ?? state.usage)
+      : state.usage,
   }
   const candidate = event.candidates?.[0]
   if (!candidate?.content)
@@ -637,7 +647,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
           events,
           "reasoning-0",
           part.text,
-          signature ? googleMetadata({ thoughtSignature: signature }) : undefined,
+          signature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: signature }) : undefined,
         )
         continue
       }
@@ -645,14 +655,16 @@ const step = (state: ParserState, event: GeminiEvent) => {
         lifecycle,
         events,
         "reasoning-0",
-        reasoningSignature ? googleMetadata({ thoughtSignature: reasoningSignature }) : undefined,
+        reasoningSignature
+          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
+          : undefined,
       )
       lifecycle = Lifecycle.textDelta(
         lifecycle,
         events,
         "text-0",
         part.text,
-        textSignature ? googleMetadata({ thoughtSignature: textSignature }) : undefined,
+        textSignature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature }) : undefined,
       )
       textSignature = undefined
       continue
@@ -672,7 +684,9 @@ const step = (state: ParserState, event: GeminiEvent) => {
         lifecycle,
         events,
         "reasoning-0",
-        reasoningSignature ? googleMetadata({ thoughtSignature: reasoningSignature }) : undefined,
+        reasoningSignature
+          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
+          : undefined,
       )
       lifecycle = Lifecycle.stepStart(lifecycle, events)
       events.push(
@@ -681,7 +695,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
           name: part.functionCall.name,
           input,
           providerMetadata: part.thoughtSignature
-            ? googleMetadata({ thoughtSignature: part.thoughtSignature })
+            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: part.thoughtSignature })
             : undefined,
         }),
       )
@@ -720,6 +734,7 @@ export const protocol = Protocol.make({
     event: Protocol.jsonEvent(GeminiEvent),
     initial: (request) => ({
       route: `${request.model.provider}/${request.model.route.id}`,
+      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
       hasToolCalls: false,
       lifecycle: Lifecycle.initial(),
     }),

packages/ai/src/protocols/open-responses.ts

@@ -346,7 +346,8 @@ export const Event = Schema.StructWithRest(
     item_id: Schema.optional(Schema.String),
     output_index: Schema.optional(Schema.Number),
     summary_index: Schema.optional(Schema.Number),
-    item: Schema.optional(StreamItem),
+    // OutputItemAdded/Done permit a null item in the Open Responses OpenAPI schema.
+    item: optionalNull(StreamItem),
     response: Schema.optional(
       Schema.StructWithRest(
         Schema.Struct({
@@ -812,7 +813,7 @@ export const providerMetadata = (state: ParserState, metadata: Record<string, un
 })
 
 const isReasoningItem = (item: StreamItem): item is StreamItem & { type: "reasoning"; id: string } =>
-  item.type === "reasoning" && typeof item.id === "string" && item.id.length > 0
+  item.type === "reasoning" && typeof item.id === "string"
 
 export type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
 
@@ -891,7 +892,7 @@ const reasoningMetadata = (state: ParserState, item: StreamItem & { id: string }
 // best-effort, not guaranteed.
 const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
   const item = event.item
-  if (item?.type === "message" && item.id) {
+  if (item?.type === "message" && item.id !== undefined) {
     const phase = messagePhase(item.phase)
     return [
       {
@@ -922,7 +923,7 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
   }
   if (item?.type !== "function_call" || !item.call_id) return [state, NO_EVENTS]
   const id = item.id ?? item.call_id
-  const metadata = item.id ? providerMetadata(state, { itemId: item.id }) : undefined
+  const metadata = item.id !== undefined ? providerMetadata(state, { itemId: item.id }) : undefined
   const events: LLMEvent[] = []
   const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
   return [
@@ -941,7 +942,7 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
 }
 
 const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResult => {
-  if (!event.item_id || event.summary_index === undefined) return [state, NO_EVENTS]
+  if (event.item_id === undefined || event.summary_index === undefined) return [state, NO_EVENTS]
   const item = state.reasoningItems[event.item_id]
   if (!item) return [state, NO_EVENTS]
   if (event.summary_index === 0) return [state, NO_EVENTS]
@@ -988,7 +989,7 @@ const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResu
 }
 
 const onReasoningSummaryPartDone = (state: ParserState, event: Event): StepResult => {
-  if (!event.item_id || event.summary_index === undefined) return [state, NO_EVENTS]
+  if (event.item_id === undefined || event.summary_index === undefined) return [state, NO_EVENTS]
   const item = state.reasoningItems[event.item_id]
   if (!item) return [state, NO_EVENTS]
   return [
@@ -1013,7 +1014,7 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenResponses.onFunctionCallArgu
   state: ParserState,
   event: Event,
 ) {
-  if (!event.item_id) return [state, NO_EVENTS] satisfies StepResult
+  if (event.item_id === undefined) return [state, NO_EVENTS] satisfies StepResult
   const tool = state.tools[event.item_id]
   if (!tool) return [state, NO_EVENTS] satisfies StepResult
   const final = event.type === "response.function_call_arguments.done" ? event.arguments : undefined
@@ -1044,7 +1045,7 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
   const item = event.item
   if (!item) return [state, NO_EVENTS] satisfies StepResult
 
-  if (item.type === "message" && item.id) {
+  if (item.type === "message" && item.id !== undefined) {
     const itemPhase = messagePhase(item.phase)
     const phase = itemPhase === undefined ? state.messagePhases[item.id] : itemPhase
     const events: LLMEvent[] = []
@@ -1075,7 +1076,7 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
       : ToolStream.start(state.tools, id, {
           id: item.call_id,
           name: item.name,
-          providerMetadata: item.id ? providerMetadata(state, { itemId: item.id }) : undefined,
+          providerMetadata: item.id !== undefined ? providerMetadata(state, { itemId: item.id }) : undefined,
         })
     const result =
       item.arguments === undefined
@@ -1136,7 +1137,7 @@ const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (
           ([current, events], item) => {
             const id = item.id ?? (item.type === "function_call" ? item.call_id : undefined)
             if (
-              !id ||
+              id === undefined ||
               ((item.type !== "function_call" || !current.tools[id]) &&
                 (item.type !== "reasoning" || !current.reasoningItems[id]))
             )
@@ -1219,12 +1220,13 @@ export const providerFailure = (id: string, event: Event, fallback: string) => {
 const providerError = (state: ParserState, event: Event, fallback: string) => providerFailure(state.id, event, fallback)
 
 export const step = (state: ParserState, input: Event) => {
+  // The OpenAPI requires string IDs but imposes no minLength; empty is not missing.
   const event =
-    input.item_id && outputItemID(state, input) !== input.item_id
+    input.item_id !== undefined && outputItemID(state, input) !== input.item_id
       ? { ...input, item_id: outputItemID(state, input) }
       : input
   if (event.type === "response.output_text.delta" || event.type === "response.output_text.done") {
-    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
     return Effect.succeed(
       event.type === "response.output_text.delta"
         ? onOutputTextDelta(state, event, event.item_id)
@@ -1233,7 +1235,7 @@ export const step = (state: ParserState, input: Event) => {
   }
   if (event.type === "response.refusal.delta" || event.type === "response.refusal.done") {
     const value = event.type === "response.refusal.delta" ? event.delta : event.refusal
-    if (!event.item_id || typeof value !== "string")
+    if (event.item_id === undefined || typeof value !== "string")
       return ProviderShared.eventError(state.id, `${event.type} is malformed`)
     return Effect.succeed(
       event.type === "response.refusal.delta"
@@ -1242,7 +1244,7 @@ export const step = (state: ParserState, input: Event) => {
     )
   }
   if (event.type === "response.reasoning.delta" || event.type === "response.reasoning_summary_text.delta") {
-    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
     return Effect.succeed(onReasoningDelta(state, event, event.item_id))
   }
   if (
@@ -1250,24 +1252,24 @@ export const step = (state: ParserState, input: Event) => {
     event.type === "response.reasoning_summary_text.done" ||
     event.type === "response.reasoning_text.done"
   ) {
-    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
     return Effect.succeed(onReasoningDone(state, event, event.item_id))
   }
   if (event.type === "response.reasoning_summary_part.added")
-    return event.item_id
+    return event.item_id !== undefined
       ? Effect.succeed(onReasoningSummaryPartAdded(state, event))
       : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
   if (event.type === "response.reasoning_summary_part.done")
-    return event.item_id
+    return event.item_id !== undefined
       ? Effect.succeed(onReasoningSummaryPartDone(state, event))
       : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
   if (event.type === "response.output_item.added") {
-    if (event.item?.type === "message" && !event.item.id)
+    if (event.item?.type === "message" && event.item.id === undefined)
       return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
     const id = event.item?.id ?? (event.item?.type === "function_call" ? event.item.call_id : undefined)
     return Effect.succeed(
       onOutputItemAdded(
-        event.output_index !== undefined && id
+        event.output_index !== undefined && id !== undefined
           ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: id } }
           : state,
         event,
@@ -1275,11 +1277,11 @@ export const step = (state: ParserState, input: Event) => {
     )
   }
   if (event.type === "response.function_call_arguments.delta" || event.type === "response.function_call_arguments.done")
-    return event.item_id
+    return event.item_id !== undefined
       ? onFunctionCallArgumentsDelta(state, event)
       : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
   if (event.type === "response.output_item.done") {
-    if (event.item?.type === "message" && !event.item.id)
+    if (event.item?.type === "message" && event.item.id === undefined)
       return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
     return onOutputItemDone(state, event)
   }

packages/ai/src/protocols/openai-chat.ts

@@ -253,6 +253,7 @@ interface PendingToolDelta {
 }
 
 export interface ParserState {
+  readonly providerMetadataKey: string
   readonly tools: ToolStream.State<number>
   readonly pendingTools: Partial<Record<number, PendingToolDelta>>
   readonly toolCallEvents: ReadonlyArray<LLMEvent>
@@ -324,17 +325,18 @@ const lowerMedia = Effect.fn("OpenAIChat.lowerMedia")(function* (part: MediaPart
 const openAICompatibleReasoningContent = (native: unknown) =>
   isRecord(native) && typeof native.reasoning_content === "string" ? native.reasoning_content : undefined
 
-const reasoningField = (part: ReasoningPart) => {
-  const field = part.providerMetadata?.openai?.reasoningField
+const reasoningField = (part: ReasoningPart, providerMetadataKey: string) => {
+  const field = part.providerMetadata?.[providerMetadataKey]?.reasoningField
   return typeof field === "string" ? field : undefined
 }
 
-const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown) => {
+const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown, providerMetadataKey: string) => {
   const observed = parts.flatMap((part) => {
-    const details = part.providerMetadata?.openai?.reasoningDetails
+    const details = part.providerMetadata?.[providerMetadataKey]?.reasoningDetails
     return Array.isArray(details) ? details : []
   })
-  if (parts.some((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))) return observed
+  if (parts.some((part) => Array.isArray(part.providerMetadata?.[providerMetadataKey]?.reasoningDetails)))
+    return observed
   if (isRecord(native) && Array.isArray(native.reasoning_details)) return native.reasoning_details
 }
 
@@ -366,7 +368,7 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
   message: OpenAIChatRequestMessage,
   configuredField: string | undefined,
   requireReasoning: boolean,
-  options: LoweringOptions,
+  options: LoweringOptions & { readonly providerMetadataKey: string },
 ) {
   const content: TextPart[] = []
   const reasoning: ReasoningPart[] = []
@@ -388,10 +390,14 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
     }
   }
   const text = reasoning.map((part) => part.text).join("")
-  const details = reasoningDetails(reasoning, message.native?.openaiCompatible)
-  const observedField = reasoning.map(reasoningField).find((value) => value !== undefined)
+  const details = reasoningDetails(reasoning, message.native?.openaiCompatible, options.providerMetadataKey)
+  const observedField = reasoning
+    .map((part) => reasoningField(part, options.providerMetadataKey))
+    .find((value) => value !== undefined)
   const nativeReasoning = openAICompatibleReasoningContent(message.native?.openaiCompatible)
-  const fullyStructured = reasoning.every((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))
+  const fullyStructured = reasoning.every((part) =>
+    Array.isArray(part.providerMetadata?.[options.providerMetadataKey]?.reasoningDetails),
+  )
   const field = (() => {
     if (configuredField !== undefined && (requireReasoning || reasoning.length > 0 || nativeReasoning !== undefined))
       return configuredField
@@ -459,7 +465,7 @@ const lowerMessage = Effect.fn("OpenAIChat.lowerMessage")(function* (
   message: OpenAIChatRequestMessage,
   reasoningField: string | undefined,
   requireReasoning: boolean,
-  options: LoweringOptions,
+  options: LoweringOptions & { readonly providerMetadataKey: string },
 ) {
   if (message.role === "user") return [yield* lowerUserMessage(message, options)]
   if (message.role === "assistant")
@@ -495,6 +501,7 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
   const mistral = ["mistral", "devstral", "codestral", "pixtral", "mixtral"].some((family) => modelID.includes(family))
   const lowering = {
     ...options,
+    providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
     toolCallID: (id: string) => {
       if (mistral)
         return id
@@ -820,7 +827,7 @@ const mapFinishReason = Effect.fn("OpenAIChat.mapFinishReason")(function* (event
 // Providers differ on cache-hit location: OpenAI uses
 // `prompt_tokens_details.cached_tokens`, DeepSeek uses
 // `prompt_cache_hit_tokens`, and Zai uses top-level `cached_tokens`.
-const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
+const mapUsage = (usage: OpenAIChatEvent["usage"], providerMetadataKey: string): Usage | undefined => {
   if (!usage) return undefined
   const input = usage.prompt_tokens ?? undefined
   const output = usage.completion_tokens ?? undefined
@@ -839,7 +846,7 @@ const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
     cacheWriteInputTokens: cacheWrite,
     reasoningTokens: reasoning,
     totalTokens: ProviderShared.totalTokens(input, output, usage.total_tokens ?? undefined),
-    providerMetadata: { openai: usage },
+    providerMetadata: { [providerMetadataKey]: usage },
   })
 }
 
@@ -913,8 +920,12 @@ const conflictingReasoningTextDetails = (previous: Record<string, unknown>, curr
 const conflictingDetailValue = (previous: unknown, current: unknown) =>
   previous !== undefined && previous !== null && current !== undefined && current !== null && previous !== current
 
-const reasoningMetadata = (field: ParserState["reasoningField"], details?: ReadonlyArray<unknown>) => ({
-  openai: {
+const reasoningMetadata = (
+  providerMetadataKey: string,
+  field: ParserState["reasoningField"],
+  details?: ReadonlyArray<unknown>,
+) => ({
+  [providerMetadataKey]: {
     ...(field ? { reasoningField: field } : {}),
     ...(details ? { reasoningDetails: details } : {}),
   },
@@ -941,7 +952,10 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
     // Moonshot (and a few other OpenAI-compatible providers) attach usage to
     // `choice.usage` instead of the top-level `usage` field.
     const choiceUsage = (choice as unknown as { usage?: OpenAIChatEvent["usage"] })?.usage
-    const usage = mapUsage(event.usage) ?? (choiceUsage ? mapUsage(choiceUsage) : undefined) ?? state.usage
+    const usage =
+      mapUsage(event.usage, state.providerMetadataKey) ??
+      (choiceUsage ? mapUsage(choiceUsage, state.providerMetadataKey) : undefined) ??
+      state.usage
     const rawFinishReason = choice?.finish_reason
     const finishReason = rawFinishReason
       ? {
@@ -979,7 +993,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
     const detailDelta = Array.isArray(delta?.reasoning_details) ? delta.reasoning_details : undefined
     if (detailDelta !== undefined) appendReasoningDetails(state.reasoningDetails, detailDelta)
     const reasoningDetailsObserved = state.reasoningDetailsObserved || detailDelta !== undefined
-    const deltaMetadata = reasoningMetadata(reasoningField)
+    const deltaMetadata = reasoningMetadata(state.providerMetadataKey, reasoningField)
     const text = detailDelta?.length ? (detailText(detailDelta) ?? reasoning?.text) : reasoning?.text
     if (text !== undefined) lifecycle = Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", text, deltaMetadata)
     else if (
@@ -995,7 +1009,11 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
         lifecycle,
         events,
         "reasoning-0",
-        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
+        reasoningMetadata(
+          state.providerMetadataKey,
+          reasoningField,
+          reasoningDetailsObserved ? state.reasoningDetails : undefined,
+        ),
       )
       lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
     }
@@ -1005,7 +1023,11 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
         lifecycle,
         events,
         "reasoning-0",
-        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
+        reasoningMetadata(
+          state.providerMetadataKey,
+          reasoningField,
+          reasoningDetailsObserved ? state.reasoningDetails : undefined,
+        ),
       )
       lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.refusal)
     }
@@ -1066,6 +1088,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
 
     return [
       {
+        providerMetadataKey: state.providerMetadataKey,
         tools: finished?.tools ?? tools,
         pendingTools,
         toolCallEvents: finished?.events ?? state.toolCallEvents,
@@ -1109,12 +1132,18 @@ const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: Pars
       }
     : { normalized: hasToolCalls ? ("tool-calls" as const) : ("stop" as const) }
   const metadata = reasoningMetadata(
+    state.providerMetadataKey,
     state.reasoningField,
     state.reasoningDetailsObserved ? state.reasoningDetails : undefined,
   )
   const started =
     state.reasoningDetailsObserved && !state.reasoningEmitted
-      ? Lifecycle.reasoningStart(state.lifecycle, events, "reasoning-0", reasoningMetadata(state.reasoningField))
+      ? Lifecycle.reasoningStart(
+          state.lifecycle,
+          events,
+          "reasoning-0",
+          reasoningMetadata(state.providerMetadataKey, state.reasoningField),
+        )
       : state.lifecycle
   const ended = Lifecycle.reasoningEnd(started, events, "reasoning-0", metadata)
   const lifecycle = toolCallEvents.length ? Lifecycle.stepStart(ended, events) : ended
@@ -1141,6 +1170,7 @@ export const protocol = Protocol.make({
   stream: {
     event: Protocol.jsonEvent(OpenAIChatEvent),
     initial: (request) => ({
+      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
       tools: ToolStream.empty<number>(),
       pendingTools: {},
       toolCallEvents: [],

packages/ai/src/protocols/openai-responses.ts

@@ -184,7 +184,7 @@ const HOSTED_TOOLS = {
 
 const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
   if (event.type === "response.reasoning_text.delta")
-    return event.item_id
+    return event.item_id !== undefined
       ? Effect.succeed(
           OpenResponses.onReasoningDelta(state, event, OpenResponses.outputItemID(state, event) ?? event.item_id),
         )

packages/ai/src/providers/amazon-bedrock-mantle.ts

@@ -23,13 +23,14 @@ export interface Settings extends ProviderPackage.Settings {
   readonly baseURL?: string
   readonly credentials?: Credentials
   readonly region?: string
+  readonly topP?: number
   readonly providerOptions?: OpenAIProviderOptionsInput
 }
 
 const responsesRoute = Route.make({
   id: "bedrock-mantle-responses",
   provider: id,
-  providerMetadataKey: OpenAIResponses.route.providerMetadataKey,
+  providerMetadataKey: "mantle",
   protocol: OpenAIResponses.protocol,
   endpoint: OpenAIResponses.route.endpoint,
   auth: OpenAIResponses.route.auth,
@@ -40,6 +41,7 @@ const responsesRoute = Route.make({
 const chatRoute = OpenAIChat.route.with({
   id: "bedrock-mantle-chat",
   provider: id,
+  providerMetadataKey: "mantle",
 })
 
 export const routes = [responsesRoute, chatRoute]
@@ -76,7 +78,7 @@ export const configure = (input: Config = {}) => {
 
   return {
     id,
-    model: chat,
+    model: responses,
     chat,
     responses,
     configure,
@@ -94,6 +96,7 @@ const config = (settings: Settings): Config => {
     apiKey: settings.auth === "sigv4" ? undefined : settings.apiKey,
     baseURL: settings.baseURL,
     credentials: settings.credentials,
+    generation: settings.topP === undefined ? undefined : { topP: settings.topP },
     headers: settings.headers === undefined ? undefined : { ...settings.headers },
     http: settings.body === undefined ? undefined : { body: { ...settings.body } },
     providerOptions: settings.providerOptions,
@@ -109,4 +112,4 @@ export const responsesModel: ProviderPackage.Definition<Settings, OpenAIProvider
   modelID,
   settings,
 ) => configure(config(settings)).responses(modelID)
-export const model = chatModel
+export const model = responsesModel

packages/ai/src/providers/amazon-bedrock.ts

@@ -35,6 +35,7 @@ const configuredRoute = (input: Config) => {
   return BedrockConverse.route.with({
     ...rest,
     provider: id,
+    providerMetadataKey: "bedrock",
     endpoint: { baseURL: baseURL ?? bedrockBaseURL(resolvedRegion) },
     auth: apiKey === undefined ? BedrockConverse.sigV4Auth(credentials) : Auth.bearer(apiKey),
   })

packages/ai/src/providers/amazon-bedrock/mantle.ts

@@ -1,2 +1,2 @@
-export { chatModel as model } from "../amazon-bedrock-mantle.js"
+export { responsesModel as model } from "../amazon-bedrock-mantle.js"
 export type { Settings } from "../amazon-bedrock-mantle.js"

packages/ai/src/providers/google-vertex-chat.ts

@@ -27,6 +27,7 @@ export interface Settings extends ProviderPackage.Settings {
 const route = OpenAICompatibleChat.route.with({
   id: "google-vertex-chat",
   provider: id,
+  providerMetadataKey: "vertex",
 })
 
 export const routes = [route]

packages/ai/src/providers/google-vertex-responses.ts

@@ -27,6 +27,7 @@ export interface Settings extends ProviderPackage.Settings {
 const route = OpenAICompatibleResponses.route.with({
   id: "google-vertex-responses",
   provider: id,
+  providerMetadataKey: "vertex",
   providerOptions: { store: false },
 })
 

packages/ai/src/providers/google-vertex.ts

@@ -68,7 +68,7 @@ const protocol = {
 const route = Route.make({
   id: "google-vertex-gemini",
   provider: id,
-  providerMetadataKey: "google",
+  providerMetadataKey: "vertex",
   protocol,
   endpoint: Endpoint.path(({ request }) => {
     const model = String(request.model.id)

packages/ai/src/providers/openrouter.ts

@@ -164,6 +164,7 @@ const bodyOptions = (input: unknown) => {
 export const route = Route.make({
   id: ADAPTER,
   provider: profile.provider,
+  providerMetadataKey: "openrouter",
   protocol,
   endpoint: Endpoint.path("/chat/completions", { baseURL: profile.baseURL }),
   framing: Framing.sse,

packages/ai/src/route/client.ts

@@ -89,6 +89,7 @@ export interface RouteDefaultsInput {
 export interface RoutePatch<Body, Prepared> extends RouteDefaultsInput {
   readonly id?: string
   readonly provider?: string | ProviderID
+  readonly providerMetadataKey?: string
   readonly auth?: Auth.Definition
   readonly transport?: Transport<Body, Prepared, unknown>
   readonly endpoint?: EndpointPatch<Body>
@@ -289,11 +290,16 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
       defaults: routeInput.defaults ?? {},
       body: protocol.body,
       with: (patch: RoutePatch<Body, Prepared>) => {
-        const { id, provider, auth, transport, endpoint, ...defaults } = patch
+        const { id, provider, providerMetadataKey, auth, transport, endpoint, ...defaults } = patch
         return build({
           ...routeInput,
           id: id ?? routeInput.id,
           provider: provider ?? routeInput.provider,
+          providerMetadataKey:
+            providerMetadataKey ??
+            (provider !== undefined && String(provider) !== String(routeInput.provider)
+              ? String(provider)
+              : routeInput.providerMetadataKey),
           auth: auth ?? routeInput.auth,
           endpoint: endpoint ? Endpoint.merge(routeInput.endpoint, endpoint) : routeInput.endpoint,
           transport: (transport as Transport<Body, Prepared, Frame> | undefined) ?? routeInput.transport,

packages/ai/src/route/executor.ts

@@ -40,17 +40,6 @@ const headerDetails = (headers: Headers.Headers) =>
 const normalizedHeaders = (headers: Headers.Headers) =>
   Object.fromEntries(Object.entries(headers).map(([key, value]) => [key.toLowerCase(), value]))
 
-const requestId = (headers: Record<string, string>) => {
-  return (
-    headers["x-request-id"] ??
-    headers["request-id"] ??
-    headers["x-amzn-requestid"] ??
-    headers["x-amz-request-id"] ??
-    headers["x-goog-request-id"] ??
-    headers["cf-ray"]
-  )
-}
-
 const retryAfterMs = (headers: Record<string, string>) => {
   const millis = Number(headers["retry-after-ms"])
   if (Number.isFinite(millis)) return Math.max(0, millis)
@@ -147,14 +136,12 @@ const responseHttp = (input: {
   readonly request: HttpClientRequest.HttpClientRequest
   readonly response: HttpClientResponse.HttpClientResponse
   readonly body: ReturnType<typeof responseBody>
-  readonly requestId?: string | undefined
   readonly rateLimit?: HttpRateLimitDetails | undefined
 }) =>
   new HttpContext({
     request: requestDetails(input.request),
     response: responseDetails(input.response),
     ...input.body,
-    requestId: input.requestId,
     rateLimit: input.rateLimit,
   })
 
@@ -179,7 +166,6 @@ const statusError =
             request,
             response,
             body: details,
-            requestId: requestId(headers),
             rateLimit,
           }),
         }),
@@ -216,7 +202,6 @@ export const classifyHttpFailure = (input: {
           ? undefined
           : new HttpResponseDetails({ status: input.status, headers: headerDetails(Headers.fromInput(headers)) }),
       ...details,
-      requestId: requestId(headers),
       rateLimit,
     }),
   })

packages/ai/src/schema/errors.ts

@@ -29,7 +29,6 @@ export class HttpContext extends Schema.Class<HttpContext>("AI.HttpContext")({
   response: Schema.optional(HttpResponseDetails),
   body: Schema.optional(Schema.String),
   bodyTruncated: Schema.optional(Schema.Boolean),
-  requestId: Schema.optional(Schema.String),
   rateLimit: Schema.optional(HttpRateLimitDetails),
 }) {}
 

packages/ai/test/executor.test.ts

@@ -327,7 +327,6 @@ describe("RequestExecutor", () => {
           retryAfterMs: 0,
           rateLimit: { retryAfterMs: 0 },
           http: {
-            requestId: "req_123",
             request: {
               method: "POST",
               url: "https://provider.test/v1/chat?api_key=secret&key=secret&debug=1",

packages/ai/test/provider/anthropic-messages.test.ts

@@ -2,7 +2,7 @@ import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { CacheHint, LLM, AIError, LLMRequest, Message, ToolCallPart, ToolDefinition, Usage } from "../../src/index.js"
-import { Auth, LLMClient } from "../../src/route.js"
+import { Auth, Endpoint, LLMClient, Route } from "../../src/route.js"
 import { compileRequest } from "../../src/route/client.js"
 import * as AnthropicMessages from "../../src/protocols/anthropic-messages.js"
 import { GoogleVertexMessages } from "../../src/providers.js"
@@ -810,6 +810,99 @@ describe("Anthropic Messages route", () => {
     }),
   )
 
+  it.effect("round-trips compatible provider metadata in its own namespace", () =>
+    Effect.gen(function* () {
+      const compatible = Route.make({
+        id: "custom-anthropic-messages",
+        provider: "custom-anthropic",
+        protocol: AnthropicMessages.protocol,
+        endpoint: Endpoint.path("/messages", { baseURL: "https://compatible.test/v1" }),
+        auth: Auth.header("x-api-key", "test"),
+        framing: AnthropicMessages.framing,
+      }).model({ id: "custom-model" })
+      const result = [
+        {
+          type: "web_search_result",
+          url: "https://example.com",
+          citations: [{ type: "web_search_result_location", cited_text: "Example" }],
+        },
+      ]
+      const response = yield* LLMClient.generate(LLM.request({ model: compatible, prompt: "Search." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "message_start", message: { usage: { input_tokens: 5, custom_start: true } } },
+              { type: "content_block_start", index: 0, content_block: { type: "thinking", thinking: "Thinking." } },
+              { type: "content_block_delta", index: 0, delta: { type: "signature_delta", signature: "custom_sig" } },
+              { type: "content_block_stop", index: 0 },
+              {
+                type: "content_block_start",
+                index: 1,
+                content_block: { type: "redacted_thinking", data: "custom_redacted" },
+              },
+              { type: "content_block_stop", index: 1 },
+              {
+                type: "content_block_start",
+                index: 2,
+                content_block: {
+                  type: "server_tool_use",
+                  id: "custom_tool",
+                  name: "web_search",
+                  input: { query: "example" },
+                },
+              },
+              { type: "content_block_stop", index: 2 },
+              {
+                type: "content_block_start",
+                index: 3,
+                content_block: { type: "web_search_tool_result", tool_use_id: "custom_tool", content: result },
+              },
+              { type: "content_block_stop", index: 3 },
+              {
+                type: "message_delta",
+                delta: { stop_reason: "end_turn", stop_sequence: "custom_stop" },
+                usage: { output_tokens: 2, custom_terminal: true },
+              },
+              { type: "message_stop" },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.message.content).toMatchObject([
+        { type: "reasoning", text: "Thinking.", providerMetadata: { "custom-anthropic": { signature: "custom_sig" } } },
+        { type: "reasoning", text: "", providerMetadata: { "custom-anthropic": { redactedData: "custom_redacted" } } },
+        { type: "tool-call", id: "custom_tool", providerExecuted: true },
+        {
+          type: "tool-result",
+          providerExecuted: true,
+          providerMetadata: { "custom-anthropic": { blockType: "web_search_tool_result", result } },
+        },
+      ])
+      expect(response.usage?.providerMetadata).toEqual({
+        "custom-anthropic": { input_tokens: 5, custom_start: true, output_tokens: 2, custom_terminal: true },
+      })
+      expect(response.events.at(-1)).toMatchObject({
+        providerMetadata: { "custom-anthropic": { stopSequence: "custom_stop" } },
+      })
+
+      const prepared = yield* compileRequest(
+        LLM.request({ model: compatible, messages: [response.message], cache: "none" }),
+      )
+      expect(prepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [
+            { type: "thinking", thinking: "Thinking.", signature: "custom_sig" },
+            { type: "redacted_thinking", data: "custom_redacted" },
+            { type: "server_tool_use", id: "custom_tool", name: "web_search", input: { query: "example" } },
+            { type: "web_search_tool_result", tool_use_id: "custom_tool", content: result },
+          ],
+        },
+      ])
+    }),
+  )
+
   it.effect("parses text, reasoning, and usage stream fixtures", () =>
     Effect.gen(function* () {
       const body = sseEvents(

packages/ai/test/provider/bedrock-converse.test.ts

@@ -569,6 +569,57 @@ describe("Bedrock Converse route", () => {
     }),
   )
 
+  it.effect("round-trips reassigned provider reasoning and usage metadata in its own namespace", () =>
+    Effect.gen(function* () {
+      const compatible = model.route.with({ provider: "custom-bedrock" }).model({ id: model.id })
+      const redactedData = "cmVkYWN0ZWQtdGhpbmtpbmc="
+      const response = yield* LLMClient.generate(LLMRequest.update(baseRequest, { model: compatible })).pipe(
+        Effect.provide(
+          fixedBytes(
+            eventStreamBody(
+              ["messageStart", { role: "assistant" }],
+              ["contentBlockDelta", { contentBlockIndex: 0, delta: { reasoningContent: { text: "Let me think." } } }],
+              ["contentBlockDelta", { contentBlockIndex: 0, delta: { reasoningContent: { signature: "custom_sig" } } }],
+              ["contentBlockStop", { contentBlockIndex: 0 }],
+              [
+                "contentBlockDelta",
+                { contentBlockIndex: 1, delta: { reasoningContent: { redactedContent: redactedData } } },
+              ],
+              ["contentBlockStop", { contentBlockIndex: 1 }],
+              ["messageStop", { stopReason: "end_turn" }],
+              ["metadata", { usage: { inputTokens: 5, outputTokens: 2, totalTokens: 7 } }],
+            ),
+          ),
+        ),
+      )
+
+      expect(response.message.content).toEqual([
+        {
+          type: "reasoning",
+          text: "Let me think.",
+          providerMetadata: { "custom-bedrock": { signature: "custom_sig" } },
+        },
+        { type: "reasoning", text: "", providerMetadata: { "custom-bedrock": { redactedData } } },
+      ])
+      expect(response.usage?.providerMetadata).toEqual({
+        "custom-bedrock": { inputTokens: 5, outputTokens: 2, totalTokens: 7 },
+      })
+
+      const prepared = yield* compileRequest(
+        LLM.request({ model: compatible, messages: [response.message], cache: "none" }),
+      )
+      expect(prepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [
+            { reasoningContent: { reasoningText: { text: "Let me think.", signature: "custom_sig" } } },
+            { reasoningContent: { redactedContent: redactedData } },
+          ],
+        },
+      ])
+    }),
+  )
+
   it.effect("preserves reasoning signatures when contentBlockStop is missing", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(baseRequest).pipe(

packages/ai/test/provider/bedrock-mantle.test.ts

@@ -3,6 +3,7 @@ import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { LLM, Message } from "../../src/index.js"
 import { AmazonBedrockMantle } from "../../src/providers.js"
+import { model } from "../../src/providers/amazon-bedrock/mantle.js"
 import { OpenAIResponses } from "../../src/protocols/openai-responses.js"
 import { compileRequest, LLMClient } from "../../src/route/client.js"
 import { it } from "../lib/effect.js"
@@ -17,13 +18,16 @@ const credentials = {
 }
 
 describe("Amazon Bedrock Mantle provider", () => {
-  it.effect("uses Chat by default and exposes Responses", () =>
+  it.effect("uses Responses by default and exposes Chat explicitly", () =>
     Effect.gen(function* () {
       const provider = AmazonBedrockMantle.configure({ credentials })
-      expect(provider.responses("openai.gpt-oss-120b").route.transport).toBe(OpenAIResponses.httpTransport)
-      const chat = yield* compileRequest(LLM.request({ model: provider.model("openai.gpt-oss-120b"), prompt: "Hi" }))
+      expect(provider.model).toBe(provider.responses)
+      expect(AmazonBedrockMantle.model).toBe(AmazonBedrockMantle.responsesModel)
+      expect(model).toBe(AmazonBedrockMantle.responsesModel)
+      expect(provider.model("openai.gpt-oss-120b").route.transport).toBe(OpenAIResponses.httpTransport)
+      const chat = yield* compileRequest(LLM.request({ model: provider.chat("openai.gpt-oss-120b"), prompt: "Hi" }))
       const responses = yield* compileRequest(
-        LLM.request({ model: provider.responses("openai.gpt-oss-120b"), prompt: "Hi" }),
+        LLM.request({ model: provider.model("openai.gpt-oss-120b"), prompt: "Hi" }),
       )
 
       expect(chat).toMatchObject({
@@ -36,6 +40,23 @@ describe("Amazon Bedrock Mantle provider", () => {
         protocol: "openai-responses",
         body: { model: "openai.gpt-oss-120b", store: false },
       })
+      expect(provider.model("openai.gpt-oss-120b").route.providerMetadataKey).toBe("mantle")
+      expect(provider.chat("openai.gpt-oss-120b").route.providerMetadataKey).toBe("mantle")
+    }),
+  )
+
+  it.effect("preserves configured top-p generation defaults for Chat and Responses", () =>
+    Effect.gen(function* () {
+      const settings = { apiKey: "test-key", topP: 0.8 }
+      const chat = yield* compileRequest(
+        LLM.request({ model: AmazonBedrockMantle.chatModel("openai.gpt-oss-safeguard-20b", settings), prompt: "Hi" }),
+      )
+      const responses = yield* compileRequest(
+        LLM.request({ model: AmazonBedrockMantle.responsesModel("openai.gpt-oss-120b", settings), prompt: "Hi" }),
+      )
+
+      expect(chat.body.top_p).toBe(0.8)
+      expect(responses.body.top_p).toBe(0.8)
     }),
   )
 
@@ -107,6 +128,9 @@ describe("Amazon Bedrock Mantle provider", () => {
         LLM.request({ model, messages: [response.message, Message.user("Continue.")] }),
       )
 
+      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
+        mantle: { itemId: "msg_95d4d0af4350432a", reasoningEncryptedContent: "mantle-state" },
+      })
       expect(prepared.body.input).toEqual([
         {
           type: "reasoning",

packages/ai/test/provider/cloudflare.test.ts

@@ -126,7 +126,7 @@ describe("Cloudflare", () => {
       expect(response.reasoning).toBe("Thinking")
       expect(response.events.filter(LLMEvent.is.reasoningDelta)).toHaveLength(2)
       expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningField: "reasoning", reasoningDetails: merged },
+        "cloudflare-ai-gateway": { reasoningField: "reasoning", reasoningDetails: merged },
       })
 
       const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))

packages/ai/test/provider/google-vertex.test.ts

@@ -6,7 +6,7 @@ import { GoogleVertex, GoogleVertexChat, GoogleVertexMessages, GoogleVertexRespo
 import { LLMClient } from "../../src/route.js"
 import { compileRequest } from "../../src/route/client.js"
 import { it } from "../lib/effect.js"
-import { dynamicResponse } from "../lib/http.js"
+import { dynamicResponse, fixedResponse } from "../lib/http.js"
 import { deltaChunk, finishChunk } from "../lib/openai-chunks.js"
 import { sseEvents } from "../lib/sse.js"
 
@@ -89,7 +89,7 @@ describe("Google Vertex providers", () => {
                 id: "call_1",
                 name: "lookup",
                 input: { query: "weather" },
-                providerMetadata: { google: { functionCallId: "provider_call_1" } },
+                providerMetadata: { vertex: { functionCallId: "provider_call_1" } },
               }),
             ]),
             Message.tool({
@@ -97,7 +97,7 @@ describe("Google Vertex providers", () => {
               name: "lookup",
               result: "sunny",
               resultType: "text",
-              providerMetadata: { google: { functionCallId: "provider_call_1" } },
+              providerMetadata: { vertex: { functionCallId: "provider_call_1" } },
             }),
           ],
         }),
@@ -122,6 +122,91 @@ describe("Google Vertex providers", () => {
     }),
   )
 
+  it.effect("round-trips Vertex Gemini metadata through signed content, tool calls, and usage", () =>
+    Effect.gen(function* () {
+      const model = GoogleVertex.configure({
+        accessToken: "vertex-token",
+        project: "vertex-project",
+      }).model("gemini-3.5-flash")
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Check the weather." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents({
+              candidates: [
+                {
+                  content: {
+                    role: "model",
+                    parts: [
+                      { text: "Thinking.", thought: true, thoughtSignature: "reasoning_sig" },
+                      { text: "Checking.", thoughtSignature: "text_sig" },
+                      {
+                        functionCall: { id: "provider_call_1", name: "lookup", args: { query: "weather" } },
+                        thoughtSignature: "tool_sig",
+                      },
+                    ],
+                  },
+                  finishReason: "STOP",
+                },
+              ],
+              promptFeedback: { blockReasonMessage: "Reviewed" },
+              usageMetadata: { promptTokenCount: 5, candidatesTokenCount: 2, thoughtsTokenCount: 1 },
+            }),
+          ),
+        ),
+      )
+      const reasoning = response.events.find((event) => event.type === "reasoning-end")
+      const text = response.events.find((event) => event.type === "text-delta")
+      const toolCall = response.toolCalls[0]
+
+      expect(reasoning?.providerMetadata).toEqual({ vertex: { thoughtSignature: "reasoning_sig" } })
+      expect(text?.providerMetadata).toEqual({ vertex: { thoughtSignature: "text_sig" } })
+      expect(toolCall).toMatchObject({
+        id: "provider_call_1",
+        providerMetadata: { vertex: { thoughtSignature: "tool_sig" } },
+      })
+      expect(response.usage?.providerMetadata).toEqual({
+        vertex: { promptTokenCount: 5, candidatesTokenCount: 2, thoughtsTokenCount: 1 },
+      })
+      expect(response.events.at(-1)?.providerMetadata).toEqual({
+        vertex: { promptFeedback: { blockReasonMessage: "Reviewed" } },
+      })
+
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            Message.assistant([
+              { type: "reasoning", text: "Thinking.", providerMetadata: reasoning?.providerMetadata },
+              { type: "text", text: "Checking.", providerMetadata: text?.providerMetadata },
+              ToolCallPart.make({
+                id: toolCall.id,
+                name: toolCall.name,
+                input: toolCall.input,
+                providerMetadata: toolCall.providerMetadata,
+              }),
+            ]),
+            Message.tool({ id: toolCall.id, name: toolCall.name, result: "sunny", resultType: "text" }),
+          ],
+        }),
+      )
+
+      expect(prepared.body.contents).toEqual([
+        {
+          role: "model",
+          parts: [
+            { text: "Thinking.", thought: true, thoughtSignature: "reasoning_sig" },
+            { text: "Checking.", thoughtSignature: "text_sig" },
+            { functionCall: { name: "lookup", args: { query: "weather" } }, thoughtSignature: "tool_sig" },
+          ],
+        },
+        {
+          role: "user",
+          parts: [{ functionResponse: { name: "lookup", response: { name: "lookup", content: "sunny" } } }],
+        },
+      ])
+    }),
+  )
+
   it.effect("projects Anthropic Messages onto the Vertex raw-predict API", () =>
     Effect.gen(function* () {
       const model = GoogleVertexMessages.configure({

packages/ai/test/provider/native-providers.test.ts

@@ -2,13 +2,80 @@ import { describe, expect } from "bun:test"
 import { ConfigProvider, Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
 import { LLM, Message, ToolDefinition } from "../../src/index.js"
-import { Cerebras, DeepInfra, Groq, TogetherAI } from "../../src/providers/index.js"
+import {
+  AmazonBedrock,
+  AmazonBedrockMantle,
+  Anthropic,
+  AnthropicCompatible,
+  Azure,
+  Cerebras,
+  CloudflareAIGateway,
+  CloudflareWorkersAI,
+  DeepInfra,
+  Google,
+  GoogleVertex,
+  GoogleVertexChat,
+  GoogleVertexMessages,
+  GoogleVertexResponses,
+  Groq,
+  OpenAI,
+  OpenAICompatible,
+  OpenAICompatibleResponses,
+  OpenRouter,
+  TogetherAI,
+  XAI,
+} from "../../src/providers/index.js"
 import { compileRequest } from "../../src/route/client.js"
 import { it } from "../lib/effect.js"
 import { dynamicResponse } from "../lib/http.js"
 import { sseEvents } from "../lib/sse.js"
 
 describe("native OpenAI-compatible providers", () => {
+  it.effect("assigns provider-owned metadata namespaces across native routes", () =>
+    Effect.gen(function* () {
+      const vertex = { project: "project", accessToken: "token" }
+      const providers = [
+        [OpenAI.configure({ apiKey: "test" }).chat("model"), "openai"],
+        [OpenAI.configure({ apiKey: "test" }).responses("model"), "openai"],
+        [Azure.configure({ resourceName: "resource", apiKey: "test" }).chat("model"), "azure"],
+        [Azure.configure({ resourceName: "resource", apiKey: "test" }).responses("model"), "azure"],
+        [AmazonBedrock.configure({ apiKey: "test" }).model("model"), "bedrock"],
+        [AmazonBedrockMantle.configure({ apiKey: "test" }).chat("model"), "mantle"],
+        [AmazonBedrockMantle.configure({ apiKey: "test" }).responses("model"), "mantle"],
+        [Google.configure({ apiKey: "test" }).model("model"), "google"],
+        [GoogleVertex.configure(vertex).model("model"), "vertex"],
+        [GoogleVertexChat.configure(vertex).model("model"), "vertex"],
+        [GoogleVertexResponses.configure(vertex).model("model"), "vertex"],
+        [GoogleVertexMessages.configure(vertex).model("model"), "anthropic"],
+        [Anthropic.configure({ apiKey: "test" }).model("model"), "anthropic"],
+        [
+          AnthropicCompatible.configure({ baseURL: "https://example.test/v1", provider: "minimax" }).model("model"),
+          "minimax",
+        ],
+        [
+          OpenAICompatible.configure({ baseURL: "https://example.test/v1", provider: "custom" }).model("model"),
+          "custom",
+        ],
+        [
+          OpenAICompatibleResponses.configure({ baseURL: "https://example.test/v1", provider: "custom" }).model(
+            "model",
+          ),
+          "custom",
+        ],
+        [Cerebras.configure({ apiKey: "test" }).model("model"), "cerebras"],
+        [DeepInfra.configure({ apiKey: "test" }).model("model"), "deepinfra"],
+        [TogetherAI.configure({ apiKey: "test" }).model("model"), "togetherai"],
+        [CloudflareAIGateway.configure({ accountId: "account" }).model("model"), "cloudflare-ai-gateway"],
+        [CloudflareWorkersAI.configure({ accountId: "account" }).model("model"), "cloudflare-workers-ai"],
+        [OpenRouter.configure({ apiKey: "test" }).model("model"), "openrouter"],
+        [XAI.configure({ apiKey: "test" }).chat("model"), "xai"],
+        [XAI.configure({ apiKey: "test" }).responses("model"), "xai"],
+      ] as const
+
+      for (const [model, key] of providers) expect(model.route.providerMetadataKey).toBe(key)
+    }),
+  )
+
   it.effect("preserves native Together AI and Cerebras provider and route identities", () =>
     Effect.gen(function* () {
       const together = TogetherAI.configure({ apiKey: "fixture" }).model("meta-llama/Llama-3.3-70B")

packages/ai/test/provider/openai-chat-reasoning.recorded.test.ts

@@ -68,11 +68,13 @@ for (const item of cases) {
           expect(response.text.replaceAll(",", "").trim()).toBe("37887")
           expect(response.reasoning.length).toBeGreaterThan(0)
           expect(response.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
-          const metadata = response.message.content.find((part) => part.type === "reasoning")?.providerMetadata
-          expect(metadata?.openai?.reasoningField).toBe(item.structured ? "reasoning" : "reasoning_content")
-          expect(Array.isArray(metadata?.openai?.reasoningDetails)).toBe(item.structured)
+          const metadata = response.message.content.find((part) => part.type === "reasoning")?.providerMetadata?.[
+            item.model.route.providerMetadataKey ?? String(item.model.provider)
+          ]
+          expect(metadata?.reasoningField).toBe(item.structured ? "reasoning" : "reasoning_content")
+          expect(Array.isArray(metadata?.reasoningDetails)).toBe(item.structured)
           if (!item.structured) return
-          const details = metadata?.openai?.reasoningDetails
+          const details = metadata?.reasoningDetails
           if (!Array.isArray(details)) return
           expect(
             details.some(
@@ -126,7 +128,11 @@ for (const item of cases) {
           ).toMatch(/^Paris is sunny\.?$/)
           const details = events
             .filter(LLMEvent.is.reasoningEnd)
-            .map((event) => event.providerMetadata?.openai?.reasoningDetails)
+            .map(
+              (event) =>
+                event.providerMetadata?.[item.model.route.providerMetadataKey ?? String(item.model.provider)]
+                  ?.reasoningDetails,
+            )
             .find(Array.isArray)
           expect(Array.isArray(details)).toBe(item.structured)
           if (!item.structured || !Array.isArray(details)) return

packages/ai/test/provider/openai-chat.test.ts

@@ -903,6 +903,70 @@ describe("OpenAI Chat route", () => {
     }),
   )
 
+  it.effect("uses the configured provider metadata namespace for reasoning and usage", () =>
+    Effect.gen(function* () {
+      const selected = LanguageModel.update(model, {
+        route: { ...model.route, providerMetadataKey: "vendor" },
+      })
+      const details = [{ type: "reasoning.text", text: "thinking", signature: "signed" }]
+      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: selected })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { choices: [{ delta: { reasoning: "thinking", reasoning_details: details } }] },
+              deltaChunk({ content: "Hello" }),
+              deltaChunk({}, "stop"),
+              usageChunk({ prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 }),
+            ),
+          ),
+        ),
+      )
+
+      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
+        vendor: { reasoningField: "reasoning", reasoningDetails: details },
+      })
+      expect(response.usage?.providerMetadata).toEqual({
+        vendor: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
+      })
+
+      const replay = yield* compileRequest(LLM.request({ model: selected, messages: [response.message] }))
+      expect(replay.body.messages).toEqual([
+        { role: "assistant", content: "Hello", reasoning: "thinking", reasoning_details: details },
+      ])
+    }),
+  )
+
+  it.effect("falls back to the selected provider for the metadata namespace", () =>
+    Effect.gen(function* () {
+      const compatible = model.route.with({ provider: "deepseek" }).model({ id: "deepseek-chat" })
+      const selected = LanguageModel.update(compatible, {
+        route: { ...compatible.route, providerMetadataKey: undefined },
+      })
+      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: selected })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              deltaChunk({ reasoning_content: "thinking" }),
+              deltaChunk({ content: "Hello" }),
+              deltaChunk({}, "stop"),
+              usageChunk({ prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 }),
+            ),
+          ),
+        ),
+      )
+
+      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
+        deepseek: { reasoningField: "reasoning_content" },
+      })
+      expect(response.usage?.providerMetadata).toEqual({
+        deepseek: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
+      })
+
+      const replay = yield* compileRequest(LLM.request({ model: selected, messages: [response.message] }))
+      expect(replay.body.messages).toEqual([{ role: "assistant", content: "Hello", reasoning_content: "thinking" }])
+    }),
+  )
+
   it.effect("parses and replays a configured custom reasoning field", () =>
     Effect.gen(function* () {
       const custom = LanguageModel.update(model, { compatibility: { reasoningField: "vendor_reasoning" } })

packages/ai/test/provider/openai-compatible-chat.test.ts

@@ -437,7 +437,7 @@ describe("OpenAI-compatible Chat route", () => {
         outputTokens: undefined,
         totalTokens: undefined,
         providerMetadata: {
-          openai: {
+          deepseek: {
             prompt_tokens: null,
             completion_tokens: null,
             total_tokens: null,

packages/ai/test/provider/openai-compatible-responses.test.ts

@@ -195,19 +195,19 @@ describe("Open Responses-compatible route", () => {
           model,
           messages: [
             Message.assistant([
-              { type: "text", text: "Kept.", providerMetadata: { openresponses: { itemId: "history_1" } } },
+              { type: "text", text: "Kept.", providerMetadata: { "openai-compatible": { itemId: "history_1" } } },
               {
                 type: "text",
                 text: "Long.",
-                providerMetadata: { openresponses: { itemId: `history_${"a".repeat(64)}` } },
+                providerMetadata: { "openai-compatible": { itemId: `history_${"a".repeat(64)}` } },
               },
               {
                 type: "text",
                 text: "Opaque.",
-                providerMetadata: { openresponses: { itemId: "provider_value/with+symbols" } },
+                providerMetadata: { "openai-compatible": { itemId: "provider_value/with+symbols" } },
               },
-              { type: "text", text: "No suffix.", providerMetadata: { openresponses: { itemId: "msg_" } } },
-              { type: "text", text: "No prefix.", providerMetadata: { openresponses: { itemId: "_item" } } },
+              { type: "text", text: "No suffix.", providerMetadata: { "openai-compatible": { itemId: "msg_" } } },
+              { type: "text", text: "No prefix.", providerMetadata: { "openai-compatible": { itemId: "_item" } } },
             ]),
           ],
         }),
@@ -267,7 +267,7 @@ describe("Open Responses-compatible route", () => {
               name: item.type,
               result: { type: "json", value: item },
               providerExecuted: true,
-              providerMetadata: { openresponses: { itemId: item.id } },
+              providerMetadata: { example: { itemId: item.id } },
             }),
           ),
         }),
@@ -302,11 +302,296 @@ describe("Open Responses-compatible route", () => {
       )
 
       expect(response.message.content).toEqual([
-        { type: "text", text: "Indexed", providerMetadata: { openresponses: { itemId: "msg_1" } } },
+        { type: "text", text: "Indexed", providerMetadata: { "openai-compatible": { itemId: "msg_1" } } },
       ])
     }),
   )
 
+  describe("stream validation", () => {
+    const request = LLM.request({
+      model: configure({ apiKey: "test-key", baseURL: "https://responses.example.test/v1" }).model("example-model"),
+      prompt: "Respond.",
+    })
+
+    const fixtures = [
+      {
+        item: { type: "message" },
+        events: [
+          { type: "response.output_text.delta", delta: "Preserved" },
+          { type: "response.output_text.done", text: "Preserved" },
+          { type: "response.refusal.delta", delta: "Preserved" },
+          { type: "response.refusal.done", refusal: "Preserved" },
+        ],
+      },
+      {
+        item: { type: "reasoning", encrypted_content: "encrypted-state" },
+        events: [
+          { type: "response.reasoning.delta", delta: "Preserved" },
+          { type: "response.reasoning.done", text: "Preserved" },
+          { type: "response.reasoning_summary_text.delta", delta: "Preserved" },
+          { type: "response.reasoning_summary_text.done", text: "Preserved" },
+          { type: "response.reasoning_text.done", text: "Preserved" },
+        ],
+      },
+      {
+        item: { type: "function_call", call_id: "call_1", name: "lookup" },
+        events: [
+          { type: "response.function_call_arguments.delta", delta: '{"query":"Preserved"}' },
+          { type: "response.function_call_arguments.done", arguments: '{"query":"Preserved"}' },
+        ],
+      },
+    ]
+
+    const routings = [
+      { name: "empty item and event IDs", id: "", item_id: "" },
+      { name: "empty event ID with registered index", id: "item_1", item_id: "", output_index: 2 },
+      { name: "empty stored ID with registered index", id: "", item_id: "wrong_item", output_index: 2 },
+      { name: "empty item and event IDs with registered index", id: "", item_id: "", output_index: 2 },
+    ]
+
+    fixtures.forEach((fixture) => {
+      fixture.events.forEach((event) => {
+        routings.forEach((routing) => {
+          it.effect(`${event.type} preserves content with ${routing.name}`, () =>
+            Effect.gen(function* () {
+              const item = { ...fixture.item, id: routing.id }
+              const response = yield* LLMClient.generate(request).pipe(
+                Effect.provide(
+                  fixedResponse(
+                    sseEvents(
+                      { type: "response.output_item.added", output_index: routing.output_index, item },
+                      { ...event, item_id: routing.item_id, output_index: routing.output_index },
+                      { type: "response.output_item.done", output_index: routing.output_index, item },
+                      { type: "response.completed", response: { id: "resp_1" } },
+                    ),
+                  ),
+                ),
+              )
+
+              const metadata = { "openai-compatible": { itemId: routing.id } }
+              if (fixture.item.type === "function_call") {
+                expect(response.toolCalls).toEqual([
+                  expect.objectContaining({
+                    id: "call_1",
+                    name: "lookup",
+                    input: { query: "Preserved" },
+                    providerMetadata: metadata,
+                  }),
+                ])
+                return
+              }
+              if (fixture.item.type === "reasoning") {
+                expect(response.message.content).toEqual([
+                  {
+                    type: "reasoning",
+                    text: "Preserved",
+                    providerMetadata: {
+                      "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "encrypted-state" },
+                    },
+                  },
+                ])
+                expect(response.events.filter(LLMEvent.is.reasoningEnd)).toHaveLength(1)
+                return
+              }
+              expect(response.message.content).toEqual([
+                { type: "text", text: "Preserved", providerMetadata: metadata },
+              ])
+              expect(response.events.filter(LLMEvent.is.textEnd)).toEqual([
+                expect.objectContaining({ id: routing.id, providerMetadata: metadata }),
+              ])
+            }),
+          )
+        })
+      })
+    })
+
+    routings.forEach((routing) => {
+      it.effect(`preserves reasoning summary boundaries and terminal metadata with ${routing.name}`, () =>
+        Effect.gen(function* () {
+          const address = { item_id: routing.item_id, output_index: routing.output_index }
+          const response = yield* LLMClient.generate(request).pipe(
+            Effect.provide(
+              fixedResponse(
+                sseEvents(
+                  {
+                    type: "response.output_item.added",
+                    output_index: routing.output_index,
+                    item: { type: "reasoning", id: routing.id },
+                  },
+                  { type: "response.reasoning_summary_part.added", ...address, summary_index: 0 },
+                  { type: "response.reasoning_summary_text.delta", ...address, summary_index: 0, delta: "First." },
+                  { type: "response.reasoning_summary_text.done", ...address, summary_index: 0, text: "First." },
+                  { type: "response.reasoning_summary_part.done", ...address, summary_index: 0 },
+                  { type: "response.reasoning_summary_part.added", ...address, summary_index: 1 },
+                  { type: "response.reasoning_summary_text.done", ...address, summary_index: 1, text: "Second." },
+                  { type: "response.reasoning_summary_part.done", ...address, summary_index: 1 },
+                  {
+                    type: "response.completed",
+                    response: { output: [{ type: "reasoning", id: routing.id, encrypted_content: "final-state" }] },
+                  },
+                ),
+              ),
+            ),
+          )
+
+          expect(response.message.content).toEqual([
+            {
+              type: "reasoning",
+              text: "First.",
+              providerMetadata: { "openai-compatible": { itemId: routing.id } },
+            },
+            {
+              type: "reasoning",
+              text: "Second.",
+              providerMetadata: {
+                "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "final-state" },
+              },
+            },
+          ])
+          expect(response.events.filter(LLMEvent.is.reasoningEnd)).toEqual([
+            expect.objectContaining({
+              id: `${routing.id}:0`,
+              providerMetadata: { "openai-compatible": { itemId: routing.id } },
+            }),
+            expect.objectContaining({
+              id: `${routing.id}:1`,
+              providerMetadata: {
+                "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "final-state" },
+              },
+            }),
+          ])
+        }),
+      )
+    })
+
+    it.effect("reconciles pending empty-ID function arguments from completed output", () =>
+      Effect.gen(function* () {
+        const item = { type: "function_call", id: "", call_id: "call_1", name: "lookup" }
+        const response = yield* LLMClient.generate(request).pipe(
+          Effect.provide(
+            fixedResponse(
+              sseEvents(
+                { type: "response.output_item.added", item },
+                { type: "response.function_call_arguments.delta", item_id: "", delta: '{"query":"partial' },
+                {
+                  type: "response.completed",
+                  response: { output: [{ ...item, arguments: '{"query":"complete"}' }] },
+                },
+              ),
+            ),
+          ),
+        )
+
+        expect(response.toolCalls).toEqual([
+          expect.objectContaining({
+            id: "call_1",
+            name: "lookup",
+            input: { query: "complete" },
+            providerMetadata: { "openai-compatible": { itemId: "" } },
+          }),
+        ])
+      }),
+    )
+
+    it.effect("treats null output items as no-ops without disturbing registered items", () =>
+      Effect.gen(function* () {
+        const response = yield* LLMClient.generate(request).pipe(
+          Effect.provide(
+            fixedResponse(
+              sseEvents(
+                { type: "response.output_item.added", output_index: 0, item: null },
+                { type: "response.output_item.done", output_index: 0, item: null },
+                { type: "response.output_item.added", output_index: 0, item: { type: "message", id: "msg_1" } },
+                { type: "response.output_text.delta", output_index: 0, item_id: "wrong_item", delta: "Before " },
+                { type: "response.output_item.added", output_index: 0, item: null },
+                { type: "response.output_item.done", output_index: 0, item: null },
+                { type: "response.output_text.delta", output_index: 0, item_id: "wrong_item", delta: "after" },
+                { type: "response.output_item.done", output_index: 0, item: { type: "message", id: "msg_1" } },
+                { type: "response.completed", response: { id: "resp_1" } },
+              ),
+            ),
+          ),
+        )
+
+        expect(response.message.content).toEqual([
+          { type: "text", text: "Before after", providerMetadata: { "openai-compatible": { itemId: "msg_1" } } },
+        ])
+        expect(response.events.map((event) => event.type)).toEqual([
+          "step-start",
+          "text-start",
+          "text-delta",
+          "text-delta",
+          "text-end",
+          "step-finish",
+          "finish",
+        ])
+      }),
+    )
+
+    it.effect("rejects missing, null, and non-string event IDs even with a registered output index", () =>
+      Effect.gen(function* () {
+        yield* Effect.forEach(
+          [
+            ...fixtures.flatMap((fixture) => fixture.events.map((event) => ({ item: fixture.item, event }))),
+            ...["response.reasoning_summary_part.added", "response.reasoning_summary_part.done"].map((type) => ({
+              item: { type: "reasoning" },
+              event: { type, summary_index: 0 },
+            })),
+          ],
+          (fixture) =>
+            Effect.forEach([undefined, null, 0, false, {}, []], (item_id) =>
+              Effect.gen(function* () {
+                const error = yield* LLMClient.generate(request).pipe(
+                  Effect.provide(
+                    fixedResponse(
+                      sseEvents(
+                        {
+                          type: "response.output_item.added",
+                          output_index: 0,
+                          item: { ...fixture.item, id: "item_1" },
+                        },
+                        { ...fixture.event, output_index: 0, item_id },
+                        { type: "response.completed", response: { id: "resp_1" } },
+                      ),
+                    ),
+                  ),
+                  Effect.flip,
+                )
+                expect(error.reason._tag).toBe("InvalidProviderOutput")
+              }),
+            ),
+        )
+      }),
+    )
+
+    it.effect("keeps malformed output item IDs invalid", () =>
+      Effect.gen(function* () {
+        yield* Effect.forEach(["response.output_item.added", "response.output_item.done"], (type) =>
+          Effect.forEach(fixtures, (fixture) =>
+            Effect.forEach(
+              fixture.item.type === "message" ? [undefined, null, 0, false, {}, []] : [null, 0, false, {}, []],
+              (id) =>
+                Effect.gen(function* () {
+                  const error = yield* LLMClient.generate(request).pipe(
+                    Effect.provide(
+                      fixedResponse(
+                        sseEvents(
+                          { type, item: { ...fixture.item, id } },
+                          { type: "response.completed", response: { id: "resp_1" } },
+                        ),
+                      ),
+                    ),
+                    Effect.flip,
+                  )
+                  expect(error.reason._tag).toBe("InvalidProviderOutput")
+                }),
+            ),
+          ),
+        )
+      }),
+    )
+  })
+
   it.effect("streams function calls without optional item ids through the shared baseline", () =>
     Effect.gen(function* () {
       const model = configure({
@@ -381,7 +666,7 @@ describe("Open Responses-compatible route", () => {
 
       expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
         input: { query: "complete" },
-        providerMetadata: { openresponses: { itemId: "item_1" } },
+        providerMetadata: { example: { itemId: "item_1" } },
       })
     }),
   )
@@ -413,7 +698,7 @@ describe("Open Responses-compatible route", () => {
       )
 
       expect(response.events.find((event) => event.type === "reasoning-end")).toMatchObject({
-        providerMetadata: { openresponses: { itemId: "rs_raw", reasoningEncryptedContent: "raw-state" } },
+        providerMetadata: { "openai-compatible": { itemId: "rs_raw", reasoningEncryptedContent: "raw-state" } },
       })
     }),
   )
@@ -462,7 +747,7 @@ describe("Open Responses-compatible route", () => {
             Message.assistant({
               type: "text",
               text: "Unclassified.",
-              providerMetadata: { openresponses: { phase: null } },
+              providerMetadata: { "openai-compatible": { phase: null } },
             }),
           ],
         }),
@@ -521,7 +806,7 @@ describe("Open Responses-compatible route", () => {
         {
           type: "text",
           text: "I can't help with that.",
-          providerMetadata: { openresponses: { itemId: "msg_refusal" } },
+          providerMetadata: { example: { itemId: "msg_refusal" } },
         },
       ])
 
@@ -610,7 +895,7 @@ describe("Open Responses-compatible route", () => {
 
       expect(response.toolCalls).toEqual([])
       expect(response.events.find(LLMEvent.is.finish)).toMatchObject({
-        providerMetadata: { openresponses: { responseId: "resp_1" } },
+        providerMetadata: { example: { responseId: "resp_1" } },
       })
     }),
   )

packages/ai/test/provider/openai-responses.test.ts

@@ -2229,6 +2229,35 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
+  it.effect("accepts empty IDs for native reasoning text deltas", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.added", output_index: 1, item: { type: "reasoning", id: "" } },
+              { type: "response.reasoning_text.delta", output_index: 1, item_id: "", delta: "Raw" },
+              {
+                type: "response.output_item.done",
+                output_index: 1,
+                item: { type: "reasoning", id: "", encrypted_content: "state" },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.message.content).toEqual([
+        {
+          type: "reasoning",
+          text: "Raw",
+          providerMetadata: { openai: { itemId: "", reasoningEncryptedContent: "state" } },
+        },
+      ])
+    }),
+  )
+
   it.effect("falls back to item ids when an output index was not registered", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(request).pipe(

packages/ai/test/provider/openrouter.test.ts

@@ -295,7 +295,7 @@ describe("OpenRouter", () => {
               {
                 type: "reasoning",
                 text: "Thinking",
-                providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: details } },
+                providerMetadata: { openrouter: { reasoningField: "reasoning", reasoningDetails: details } },
               },
             ]),
           ],
@@ -328,7 +328,7 @@ describe("OpenRouter", () => {
             Message.assistant({
               type: "reasoning",
               text: "Thinking",
-              providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: details } },
+              providerMetadata: { openrouter: { reasoningField: "reasoning", reasoningDetails: details } },
             }),
           ],
         }),
@@ -354,7 +354,7 @@ describe("OpenRouter", () => {
             Message.assistant({
               type: "reasoning",
               text: "AB",
-              providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: details } },
+              providerMetadata: { openrouter: { reasoningField: "reasoning", reasoningDetails: details } },
             }),
           ],
         }),

packages/ai/test/route.test.ts

@@ -40,4 +40,13 @@ describe("Route.with", () => {
       "x-patch": "patch",
     })
   })
+
+  test("assigns metadata ownership to a replacement provider and preserves explicit overrides", () => {
+    const route = OpenAIChat.route.with({ provider: "azure" })
+    const overridden = route.with({ providerMetadataKey: "custom-azure" }).with({ headers: { "x-test": "value" } })
+
+    expect(route.providerMetadataKey).toBe("azure")
+    expect(overridden.providerMetadataKey).toBe("custom-azure")
+    expect(overridden.defaults).not.toHaveProperty("providerMetadataKey")
+  })
 })

packages/app/.gitignore

@@ -1,3 +1,5 @@
 src/assets/theme.css
 e2e/test-results
 e2e/playwright-report
+component-tests/test-results
+component-tests/playwright-report

packages/app/component-tests/composer.spec.ts

@@ -0,0 +1,26 @@
+import { expect, story } from "../../storybook/playwright/story"
+
+// Moved from packages/app/e2e/regression/prompt-thinking-level.spec.ts
+story("shows the thinking level control while relevant", async ({ mount, page }) => {
+  const component = await mount("opencode-composer-flow--model-and-variant")
+  const composer = component.locator('[data-component="composer"]')
+  const input = composer.locator('[data-component="composer-editor"]')
+  const control = composer.getByRole("button", { name: "Choose model variant" })
+
+  await page.mouse.move(0, 0)
+  await page.evaluate(() => (document.activeElement as HTMLElement | null)?.blur())
+  await expect(control).toBeVisible()
+
+  await control.click()
+  const high = page.getByRole("menuitemradio", { name: "high" })
+  await expect(high).toBeVisible()
+  await page.mouse.move(0, 0)
+  await expect(control).toBeVisible()
+  await expect(high).toBeVisible()
+  await high.click()
+
+  await input.focus()
+  await expect(control).toBeVisible()
+  await input.blur()
+  await expect(control).toBeVisible()
+})

packages/app/e2e/regression/review-line-comment.spec.ts

@@ -12,54 +12,6 @@ test.beforeEach(async ({ page }) => {
   await openReview(page)
 })
 
-test("opens the comment editor when code is clicked", async ({ page }) => {
-  const review = page.locator('[data-component="session-review"]')
-  const line = review.getByText("export const value = 'after'", { exact: true })
-  await expectAppVisible(line)
-  await line.click()
-
-  await expect(review.getByRole("textbox")).toBeVisible()
-  await expect(review.locator('[data-slot="line-comment-editor-label"]')).toHaveText("Commenting on line 2")
-})
-
-test("opens the comment editor when a line number is clicked", async ({ page }) => {
-  const review = page.locator('[data-component="session-review"]')
-  const lineNumber = review.locator('[data-column-number="1"]').last()
-  await expectAppVisible(lineNumber)
-  await lineNumber.click()
-
-  await expect(review.getByRole("textbox")).toBeVisible()
-  await expect(review.locator('[data-slot="line-comment-editor-label"]')).toHaveText("Commenting on line 1")
-})
-
-test("opens the comment editor for a line number range", async ({ page }) => {
-  const review = page.locator('[data-component="session-review"]')
-  const start = review.locator('[data-column-number="1"]').last()
-  const end = review.locator('[data-column-number="3"]').last()
-  await expectAppVisible(start)
-  await expectAppVisible(end)
-
-  await start.dragTo(end)
-
-  await expect(review.getByRole("textbox")).toBeVisible()
-  await expect(review.locator('[data-slot="line-comment-editor-label"]')).toHaveText("Commenting on lines 1-3")
-})
-
-test("shows a comment button when a diff line is hovered", async ({ page }) => {
-  const review = page.locator('[data-component="session-review"]')
-  const line = review.getByText("export const first = 1", { exact: true })
-  await expectAppVisible(line)
-
-  const comment = review.getByRole("button", { name: "Comment", exact: true, includeHidden: true })
-  await expect(comment).toHaveCount(1)
-  await line.dispatchEvent("pointermove", { pointerType: "mouse", bubbles: true, composed: true })
-  await expect(comment).toBeVisible()
-  await expect(comment).toHaveCSS("pointer-events", "auto")
-  await comment.dispatchEvent("click")
-  await expect(review.getByRole("textbox")).toBeVisible()
-  await expect(review.locator('[data-slot="line-comment-editor-label"]')).toHaveText("Commenting on line 1")
-})
-
 test("stages a submitted line comment in the prompt context", async ({ page }) => {
   page.on("request", (request) => {
     expect.soft(request.method(), `unexpected ${request.method()} ${new URL(request.url()).pathname}`).toBe("GET")

packages/app/e2e/regression/session-queue.spec.ts

@@ -201,12 +201,15 @@ test("editing restores the existing draft and replaces only the original queue p
   await view.input.fill("my in-progress draft")
   await original.click()
   await expect(view.input).toHaveText("tighten the error copy")
+  await expect(view.input).toBeFocused()
   await view.input.press("Escape")
   await expect(view.input).toHaveText("my in-progress draft")
 
   await original.click()
   await expect(view.input).toHaveText("tighten the error copy")
+  await expect(view.input).toBeFocused()
   await view.input.fill("tighten the error copy and add a retry hint")
+  await expect(view.input).toHaveText("tighten the error copy and add a retry hint")
   await view.input.press("Enter")
 
   await expect(view.rows.locator('[data-action="session-queue-edit"]')).toHaveText([

packages/app/e2e/regression/session-timeline-accessibility.spec.ts

@@ -1,16 +1,43 @@
 import { expect, test } from "@playwright/test"
-import { assistantMessage, setupTimeline, shell, userMessage } from "../performance/timeline-stability/fixture"
+import {
+  assistantMessage,
+  setupTimeline,
+  shell,
+  textPart,
+  userMessage,
+} from "../performance/timeline-stability/fixture"
 
 test("space activates a focused timeline button instead of scrolling", async ({ page }) => {
   const shellID = "prt_space_button_shell"
   await setupTimeline(page, {
-    messages: [userMessage(), assistantMessage([shell(shellID, "completed", lines(5))])],
+    messages: [
+      userMessage(),
+      assistantMessage([
+        shell(shellID, "completed", lines(5)),
+        textPart(
+          "prt_space_following",
+          "Following content leaves room to focus the command away from the bottom. ".repeat(40),
+        ),
+      ]),
+    ],
     settings: { shellToolPartsExpanded: false },
     reducedMotion: true,
+    seedHistory: true,
   })
   const scroller = page.locator(".scroll-view__viewport", { has: page.locator("[data-timeline-row]") })
   const trigger = page.getByRole("button", { name: "Used Shell" })
+  await expect
+    .poll(() => scroller.evaluate((element) => element.scrollHeight - element.clientHeight))
+    .toBeGreaterThan(300)
+  await trigger.scrollIntoViewIfNeeded()
+  await scroller.hover()
+  await page.mouse.wheel(0, -100)
+  await expect
+    .poll(() => scroller.evaluate((element) => element.scrollHeight - element.clientHeight - element.scrollTop))
+    .toBeGreaterThan(50)
+  await expect(trigger).toBeInViewport()
   await trigger.focus()
+  await expect(trigger).toBeFocused()
   const before = await scroller.evaluate((element) => element.scrollTop)
   await trigger.press("Space")
   await expect(trigger).toHaveAttribute("aria-expanded", "true")

packages/app/e2e/regression/session-timeline-file-projection.spec.ts

@@ -8,21 +8,6 @@ import {
   userText,
 } from "../performance/timeline-stability/fixture"
 
-test("renders completed write content", async ({ page }) => {
-  const id = "prt_file_projection_write"
-  await setupTimeline(page, {
-    messages: [
-      userMessage(),
-      assistantMessage([
-        toolPart(id, "write", "completed", { path: "src/write.ts", content: "export const written = true\n" }),
-      ]),
-    ],
-    settings: { editToolPartsExpanded: true },
-  })
-
-  await expect(page.locator(`[data-timeline-part-id="${id}"] [data-component="write-content"]`)).toBeVisible()
-})
-
 test("renders a completed single-file patch", async ({ page }) => {
   const id = "prt_file_projection_single_patch"
   await setupTimeline(page, {

packages/app/e2e/regression/session-timeline-file-state.spec.ts

@@ -1,53 +0,0 @@
-import { expect, test } from "@playwright/test"
-import { assistantMessage, setupTimeline, toolPart, userMessage } from "../performance/timeline-stability/fixture"
-import { createTwoFilesPatch } from "diff"
-
-test("keeps patch file disclosures independent", async ({ page }) => {
-  const patchID = "prt_nested_patch"
-  const files = [patchFile("src/a.ts", "modified"), patchFile("src/b.ts", "added"), patchFile("src/old.ts", "deleted")]
-  await setupTimeline(page, {
-    messages: [
-      userMessage(),
-      assistantMessage([
-        toolPart(
-          patchID,
-          "patch",
-          "completed",
-          { patchText: "Update three files" },
-          { metadata: { files } },
-        ),
-      ]),
-    ],
-    settings: { editToolPartsExpanded: true },
-  })
-  const wrapper = page.locator(`[data-timeline-part-id="${patchID}"]`)
-  const modified = wrapper.locator('[data-scope="apply-patch"] [data-type="update"]')
-  const deleted = wrapper.locator('[data-scope="apply-patch"] [data-type="delete"]')
-  await expect(wrapper.locator('[data-scope="apply-patch"] [aria-expanded="false"]')).toHaveCount(3)
-  await deleted.getByRole("button").click()
-  await expect(deleted.getByRole("button")).toHaveAttribute("aria-expanded", "true")
-  await expect(modified.getByRole("button")).toHaveAttribute("aria-expanded", "false")
-  await modified.getByRole("button").click()
-  await expect(modified.getByRole("button")).toHaveAttribute("aria-expanded", "true")
-  await deleted.getByRole("button").click()
-  await expect(deleted.getByRole("button")).toHaveAttribute("aria-expanded", "false")
-  await expect(modified.getByRole("button")).toHaveAttribute("aria-expanded", "true")
-})
-
-function patchFile(file: string, status: "added" | "modified" | "deleted") {
-  const before = status === "added" ? "" : source(false)
-  const after = status === "deleted" ? "" : source(true)
-  return {
-    file,
-    status,
-    patch: createTwoFilesPatch(`a/${file}`, `b/${file}`, before, after),
-    additions: status === "deleted" ? 0 : 4,
-    deletions: status === "added" ? 0 : 3,
-  }
-}
-
-function source(changed: boolean) {
-  return Array.from({ length: 12 }, (_, index) => `export const value${index} = ${changed ? index + 1 : index}\n`).join(
-    "",
-  )
-}

packages/app/e2e/regression/session-timeline-lifecycle-state.spec.ts

@@ -122,6 +122,7 @@ test("transitions thinking and hidden reasoning through busy to idle", async ({
   await expect(page.locator('[data-timeline-row="Thinking"]')).toBeVisible()
   await expect(page.getByText("Inspecting stability", { exact: true })).toBeVisible()
   await expect(page.locator(`[data-timeline-part-id="${reasoningID}"]`)).toHaveCount(0)
+  await expect(page.locator(`[data-timeline-part-id="${renderedPartID(reasoningID)}"]`)).toHaveCount(0)
   await timeline.send(partUpdated(shell("prt_reasoning_shell", "running")), 160)
   await expect(page.locator('[data-timeline-row="Thinking"]')).toBeVisible()
   await timeline.send(partUpdated(shell("prt_reasoning_shell", "completed", "done")), 180)
@@ -129,6 +130,7 @@ test("transitions thinking and hidden reasoning through busy to idle", async ({
   await timeline.send(status("idle"), 300)
   await expect(page.locator('[data-timeline-row="Thinking"]')).toHaveCount(0)
   await expect(page.locator(`[data-timeline-part-id="${reasoningID}"]`)).toHaveCount(0)
+  await expect(page.locator(`[data-timeline-part-id="${renderedPartID(reasoningID)}"]`)).toHaveCount(0)
 })
 
 test("moves busy through retry and recovery to final idle content", async ({ page }) => {

packages/app/e2e/regression/session-timeline-notices.spec.ts

@@ -122,6 +122,7 @@ test("updates running compactions to failed and cancelled boundaries", async ({
 
   await timeline.send(compactionStarted({ sessionID, reason: "auto", recent: "" }))
   await timeline.send(compactionDelta({ sessionID, text: "Partial summary that should be discarded." }))
+  await expect(page.getByText("Partial summary that should be discarded.", { exact: true })).toBeVisible()
   await timeline.send(
     compactionFailed({
       sessionID,
@@ -140,6 +141,9 @@ test("updates running compactions to failed and cancelled boundaries", async ({
   await expect(failed).not.toContainText("Partial summary that should be discarded.")
 
   await timeline.send(compactionStarted({ sessionID, reason: "manual", recent: "" }))
+  await expect(compactions).toHaveCount(2)
+  await timeline.send(compactionDelta({ sessionID, text: "Summary before cancellation." }))
+  await expect(page.getByText("Summary before cancellation.", { exact: true })).toBeVisible()
   await timeline.send(
     compactionFailed({
       sessionID,
@@ -152,88 +156,7 @@ test("updates running compactions to failed and cancelled boundaries", async ({
   const cancelled = compactions.filter({ hasNotText: "The provider rejected the summary." })
   await expect(cancelled.getByText("Session compacted", { exact: true })).toBeVisible()
   await expect(cancelled).not.toContainText("Cancellation detail should stay hidden.")
-})
-
-test("shows a delegating row while subagent input streams", async ({ page }) => {
-  await setupTimeline(page, {
-    sessionMessages: [
-      user,
-      {
-        ...assistant(false),
-        content: [
-          {
-            type: "tool",
-            id: "call_subagent",
-            name: "subagent",
-            state: { status: "streaming", input: "" },
-            time: { created: 2 },
-          },
-        ],
-      },
-    ],
-  })
-
-  const delegating = page.locator('[data-component="task-tool-delegating"]')
-  await expect(delegating).toBeVisible()
-  const shimmer = delegating.locator('[data-component="text-shimmer"]')
-  await expect(shimmer).toHaveAttribute("aria-label", "Delegating agent...")
-  await expect(shimmer).toHaveCSS("line-height", "16px")
-  const icon = delegating.locator('[data-slot="icon-svg"]')
-  await expect(icon.locator('use[href="#opencode-v2-icon-subagent"]')).toBeVisible()
-  await expect(icon).toHaveCSS("color", "rgb(174, 174, 174)")
-  await expect(page.locator('[data-component="task-tool-card"]')).toHaveCount(0)
-  await expect(page.locator('[data-timeline-row="Thinking"]')).toHaveCount(0)
-})
-
-test("renders the moved location notice in its compact timeline style", async ({ page }) => {
-  const directory = `/Users/usrnk1/Developer/opencode/${"nested-directory/".repeat(24)}session`
-  await page.setViewportSize({ width: 480, height: 720 })
-  await setupTimeline(page, {
-    sessionMessages: [
-      user,
-      {
-        id: "msg_location",
-        type: "location-switched",
-        location: { directory },
-        time: { created: 2 },
-      },
-    ],
-  })
-
-  const notice = page.locator('[data-slot="session-timeline-notice"][data-type="location-switched"]')
-  const label = notice.locator('[data-slot="session-timeline-notice-label"]')
-  const value = notice.locator('[data-slot="session-timeline-notice-value"]')
-  const tooltipTrigger = notice.locator('[data-component="tooltip-v2-trigger"]')
-
-  await expect(label).toHaveText("Moved to")
-  await expect(value).toHaveText(directory)
-  await expect(notice).not.toContainText("·")
-  await expect(notice.locator("svg")).toHaveCount(0)
-  await expect(notice).toHaveCSS("height", "28px")
-  await expect(notice).toHaveCSS("gap", "8px")
-  await expect(notice).toHaveCSS("padding-top", "4px")
-  await expect(notice).toHaveCSS("padding-bottom", "4px")
-  await expect(label).toHaveCSS("font-size", "13px")
-  await expect(label).toHaveCSS("font-weight", "530")
-  await expect(label).toHaveCSS("line-height", "16px")
-  await expect(label).toHaveCSS("color", "rgb(128, 128, 128)")
-  await expect(value).toHaveCSS("font-size", "13px")
-  await expect(value).toHaveCSS("font-weight", "440")
-  await expect(value).toHaveCSS("line-height", "16px")
-  await expect(value).toHaveCSS("color", "rgb(128, 128, 128)")
-  await expect(value).toHaveCSS("text-overflow", "ellipsis")
-  await expect(value).toHaveCSS("white-space", "nowrap")
-  await expect(value).toHaveAttribute("dir", "ltr")
-  await expect.poll(() => value.evaluate((element) => element.scrollWidth > element.clientWidth)).toBe(true)
-
-  const tooltip = page.getByText("Session working directory changed", { exact: true })
-  await label.hover()
-  await expect(tooltip).toBeVisible()
-  await page.mouse.move(0, 0)
-  await expect(tooltip).toBeHidden()
-  await tooltipTrigger.focus()
-  await expect(tooltipTrigger).toBeFocused()
-  await expect(tooltip).toBeVisible()
+  await expect(cancelled).not.toContainText("Summary before cancellation.")
 })
 
 test("moves blocking work to the background with Ctrl+B", async ({ page }) => {
@@ -271,11 +194,6 @@ test("moves blocking work to the background with Ctrl+B", async ({ page }) => {
   await request
 })
 
-test("waits for completion before labeling requested background work", async ({ page }) => {
-  await setupTimeline(page, { sessionMessages: [user, assistant(false, true, undefined, true)] })
-  await expect(page.locator('[data-component="task-tool-card"]')).not.toContainText("(background)")
-})
-
 test("navigates from a running subagent card and hides background controls in the child", async ({ page }) => {
   const childID = "ses_running_child"
   await setupTimeline(page, {

packages/app/e2e/regression/session-timeline-projection.spec.ts

@@ -7,86 +7,9 @@ import {
   toolPart,
   userMessage,
   userText,
-  type PartSeed,
 } from "../performance/timeline-stability/fixture"
 
 test.describe("session timeline projection", () => {
-  test("renders every admitted tool family and hides timeline-only exclusions", async ({ page }) => {
-    const parts = [
-      toolPart("prt_01_read", "read", "completed", { path: "src/a.ts" }),
-      toolPart("prt_02_glob", "glob", "completed", { path: ".", pattern: "**/*.ts" }),
-      toolPart("prt_03_grep", "grep", "completed", { path: ".", pattern: "value" }),
-      toolPart("prt_04_list", "list", "completed", { path: "src" }),
-      toolPart("prt_webfetch", "webfetch", "completed", { url: "https://example.com" }),
-      toolPart(
-        "prt_websearch",
-        "websearch",
-        "completed",
-        { query: "timeline stability" },
-        { output: "https://example.com/result" },
-      ),
-      toolPart("prt_task", "subagent", "completed", {
-        description: "Inspect timeline",
-        agent: "explore",
-        prompt: "Inspect the timeline implementation.",
-      }),
-      toolPart(
-        "prt_bash",
-        "shell",
-        "completed",
-        { command: "printf stable" },
-        { output: "stable", title: "printf stable" },
-      ),
-      editPart("prt_edit"),
-      toolPart("prt_write", "write", "completed", { path: "src/new.ts", content: "export const stable = true\n" }),
-      patchPart("prt_patch"),
-      toolPart("prt_todo", "todowrite", "completed", { todos: [{ content: "Hidden", status: "pending" }] }),
-      toolPart(
-        "prt_question",
-        "question",
-        "completed",
-        { questions: [{ question: "Keep stable?", header: "Stability", options: [] }] },
-        { metadata: { answers: [["Yes"]] } },
-      ),
-      toolPart("prt_skill", "skill", "completed", { name: "stability" }),
-      toolPart("prt_custom", "custom_mcp_tool", "completed", { target: "timeline", count: 2 }),
-    ]
-    await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
-
-    const first = page.locator(
-      '[data-timeline-part-ids="prt_01_read,prt_02_glob,prt_03_grep,prt_04_list,prt_webfetch,prt_websearch,prt_task,prt_bash,prt_edit,prt_write,prt_patch"]',
-    )
-    const second = page.locator('[data-timeline-part-ids="prt_skill,prt_custom"]')
-    await expect(first).toBeVisible()
-    await expect(second).toBeVisible()
-    await first.getByRole("button").click()
-    await second.getByRole("button").click()
-    for (const id of [
-      "prt_webfetch",
-      "prt_websearch",
-      "prt_task",
-      "prt_bash",
-      "prt_edit",
-      "prt_write",
-      "prt_patch",
-      "prt_question",
-      "prt_skill",
-      "prt_custom",
-    ]) {
-      await expect(page.locator(`[data-timeline-part-id="${id}"]`).first(), id).toBeVisible()
-    }
-    const patch = page.locator('[data-timeline-part-id="prt_patch"]')
-    await expect(patch.getByText("1 file", { exact: true })).toBeVisible()
-    await expect(patch.getByRole("button", { name: "Patch 1 file", exact: true })).toHaveCount(0)
-    await expect(patch.getByRole("button")).toHaveCount(1)
-    await expect(patch.locator('[data-scope="apply-patch"] button[aria-expanded="false"]')).toHaveCount(1)
-    await expect(patch.locator('[data-slot="message-part-title-filename"]')).toHaveCount(0)
-    await expect(patch.locator('[data-slot="message-part-actions"]')).toHaveCount(0)
-    const edit = page.locator('[data-timeline-part-id="prt_edit"]')
-    await expect(edit).toContainText("Edit")
-    await expect(page.locator('[data-timeline-part-id="prt_todo"]')).toHaveCount(0)
-  })
-
   test("combines adjacent patch calls and repeated files into one group", async ({ page }) => {
     const first = "prt_patch_first"
     const second = "prt_patch_second"
@@ -158,43 +81,6 @@ test.describe("session timeline projection", () => {
     await expect(page.locator(`[data-timeline-part-id="${first}"], [data-timeline-part-id="${second}"]`)).toHaveCount(0)
   })
 
-  test("combines adjacent edit calls and repeated files into one group", async ({ page }) => {
-    const first = "prt_edit_first"
-    const second = "prt_edit_second"
-    await setupTimeline(page, {
-      messages: [
-        userMessage(),
-        assistantMessage([
-          toolPart(
-            first,
-            "edit",
-            "completed",
-            { path: "src/first.ts", oldString: "one", newString: "two" },
-            {
-              metadata: { files: [patchFile("src/first.ts", "modified")] },
-            },
-          ),
-          toolPart(
-            second,
-            "edit",
-            "completed",
-            { path: "src/first.ts", oldString: "two", newString: "three" },
-            {
-              metadata: { files: [patchFile("src/first.ts", "modified")] },
-            },
-          ),
-        ]),
-      ],
-      settings: { editToolPartsExpanded: true },
-    })
-
-    const group = page.locator(`[data-timeline-part-ids="${first},${second}"]`)
-    await expect(group.locator('[data-slot="basic-tool-tool-title"]')).toContainText("Edit")
-    await expect(group.getByText("1 file", { exact: true })).toBeVisible()
-    await expect(group.locator('[data-slot="apply-patch-filename"]')).toHaveText(["first.ts"])
-    await expect(group.locator('[data-scope="apply-patch"] button')).toHaveAttribute("aria-expanded", "true")
-  })
-
   test("projects gaps, dividers, assistant parts, and errors together", async ({ page }) => {
     const firstUser = userMessage(
       [
@@ -236,25 +122,6 @@ test.describe("session timeline projection", () => {
     await expect(page.locator('[data-timeline-row="TurnGap"]')).toBeVisible()
   })
 
-  test("renders interruption independently when the turn is not compacted", async ({ page }) => {
-    const user = userMessage()
-    const before = assistantMessage([{ id: "prt_before", type: "text", text: "Before" }], {
-      id: "msg_1001_before",
-      error: { type: "MessageAbortedError", message: "Stopped" },
-    })
-    const after = assistantMessage([{ id: "prt_after", type: "text", text: "After" }], {
-      id: "msg_1002_after",
-      created: 1700000003000,
-    })
-    await setupTimeline(page, { messages: [user, before, after] })
-
-    await expect(page.getByText("Interrupted", { exact: true })).toBeVisible()
-    const rows = await page
-      .locator('[data-timeline-row="AssistantPart"], [data-timeline-row="TurnDivider"]')
-      .evaluateAll((elements) => elements.map((element) => element.getAttribute("data-timeline-row")))
-    expect(rows).toEqual(["AssistantPart", "TurnDivider", "AssistantPart"])
-  })
-
   test("renders aliased and long custom model notices", async ({ page }) => {
     const shortName = "GPT-5.4 nano"
     const longName = "Company Gateway Extra Long Context Model for Narrow Timeline Layouts"
@@ -291,77 +158,8 @@ test.describe("session timeline projection", () => {
     await expect(longNotice.locator("[title]")).toHaveAttribute("title", `Switched to ${longName}`)
     await expect.poll(() => longNotice.evaluate((element) => element.scrollWidth <= element.clientWidth)).toBe(true)
   })
-
-  test("renders user image, file attachment, file reference, and agent reference", async ({ page }) => {
-    const text = "Use @explore with @src/a.ts and inspect the attachments"
-    const parts: PartSeed<"user">[] = [
-      userText(text, { id: "prt_user_rich" }),
-      {
-        id: "prt_user_image",
-        type: "file",
-        mime: "image/png",
-        filename: "pixel.png",
-        url: "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==",
-      },
-      {
-        id: "prt_user_attachment",
-        type: "file",
-        mime: "application/json",
-        filename: "tsconfig.json",
-        url: "data:application/json;base64,e30=",
-      },
-      {
-        id: "prt_user_reference",
-        type: "file",
-        mime: "text/plain",
-        filename: "a.ts",
-        url: "src/a.ts",
-        source: { type: "file", path: "src/a.ts", text: { value: "@src/a.ts", start: 18, end: 27 } },
-      },
-      {
-        id: "prt_user_agent",
-        type: "agent",
-        name: "explore",
-        source: { value: "@explore", start: 4, end: 12 },
-      },
-    ]
-    await setupTimeline(page, { messages: [userMessage(parts), assistantMessage()] })
-
-    await expect(page.getByAltText("pixel.png")).toBeVisible()
-    await expect(page.getByText("tsconfig.json")).toBeVisible()
-    await expect(page.getByText("@src/a.ts", { exact: true })).toBeVisible()
-    await expect(page.getByText("@explore", { exact: true })).toBeVisible()
-  })
 })
 
-function editPart(id: string) {
-  return toolPart(
-    id,
-    "edit",
-    "completed",
-    { path: "src/a.ts", oldString: "export const value = 1", newString: "export const value = 2" },
-    {
-      metadata: {
-        files: [patchFile("src/a.ts", "modified")],
-      },
-    },
-  )
-}
-
-function patchPart(id: string) {
-  return toolPart(
-    id,
-    "patch",
-    "completed",
-    { patchText: "Update the projected files" },
-    {
-      metadata: {
-        files: [patchFile("src/a.ts", "modified")],
-      },
-    },
-  )
-}
-
 function patchFile(file: string, status: "added" | "modified" | "deleted") {
   return {
     file,

packages/app/e2e/regression/session-timeline-reducer-projection.spec.ts

@@ -7,33 +7,25 @@ import {
   renderedPartID,
   setupTimeline,
   shell,
+  toolPart,
   status,
   textPart,
-  toolPart,
   userMessage,
 } from "../performance/timeline-stability/fixture"
 
-test("groups every collapsed tool until visible text separates the stack", async ({ page }) => {
-  const parts = [
-    toolPart("prt_boundary_01_read", "read", "completed", { path: "src/a.ts" }),
-    textPart("prt_boundary_02_text", "Boundary text"),
-    toolPart("prt_boundary_03_glob", "glob", "completed", { path: ".", pattern: "**/*.ts" }),
-    toolPart("prt_boundary_04_grep", "grep", "completed", { path: ".", pattern: "stable" }),
-    shell("prt_boundary_05_shell", "completed", "done"),
-    toolPart("prt_boundary_06_list", "list", "completed", { path: "src" }),
-  ]
-  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
+test("reducer-hardening: converges when idle arrives before final part and message completion", async ({ page }) => {
+  const textID = "prt_event_order_text"
+  const assistant = assistantMessage([textPart(textID, "Partial")], { completed: false })
+  const timeline = await setupTimeline(page, { messages: [userMessage(), assistant] })
+  await timeline.send(status("busy"), 100)
+  await timeline.send(status("idle"), 100)
+  await timeline.send(partUpdated(textPart(textID, "Final after early idle")), 120)
+  await timeline.send(messageUpdated(completedAssistantInfo(assistant)), 250)
 
-  await expect(page.locator('[data-timeline-part-ids="prt_boundary_01_read"]')).toBeVisible()
-  const group = page.locator(
-    '[data-timeline-part-ids="prt_boundary_03_glob,prt_boundary_04_grep,prt_boundary_05_shell,prt_boundary_06_list"]',
+  await expect(page.locator('[data-timeline-row="Thinking"]')).toHaveCount(0)
+  await expect(page.locator(`[data-timeline-part-id="${renderedPartID(textID)}"]`)).toContainText(
+    "Final after early idle",
   )
-  await expect(group).toBeVisible()
-  await expect(group.getByRole("button")).toHaveAccessibleName("Used Glob, Grep, Shell, List")
-  await expect(group.locator('[data-component="tag"]')).toHaveText("4")
-  await expect(page.locator('[data-timeline-row="AssistantPart"]')).toHaveCount(3)
-  await expect(page.locator('[data-timeline-spacing="content"]')).toHaveCount(2)
-  await expect(page.locator('[data-timeline-spacing="content"]').nth(0)).toHaveCSS("padding-top", "16px")
 })
 
 test("expands a mixed collapsed tool stack without expanding its individual calls", async ({ page }) => {
@@ -132,18 +124,3 @@ test("keeps failed search calls and their error cards inside the collapsed stack
     "Search timed out after 30 seconds",
   )
 })
-
-test("reducer-hardening: converges when idle arrives before final part and message completion", async ({ page }) => {
-  const textID = "prt_event_order_text"
-  const assistant = assistantMessage([textPart(textID, "Partial")], { completed: false })
-  const timeline = await setupTimeline(page, { messages: [userMessage(), assistant] })
-  await timeline.send(status("busy"), 100)
-  await timeline.send(status("idle"), 100)
-  await timeline.send(partUpdated(textPart(textID, "Final after early idle")), 120)
-  await timeline.send(messageUpdated(completedAssistantInfo(assistant)), 250)
-
-  await expect(page.locator('[data-timeline-row="Thinking"]')).toHaveCount(0)
-  await expect(page.locator(`[data-timeline-part-id="${renderedPartID(textID)}"]`)).toContainText(
-    "Final after early idle",
-  )
-})

packages/app/e2e/regression/session-timeline-tool-projection.spec.ts

@@ -7,31 +7,6 @@ import {
   userMessage,
 } from "../performance/timeline-stability/fixture"
 
-test("renders every tool error outcome without leaking hidden tools", async ({ page }) => {
-  const ordinary = ["shell", "edit", "write", "patch", "webfetch", "websearch", "subagent", "skill", "mcp_probe"]
-  const parts = ordinary.map((tool, index) =>
-    toolPart(`prt_error_${index}`, tool, "error", errorInput(tool), { error: `${tool} failed visibly` }),
-  )
-  parts.push(
-    toolPart("prt_question_dismissed", "question", "error", questionInput(), {
-      error: "The user dismissed this question",
-    }),
-    toolPart("prt_question_error", "question", "error", questionInput(), { error: "Question transport failed" }),
-    toolPart("prt_todo_error", "todowrite", "error", { todos: [] }, { error: "Hidden todo failure" }),
-  )
-  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
-
-  const group = page.locator(`[data-timeline-part-ids="${ordinary.map((_, index) => `prt_error_${index}`).join(",")}"]`)
-  await expect(group.locator('[data-component="tag"]')).toHaveText(String(ordinary.length))
-  await group.getByRole("button").click()
-  await expect(page.locator('[data-kind="tool-error-card"]')).toHaveCount(ordinary.length + 1)
-  await expect(page.getByText(/dismissed/i)).toBeVisible()
-  await expect(page.locator('[data-timeline-part-id="prt_todo_error"]')).toHaveCount(0)
-  for (let index = 0; index < ordinary.length; index++) {
-    await expect(page.locator(`[data-timeline-part-id="prt_error_${index}"]`)).toBeVisible()
-  }
-})
-
 test("transitions shell and question through running error outcomes", async ({ page }) => {
   const shellID = "prt_transition_error_shell"
   const questionID = "prt_transition_error_question"
@@ -138,62 +113,6 @@ test("preserves surviving grouped patch state when its first patch fails", async
     .toBeGreaterThanOrEqual(-0.5)
 })
 
-test("labels all web search provider variants", async ({ page }) => {
-  const parts = [
-    toolPart(
-      "prt_search_parallel",
-      "websearch",
-      "completed",
-      { query: "parallel" },
-      { metadata: { provider: "parallel" } },
-    ),
-    toolPart("prt_search_exa", "websearch", "completed", { query: "exa" }, { metadata: { provider: "exa" } }),
-    toolPart("prt_search_generic", "websearch", "completed", { query: "generic" }),
-  ]
-  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
-  await page.getByRole("button", { name: "Used Parallel Web Search, Exa Web Search, Web Search" }).click()
-
-  const tools = page.locator('[data-component="context-tool-group-list"]')
-  await expect(tools.getByRole("button", { name: /Parallel Web Search/ })).toBeVisible()
-  await expect(tools.getByRole("button", { name: /Exa Web Search/ })).toBeVisible()
-  await expect(tools.getByRole("button", { name: /^Web Search/ })).toBeVisible()
-})
-
-test("labels completed searches with result counts", async ({ page }) => {
-  const glob = "prt_glob_count"
-  const grep = "prt_grep_count"
-  await setupTimeline(page, {
-    messages: [
-      userMessage(),
-      assistantMessage([
-        toolPart(glob, "glob", "completed", { path: ".", pattern: "**/*.ts" }, { metadata: { count: 1 } }),
-        toolPart(grep, "grep", "completed", { path: ".", pattern: "value" }, { metadata: { matches: 12 } }),
-      ]),
-    ],
-  })
-
-  const group = page.locator(`[data-timeline-part-ids="${glob},${grep}"]`)
-  await group.locator('[data-slot="collapsible-trigger"]').click()
-  const rows = group.locator('[data-component="context-tool-group-list"] [data-component="tool-trigger"]')
-  await expect(rows.filter({ hasText: "Glob" })).toContainText("(1 match)")
-  await expect(rows.filter({ hasText: "Grep" })).toContainText("(12 matches)")
-})
-
-test("labels read tools from their path input", async ({ page }) => {
-  const id = "prt_read_path"
-  await setupTimeline(page, {
-    messages: [userMessage(), assistantMessage([toolPart(id, "read", "completed", { path: "src/a.ts" })])],
-  })
-
-  const group = page.locator(`[data-timeline-part-ids="${id}"]`)
-  await group.locator('[data-slot="collapsible-trigger"]').click()
-  await expect(
-    group
-      .locator('[data-component="context-tool-group-list"] [data-component="tool-trigger"]')
-      .filter({ hasText: "Read" }),
-  ).toContainText("a.ts")
-})
-
 test("groups instruction files loaded by the same read", async ({ page }) => {
   const id = "prt_read_instructions"
   await setupTimeline(page, {
@@ -221,36 +140,6 @@ test("groups instruction files loaded by the same read", async ({ page }) => {
   await expect(loaded.locator('[data-slot="tool-loaded-kind"]')).toHaveCount(0)
 })
 
-test("labels skill tools from IDs and result metadata", async ({ page }) => {
-  const pending = "prt_skill_id"
-  const completed = "prt_skill_name"
-  await setupTimeline(page, {
-    messages: [
-      userMessage(),
-      assistantMessage([
-        toolPart(pending, "skill", "running", { id: "frontend-design" }),
-        toolPart(completed, "skill", "completed", { id: "opencode" }, { metadata: { name: "OpenCode" } }),
-      ]),
-    ],
-  })
-
-  const group = page.locator(`[data-timeline-part-ids="${pending},${completed}"]`)
-  await expect(group.getByRole("button")).toHaveAccessibleName("Used Skill")
-  await expect(group.locator('[data-component="tag"]')).toHaveText("2")
-  await group.getByRole("button").click()
-
-  const loaded = group.locator('[data-component="tool-loaded-item"]')
-  await expect(loaded).toHaveCount(1)
-  await expect(loaded).toHaveAttribute("aria-label", "Loaded frontend-design, OpenCode skills")
-  await expect(loaded).toHaveCSS("line-height", "16px")
-  await expect(loaded.locator('[data-slot="tool-loaded-label"]')).toHaveText("Loaded")
-  await expect(loaded.locator('[data-slot="tool-loaded-kind"]')).toHaveText("skills")
-  const names = loaded.locator('[data-component="text-shimmer"]')
-  await expect(names).toHaveCount(2)
-  await expect(names.nth(0)).toHaveAttribute("aria-label", "frontend-design")
-  await expect(names.nth(1)).toHaveAttribute("aria-label", "OpenCode")
-})
-
 test("groups only consecutive successful skill tools", async ({ page }) => {
   const parts = [
     toolPart("prt_skill_first", "skill", "completed", { id: "ocpr" }),
@@ -273,14 +162,3 @@ test("groups only consecutive successful skill tools", async ({ page }) => {
 function questionInput() {
   return { questions: [{ header: "Stability", question: "Keep it stable?", options: [] }] }
 }
-
-function errorInput(tool: string) {
-  if (tool === "shell") return { command: "exit 1" }
-  if (["edit", "write"].includes(tool)) return { path: "src/error.ts", content: "" }
-  if (tool === "patch") return { patchText: "Update src/error.ts" }
-  if (tool === "webfetch") return { url: "https://example.com" }
-  if (tool === "websearch") return { query: "failure" }
-  if (tool === "subagent") return { description: "Fail subagent", agent: "explore", prompt: "Inspect the failure." }
-  if (tool === "skill") return { name: "failure" }
-  return { target: "failure" }
-}

packages/app/e2e/regression/terminal-composer-focus.spec.ts

@@ -87,6 +87,7 @@ test("clears the terminal line with Command+Delete", async ({ page }) => {
   const terminal = page.locator('[data-component="terminal"]')
   await page.keyboard.press("Control+Backquote")
   await expect(terminal.locator("textarea")).toHaveCount(1)
+  await expect.poll(() => sendPtyOutput).toBeDefined()
 
   await page.keyboard.press("Meta+Backspace")
 

packages/app/e2e/regression/terminal-hidden.spec.ts

@@ -127,7 +127,6 @@ test("animates review and terminal panels while caching hidden terminal content"
   await expectStackedGeometry(page)
   await expectPanelGapHeld(page)
 
-  await resetTerminalTopMotion(page)
   await resetTerminalBottomMotion(page)
   await resetTerminalAnchorGaps(page)
   await resetPanelGaps(page)
@@ -139,7 +138,6 @@ test("animates review and terminal panels while caching hidden terminal content"
   await expect(panel).toBeVisible()
   await expectHeightMotions(page, "session-side-region", 2)
   await expectHeightMotions(page, "session-side-terminal-region", 2)
-  await expectTerminalTopMotion(page)
   await expectTerminalBottomFixed(page)
   await expectTerminalTopAnchored(page)
   await expectPanelGapHeld(page)
@@ -225,7 +223,6 @@ type MotionProbe = {
   terminalAnchorGaps: number[]
   resetAnchorOnMotion: boolean
   panelGaps: number[]
-  terminalTops: number[]
   terminalBottoms: number[]
   heights: string[]
   animations: string[]
@@ -243,7 +240,6 @@ async function installMotionProbe(page: Page) {
       terminalAnchorGaps: [],
       resetAnchorOnMotion: false,
       panelGaps: [],
-      terminalTops: [],
       terminalBottoms: [],
       heights: [],
       animations: [],
@@ -270,7 +266,6 @@ async function installMotionProbe(page: Page) {
         const terminalContent = document.querySelector<HTMLElement>('[data-slot="terminal-panel-content"]')
         const panelGap = document.querySelector<HTMLElement>('[data-slot="session-side-panel-gap"]')
         if (!terminal || !terminalContent) return
-        probe.terminalTops.push(terminal.getBoundingClientRect().top)
         probe.terminalBottoms.push(terminal.getBoundingClientRect().bottom)
         probe.terminalContentSizes.push({
           width: terminalContent.getBoundingClientRect().width,
@@ -446,13 +441,6 @@ async function expectStackPainted(page: Page) {
   expect(Math.max(...gaps.map((gap) => gap.terminalSurface)), JSON.stringify(gaps)).toBeLessThanOrEqual(1)
 }
 
-async function resetTerminalTopMotion(page: Page) {
-  await page.evaluate(() => {
-    const probe = (window as Window & { __panelMotion?: MotionProbe }).__panelMotion
-    if (probe) probe.terminalTops = []
-  })
-}
-
 async function resetTerminalBottomMotion(page: Page) {
   await page.evaluate(() => {
     const probe = (window as Window & { __panelMotion?: MotionProbe }).__panelMotion
@@ -516,17 +504,6 @@ async function expectTerminalContentCachedSize(page: Page) {
   expect(Math.min(...sizes.map((size) => size.height))).toBeGreaterThan(100)
 }
 
-async function expectTerminalTopMotion(page: Page) {
-  const tops = await page.evaluate(
-    () => (window as Window & { __panelMotion?: MotionProbe }).__panelMotion?.terminalTops.map(Math.round) ?? [],
-  )
-  const unique = [...new Set(tops)]
-  const range = Math.max(...unique) - Math.min(...unique)
-  const maxDelta = Math.max(...unique.slice(1).map((value, index) => Math.abs(value - unique[index])))
-  expect(unique.length, JSON.stringify(unique)).toBeGreaterThan(6)
-  expect(maxDelta, JSON.stringify({ unique, range, maxDelta })).toBeLessThan(range * 0.3)
-}
-
 async function expectHeightMotions(page: Page, slot: string, count: number) {
   await expect
     .poll(() =>

packages/app/e2e/service-worker/cache.spec.ts

@@ -0,0 +1,137 @@
+import { expect, test } from "@playwright/test"
+import { readFile } from "node:fs/promises"
+import { createServer } from "node:http"
+import { once } from "node:events"
+
+const legacy = `
+self.addEventListener("install", event => event.waitUntil(
+  caches.open("workbox-precache-v2-" + self.registration.scope).then(cache =>
+    cache.addAll(["/index.html", "/assets/app-old.js", "/assets/lazy-old.js"])
+  )
+))
+self.addEventListener("fetch", event => {
+  if (event.request.mode === "navigate") {
+    event.respondWith(caches.match("/index.html"))
+    return
+  }
+  event.respondWith(caches.match(event.request).then(response => response || fetch(event.request)))
+})
+`
+
+const fixture = test.extend<{ site: { url: string; upgrade: () => void; repair: () => void } }>({
+  site: async ({}, use) => {
+    const worker = await readFile(new URL("../../dist/sw.js", import.meta.url), "utf8")
+    const state = { version: "old", repaired: false }
+    const server = createServer((request, response) => {
+      const pathname = new URL(request.url ?? "/", "http://localhost").pathname
+      const prefix = state.version === "old" ? "/assets" : "/_assets"
+      response.setHeader("cache-control", "no-store")
+      if (pathname === "/sw.js") {
+        response.setHeader("content-type", "text/javascript")
+        response.end(state.version === "old" ? legacy : worker)
+        return
+      }
+      if (pathname === `${prefix}/app-${state.version}.js`) {
+        response.setHeader("content-type", "text/javascript")
+        response.end(`import "${prefix}/startup-${state.version}.js"`)
+        return
+      }
+      if (pathname === `${prefix}/startup-${state.version}.js`) {
+        response.setHeader("content-type", "text/javascript")
+        response.end(`
+          document.getElementById("root").innerHTML = '<h1>${state.version}</h1><label>Draft<input></label><button>Load older chunk</button><output></output>'
+          document.querySelector("button").onclick = () => import("/assets/lazy-old.js")
+        `)
+        return
+      }
+      if (
+        (pathname === "/assets/lazy-old.js" && state.version === "old") ||
+        (pathname === "/_assets/retry.js" && state.repaired)
+      ) {
+        response.setHeader("content-type", "text/javascript")
+        response.end('document.querySelector("output").textContent = "Older chunk loaded"')
+        return
+      }
+      // Deliberately retain the old server's fallback so the worker must reject HTML asset responses itself.
+      response.setHeader("content-type", "text/html")
+      response.end(`<div id="root"></div><script type="module" src="${prefix}/app-${state.version}.js"></script>`)
+    })
+    server.listen(0, "127.0.0.1")
+    await once(server, "listening")
+    const address = server.address()
+    if (!address || typeof address === "string") throw new Error("Expected a TCP address")
+    try {
+      await use({
+        url: `http://127.0.0.1:${address.port}`,
+        upgrade: () => (state.version = "new"),
+        repair: () => (state.repaired = true),
+      })
+    } finally {
+      server.closeAllConnections()
+      await new Promise<void>((resolve, reject) => server.close((error) => (error ? reject(error) : resolve())))
+    }
+  },
+})
+
+fixture("updates a legacy worker without reloading drafts or deleting old chunks", async ({ page, site }) => {
+  await page.goto(site.url)
+  await expect(page.getByRole("heading")).toHaveText("old")
+  await page.evaluate(async () => {
+    await navigator.serviceWorker.register("/sw.js")
+    await navigator.serviceWorker.ready
+  })
+  await page.goto(site.url)
+  await expect(page.getByRole("heading")).toHaveText("old")
+  await page.getByLabel("Draft").fill("Keep this unsent prompt")
+
+  site.upgrade()
+  await page.evaluate(async () => {
+    const cache = await caches.open("opencode-assets")
+    await cache.put(
+      "/_assets/startup-new.js",
+      new Response("<html>stale fallback</html>", {
+        headers: { "content-type": "text/html" },
+      }),
+    )
+    const changed = new Promise<void>((resolve) =>
+      navigator.serviceWorker.addEventListener("controllerchange", () => resolve(), { once: true }),
+    )
+    const registration = await navigator.serviceWorker.getRegistration()
+    if (!registration) throw new Error("Missing legacy worker")
+    await registration.update()
+    await changed
+  })
+
+  await expect(page.getByLabel("Draft")).toHaveValue("Keep this unsent prompt")
+  await page.getByRole("button", { name: "Load older chunk" }).click()
+  await expect(page.getByRole("status")).toHaveText("Older chunk loaded")
+
+  await page.goto(`${site.url}/workspace/example`)
+  await expect(page.getByRole("heading")).toHaveText("new")
+  await expect
+    .poll(() =>
+      page.evaluate(async () =>
+        (await (await caches.open("opencode-assets")).match("/_assets/startup-new.js"))?.headers.get("content-type"),
+      ),
+    )
+    .toBe("text/javascript")
+})
+
+fixture("does not cache HTML responses under asset URLs", async ({ page, site }) => {
+  site.upgrade()
+  await page.goto(site.url)
+  await expect(page.getByRole("heading")).toHaveText("new")
+  await page.evaluate(async () => {
+    await navigator.serviceWorker.register("/sw.js")
+    await navigator.serviceWorker.ready
+  })
+  await page.goto(site.url)
+  await expect(page.getByRole("heading")).toHaveText("new")
+  expect(await page.evaluate(async () => (await fetch("/_assets/retry.js")).headers.get("content-type"))).toBe(
+    "text/html",
+  )
+  site.repair()
+  expect(await page.evaluate(async () => (await fetch("/_assets/retry.js")).headers.get("content-type"))).toBe(
+    "text/javascript",
+  )
+})

packages/app/e2e/service-worker/playwright.config.ts

@@ -0,0 +1,9 @@
+import { defineConfig } from "@playwright/test"
+
+export default defineConfig({
+  testDir: ".",
+  testMatch: "*.spec.ts",
+  outputDir: "../test-results/service-worker",
+  timeout: 30_000,
+  use: { browserName: "chromium" },
+})

packages/app/e2e/tsconfig.json

@@ -7,5 +7,5 @@
     "rootDir": "..",
     "types": ["node", "bun"]
   },
-  "include": ["./**/*.ts", "./**/*.tsx", "../src/types.ts"]
+  "include": ["./**/*.ts", "./**/*.tsx", "../component-tests/**/*.ts", "../src/types.ts"]
 }

packages/app/package.json

@@ -26,8 +26,11 @@
     "test:unit:watch": "bun test --conditions=solid --watch --preload ./happydom.ts ./src",
     "test:e2e": "playwright test",
     "test:e2e:local": "playwright test",
+    "test:components": "playwright test --config playwright.components.config.ts",
+    "test:components:ui": "playwright test --config playwright.components.config.ts --ui",
     "test:e2e:ui": "playwright test --ui",
     "test:e2e:report": "playwright show-report e2e/playwright-report",
+    "test:service-worker": "bun run build && playwright test --config e2e/service-worker/playwright.config.ts",
     "test:stability": "bun test ./e2e/performance/unit/visual-stability.test.ts && playwright test --config e2e/performance/timeline-stability/playwright.config.ts",
     "test:bench": "bun test ./e2e/performance/unit && playwright test --config e2e/performance/playwright.config.ts",
     "test:bench:devex": "bun test ./e2e/performance/unit/desktop-startup.test.ts && playwright test --config e2e/performance/devex/playwright.config.ts"

packages/app/playwright.components.config.ts

@@ -0,0 +1,4 @@
+import { fileURLToPath } from "node:url"
+import { componentConfig } from "../storybook/playwright/config"
+
+export default componentConfig(fileURLToPath(new URL(".", import.meta.url)))

packages/app/playwright.config.ts

@@ -9,7 +9,10 @@ const reuse = !process.env.CI
 const workers = Number(process.env.PLAYWRIGHT_WORKERS ?? (process.env.CI ? 5 : 0)) || undefined
 export default defineConfig({
   testDir: "./e2e",
-  testIgnore: process.env.OPENCODE_PERFORMANCE === "1" ? "performance/**/*.test.ts" : "performance/**",
+  testIgnore: [
+    "service-worker/**",
+    process.env.OPENCODE_PERFORMANCE === "1" ? "performance/**/*.test.ts" : "performance/**",
+  ],
   outputDir: "./e2e/test-results",
   timeout: 60_000,
   expect: {

packages/app/public/_headers

@@ -1,10 +1,10 @@
-/assets/*.js
+/_assets/*.js
   Content-Type: application/javascript
 
-/assets/*.mjs
+/_assets/*.mjs
   Content-Type: application/javascript
 
-/assets/*.css
+/_assets/*.css
   Content-Type: text/css
 
 /*.js

packages/app/tsconfig.json

@@ -21,6 +21,12 @@
       "@/*": ["./src/*"]
     }
   },
-  "include": ["src", "package.json"],
+  "include": [
+    "src",
+    "component-tests",
+    "playwright.components.config.ts",
+    "../storybook/playwright/*.ts",
+    "package.json"
+  ],
   "exclude": ["dist", "ts-dist"]
 }

packages/app/vite.config.ts

@@ -29,33 +29,44 @@ export default defineConfig({
       injectRegister: false,
       manifest: false,
       workbox: {
-        cleanupOutdatedCaches: true,
         clientsClaim: false,
-        skipWaiting: false,
+        skipWaiting: true,
         inlineWorkboxRuntime: true,
-        navigateFallback: "/index.html",
-        navigateFallbackDenylist: [/^\/api(?:\/|$)/],
-        globPatterns: [
-          "index.html",
-          "site.webmanifest",
-          "favicon*",
-          "apple-touch-icon*",
-          "web-app-manifest*",
-          "assets/index-*.{js,css}",
-          "assets/session-*.js",
-          "assets/IBMPlexMono-Text-*.woff2",
-          "assets/Inter.ttf",
-          "assets/JetBrainsMonoNerdFontMono-Regular.woff2",
-        ],
+        // Always fetch the current HTML. Precaching a partial build can strand it without its chunks after an upgrade.
+        navigateFallback: null,
+        globPatterns: [],
         runtimeCaching: [
           {
-            urlPattern: ({ url }) => url.origin === self.location.origin && url.pathname.startsWith("/assets/"),
+            urlPattern: ({ url }) =>
+              url.origin === self.location.origin &&
+              (url.pathname.startsWith("/_assets/") || url.pathname.startsWith("/assets/")),
             handler: "CacheFirst",
             options: {
               cacheName: "opencode-assets",
-              cacheableResponse: {
-                statuses: [200],
-              },
+              plugins: [
+                {
+                  cachedResponseWillBeUsed: async ({ request, cachedResponse }) => {
+                    if (
+                      cachedResponse?.status === 200 &&
+                      !/^(text\/html|application\/xhtml\+xml)\b/i.test(cachedResponse.headers.get("content-type") ?? "")
+                    )
+                      return cachedResponse
+                    // Keep old tabs' precached chunks usable without retaining their stale HTML navigation handler.
+                    const response = await caches.match(request, {
+                      cacheName: `workbox-precache-v2-${self.location.origin}/`,
+                    })
+                    return response?.status === 200 &&
+                      !/^(text\/html|application\/xhtml\+xml)\b/i.test(response.headers.get("content-type") ?? "")
+                      ? response
+                      : null
+                  },
+                  cacheWillUpdate: async ({ response }) =>
+                    response.status === 200 &&
+                    !/^(text\/html|application\/xhtml\+xml)\b/i.test(response.headers.get("content-type") ?? "")
+                      ? response
+                      : null,
+                },
+              ],
               expiration: {
                 maxEntries: 1000,
               },
@@ -72,6 +83,7 @@ export default defineConfig({
     port: 3000,
   },
   build: {
+    assetsDir: "_assets",
     target: "esnext",
     sourcemap: true,
   },

packages/cli/package.json

@@ -27,7 +27,7 @@
     "@effect/platform-node": "catalog:",
     "@opencode-ai/client": "workspace:*",
     "@opencode-ai/plugin": "workspace:*",
-    "@opencode-ai/pty": "0.1.9",
+    "@opencode-ai/pty": "0.1.11",
     "@opencode-ai/schema": "workspace:*",
     "@opencode-ai/server": "workspace:*",
     "@opencode-ai/tui": "workspace:*",

packages/cli/src/commands/handlers/service/restart.ts

@@ -4,11 +4,14 @@ import { Service } from "@opencode-ai/client/effect/service"
 import { Commands } from "../../commands"
 import { Runtime } from "../../../framework/runtime"
 import { ServiceConfig } from "../../../services/service-config"
+import { ServerConnection } from "../../../services/server-connection"
 
 export default Runtime.handler(
   Commands.commands.service.commands.restart,
   Effect.fn("cli.service.restart")(function* () {
     const options = yield* ServiceConfig.options()
+    // Keep this explicit: automatic service replacement must preserve terminals.
+    yield* ServerConnection.shutdownPersistentPty(options).pipe(Effect.ignore)
     yield* Service.stop(options)
     const transport = yield* Service.ensure(options)
     process.stdout.write(transport.url + EOL)

packages/cli/src/services/web-ui.ts

@@ -26,6 +26,8 @@ export const handler = Effect.fn("cli.web-ui.handler")(function* (options?: { re
 
 function serveUI(request: HttpServerRequest.HttpServerRequest, url: URL, assets: AssetMap) {
   const key = url.pathname.replace(/^\//, "")
+  if (key.startsWith("_assets/") && assets[key] === undefined)
+    return Effect.succeed(HttpServerResponse.empty({ status: 404, headers: { "cache-control": "no-store" } }))
   const name = assets[key] !== undefined ? key : "index.html"
   const file = assets[name]
   if (!file) return Effect.succeed(HttpServerResponse.empty({ status: 404 }))

packages/cli/test/web-ui.test.ts

@@ -19,7 +19,7 @@ describe("web UI", () => {
     await writeFile(asset, "console.log('embedded')")
     const assets = {
       "index.html": await Bun.file(index).text(),
-      "app.js": await Bun.file(asset).text(),
+      "_assets/app.js": await Bun.file(asset).text(),
       "sw.js": "service worker",
       "registerSW.js": "registration",
       "font.woff2": new Uint8Array([0, 1, 2, 255]),
@@ -53,7 +53,16 @@ describe("web UI", () => {
           expect(missing.status).toBe(404)
           expect(yield* Effect.promise(() => missing.text())).toBe("")
 
-          const script = yield* Effect.promise(() => fetch(`${origin}/app.js`))
+          yield* Effect.forEach(["/_assets/old.js", "/_assets/old.css", "/_assets/missing"], (pathname) =>
+            Effect.gen(function* () {
+              const missing = yield* Effect.promise(() => fetch(`${origin}${pathname}`))
+              expect(missing.status).toBe(404)
+              expect(missing.headers.get("cache-control")).toBe("no-store")
+              expect(yield* Effect.promise(() => missing.text())).toBe("")
+            }),
+          )
+
+          const script = yield* Effect.promise(() => fetch(`${origin}/_assets/app.js`))
           expect(yield* Effect.promise(() => script.text())).toBe("console.log('embedded')")
           expect(script.headers.get("content-type")).toContain("javascript")
           expect(script.headers.get("cache-control")).toBe("public, max-age=31536000, immutable")
@@ -74,6 +83,14 @@ describe("web UI", () => {
           expect(yield* Effect.promise(() => fallback.text())).toContain("embedded")
           expect(fallback.headers.get("content-security-policy")).toContain("default-src 'self'")
           expect(fallback.headers.get("content-security-policy")).toContain("connect-src * data: blob:")
+
+          const dotted = yield* Effect.promise(() => fetch(`${origin}/workspace/example.js`))
+          expect(dotted.status).toBe(200)
+          expect(yield* Effect.promise(() => dotted.text())).toContain("embedded")
+
+          const legacy = yield* Effect.promise(() => fetch(`${origin}/assets/missing.js`))
+          expect(legacy.status).toBe(200)
+          expect(yield* Effect.promise(() => legacy.text())).toContain("embedded")
         }),
       ).pipe(Effect.provide(NodeFileSystem.layer)),
     )

packages/client/test/promise.test.ts

@@ -514,6 +514,115 @@ test("event.subscribe exposes the Promise event stream wire projection", async (
   expect(events[1]?.type === "session.model.selected" && events[1].created).toBe(1_717_171_717_000)
 })
 
+// Moved from packages/app/e2e/regression/session-timeline-transport.spec.ts
+test("event.subscribe keeps one request open while delivering multiple events", async () => {
+  const requests: Request[] = []
+  const events = [
+    { id: "evt_first", created: 1, type: "server.connected", data: {} },
+    { id: "evt_second", created: 2, type: "server.connected", data: {} },
+  ]
+  const client = OpenCode.make({
+    baseUrl: "http://localhost:3000",
+    fetch: async (input, init) => {
+      requests.push(input instanceof Request ? input : new Request(input, init))
+      return new Response(events.map((event) => `data: ${JSON.stringify(event)}\n\n`).join(""), {
+        headers: { "content-type": "text/event-stream" },
+      })
+    },
+  })
+  const received = []
+  for await (const event of client.event.subscribe()) received.push(event)
+  expect(received).toEqual(events)
+  expect(requests).toHaveLength(1)
+})
+
+// Moved from packages/app/e2e/regression/session-timeline-transport.spec.ts
+test("event.subscribe delivers every event from one stream chunk", async () => {
+  const events = Array.from({ length: 4 }, (_, index) => ({
+    id: `evt_burst_${index}`,
+    created: index,
+    type: "server.connected",
+    data: {},
+  }))
+  const client = OpenCode.make({
+    baseUrl: "http://localhost:3000",
+    fetch: async () =>
+      new Response(new TextEncoder().encode(events.map((event) => `data: ${JSON.stringify(event)}\n\n`).join("")), {
+        headers: { "content-type": "text/event-stream" },
+      }),
+  })
+  const received = []
+  for await (const event of client.event.subscribe()) received.push(event)
+  expect(received).toEqual(events)
+  expect(new Set(received.map((event) => event.id)).size).toBe(4)
+})
+
+// Moved from packages/app/e2e/regression/session-timeline-transport.spec.ts
+test("event.subscribe parses split JSON and a split multibyte code point", async () => {
+  const event = {
+    id: "evt_split",
+    created: 1,
+    type: "server.connected",
+    data: { text: "split snowman \u2603\u2603\u2603" },
+  }
+  const encoded = new TextEncoder().encode(`data: ${JSON.stringify(event)}\n\n`)
+  const multibyte = encoded.indexOf(new TextEncoder().encode("\u2603")[0]!)
+  const boundaries = [9, multibyte + 1, multibyte + 2, encoded.length]
+  const client = OpenCode.make({
+    baseUrl: "http://localhost:3000",
+    fetch: async () =>
+      new Response(
+        new ReadableStream({
+          start(controller) {
+            boundaries.forEach((end, index) =>
+              controller.enqueue(encoded.slice(index ? boundaries[index - 1] : 0, end)),
+            )
+            controller.close()
+          },
+        }),
+        { headers: { "content-type": "text/event-stream" } },
+      ),
+  })
+  await expect(client.event.subscribe()[Symbol.asyncIterator]().next()).resolves.toEqual({ done: false, value: event })
+})
+
+// Moved from packages/app/e2e/regression/session-timeline-transport.spec.ts
+test("event.subscribe ignores server heartbeat comments", async () => {
+  const event = { id: "evt_sentinel", created: 1, type: "server.connected", data: {} }
+  const client = OpenCode.make({
+    baseUrl: "http://localhost:3000",
+    fetch: async () =>
+      new Response(`: heartbeat\n\ndata: ${JSON.stringify(event)}\n\n: heartbeat\n\n`, {
+        headers: { "content-type": "text/event-stream" },
+      }),
+  })
+  const received = []
+  for await (const item of client.event.subscribe()) received.push(item)
+  expect(received).toEqual([event])
+})
+
+// Moved from packages/app/e2e/regression/session-timeline-transport.spec.ts
+test("event transport passes through ordinary health requests", async () => {
+  const requests: string[] = []
+  const event = { id: "evt_connected", created: 1, type: "server.connected", data: {} }
+  const client = OpenCode.make({
+    baseUrl: "http://localhost:3000",
+    fetch: async (input, init) => {
+      const request = input instanceof Request ? input : new Request(input, init)
+      requests.push(new URL(request.url).pathname)
+      if (new URL(request.url).pathname === "/api/event") {
+        return new Response(`data: ${JSON.stringify(event)}\n\n`, {
+          headers: { "content-type": "text/event-stream" },
+        })
+      }
+      return Response.json({ healthy: true, version: "2.0.0", pid: 1 })
+    },
+  })
+  await expect(client.event.subscribe()[Symbol.asyncIterator]().next()).resolves.toEqual({ done: false, value: event })
+  await expect(client.health.get()).resolves.toEqual({ healthy: true, version: "2.0.0", pid: 1 })
+  expect(requests).toEqual(["/api/event", "/api/health"])
+})
+
 test("event.subscribe terminates on malformed Promise SSE data", async () => {
   const client = OpenCode.make({
     baseUrl: "http://localhost:3000",

packages/core/package.json

@@ -118,7 +118,7 @@
     "@ff-labs/fff-node": "0.10.5",
     "@opencode-ai/codemode": "workspace:*",
     "@opencode-ai/ai": "workspace:*",
-    "@opencode-ai/pty": "0.1.9",
+    "@opencode-ai/pty": "0.1.11",
     "@opencode-ai/schema": "workspace:*",
     "@opencode-ai/plugin": "workspace:*",
     "@opencode-ai/util": "workspace:*",

packages/core/src/database/migration/20260805200742_import_legacy_credentials.ts

@@ -80,10 +80,11 @@ export function importLegacyCredentials(tx: Parameters<DatabaseMigration.Migrati
                       }
                     : undefined,
               })
+      const label = credential.type === "oauth" ? "OAuth" : "API key"
       const now = Date.now()
       yield* tx.run(sql`
         INSERT INTO credential (id, integration_id, label, value, time_created, time_updated)
-        VALUES (${Credential.ID.create()}, ${integrationID}, 'default', ${JSON.stringify(credential)}, ${now}, ${now})
+        VALUES (${Credential.ID.create()}, ${integrationID}, ${label}, ${JSON.stringify(credential)}, ${now}, ${now})
       `)
     }
 

packages/core/src/job.ts

@@ -1,11 +1,41 @@
 export * as Job from "./job.js"
 
-import { Cause, Clock, Context, Deferred, Effect, Exit, Layer, Scope, SynchronizedRef } from "effect"
+import { Array, Cause, Clock, Context, Deferred, Effect, Exit, Layer, Schema, Scope, SynchronizedRef } from "effect"
 import { makeGlobalNode } from "@opencode-ai/util/effect/app-node"
 import { Identifier } from "./id/id.js"
+import { KV } from "./kv.js"
+import { SessionMessage } from "./session/message.js"
 import { SessionSchema } from "./session/schema.js"
 
-export type Status = "running" | "completed" | "error" | "cancelled"
+const Background = Schema.Struct({
+  id: Schema.String,
+  notificationID: SessionMessage.ID,
+  recovery: Schema.Union([
+    Schema.Struct({
+      kind: Schema.Literal("shell"),
+      sessionID: SessionSchema.ID,
+      shellID: Schema.String,
+      command: Schema.String,
+    }),
+    Schema.Struct({
+      kind: Schema.Literal("subagent"),
+      parentSessionID: SessionSchema.ID,
+      childSessionID: SessionSchema.ID,
+      agent: Schema.String,
+      description: Schema.String,
+    }),
+  ]),
+  status: Schema.Literals(["running", "completed", "error", "cancelled"]),
+  output: Schema.optionalKey(Schema.String),
+  error: Schema.optionalKey(Schema.String),
+})
+
+export type Background = typeof Background.Type
+export type Recovery = Background["recovery"]
+export type Status = Background["status"]
+
+const decodeBackground = Schema.decodeUnknownResult(Background)
+const backgroundPrefix = "job.background/"
 
 export type Info = {
   id: string
@@ -17,6 +47,7 @@ export type Info = {
   output?: string
   error?: string
   metadata?: Record<string, unknown>
+  notificationID?: SessionMessage.ID
 }
 
 type Active = {
@@ -27,6 +58,7 @@ type Active = {
   token: object
   blockingSessions: Map<SessionSchema.ID, number>
   isBackgrounded: boolean
+  recovery?: Recovery
 }
 
 type State = {
@@ -63,6 +95,8 @@ export type StartInput = {
   type: string
   title?: string
   metadata?: Record<string, unknown>
+  recovery?: Recovery
+  notificationID?: SessionMessage.ID
   run: Effect.Effect<string, unknown>
 }
 
@@ -96,6 +130,8 @@ export interface Interface {
   readonly background: (id: string) => Effect.Effect<Info | undefined>
   readonly backgroundAll: (input: BackgroundAllInput) => Effect.Effect<Info[]>
   readonly cancel: (id: string) => Effect.Effect<Info | undefined>
+  readonly pendingBackground: Effect.Effect<readonly Background[]>
+  readonly completeBackground: (notificationID: SessionMessage.ID) => Effect.Effect<void>
 }
 
 export class Service extends Context.Service<Service, Interface>()("@opencode/Job") {}
@@ -126,43 +162,57 @@ function decrementSession(input: Map<SessionSchema.ID, number>, sessionID: Sessi
 }
 
 /**
- * Makes one scoped, process-local registry. Entries are intentionally not
- * durable: process restart or owner-scope closure loses status and interrupts
- * live work. Persisted observation, restart recovery, and remote workers need a
- * separate durable ownership slice rather than pretending this registry has
- * those semantics.
+ * Makes one scoped, process-local registry. Explicitly recoverable background
+ * work also owns a durable notification marker until its notification is admitted.
  */
 export const make = Effect.gen(function* () {
+  const kv = yield* KV.Service
   const state: State = {
     jobs: yield* SynchronizedRef.make(new Map()),
     scope: yield* Scope.Scope,
   }
 
+  const persistBackground = Effect.fnUntraced(function* (job: Active) {
+    if (!job.recovery || !job.info.notificationID) return
+    yield* kv.set(`${backgroundPrefix}${job.info.notificationID}`, {
+      id: job.info.id,
+      notificationID: job.info.notificationID,
+      recovery: job.recovery,
+      status: job.info.status,
+      ...(job.info.output !== undefined ? { output: job.info.output } : {}),
+      ...(job.info.error !== undefined ? { error: job.info.error } : {}),
+    })
+  })
+
   const settle = Effect.fnUntraced(function* (id: string, token: object, exit: Exit.Exit<string, unknown>) {
     const completed_at = yield* Clock.currentTimeMillis
-    const result = yield* SynchronizedRef.modify(state.jobs, (jobs): readonly [FinishResult, Map<string, Active>] => {
-      const job = jobs.get(id)
-      if (!job) return [{}, jobs]
-      if (job.token !== token) return [{}, jobs]
-      if (job.info.status !== "running") return [{ info: snapshot(job) }, jobs]
-      const status: Exclude<Status, "running"> = Exit.isSuccess(exit)
-        ? "completed"
-        : Cause.hasInterruptsOnly(exit.cause)
-          ? "cancelled"
-          : "error"
-      const next = {
-        ...job,
-        blockingSessions: new Map<SessionSchema.ID, number>(),
-        info: {
-          ...job.info,
-          status,
-          completed_at,
-          ...(Exit.isSuccess(exit) ? { output: exit.value } : {}),
-          ...(Exit.isFailure(exit) ? { error: errorText(Cause.squash(exit.cause)) } : {}),
-        },
-      }
-      return [{ info: snapshot(next), done: job.done, scope: job.scope }, new Map(jobs).set(id, next)]
-    })
+    const result = yield* SynchronizedRef.modifyEffect(
+      state.jobs,
+      Effect.fnUntraced(function* (jobs): Effect.fn.Return<readonly [FinishResult, Map<string, Active>]> {
+        const job = jobs.get(id)
+        if (!job) return [{}, jobs]
+        if (job.token !== token) return [{}, jobs]
+        if (job.info.status !== "running") return [{ info: snapshot(job) }, jobs]
+        const status: Exclude<Status, "running"> = Exit.isSuccess(exit)
+          ? "completed"
+          : Cause.hasInterruptsOnly(exit.cause)
+            ? "cancelled"
+            : "error"
+        const next = {
+          ...job,
+          blockingSessions: new Map<SessionSchema.ID, number>(),
+          info: {
+            ...job.info,
+            status,
+            completed_at,
+            ...(Exit.isSuccess(exit) ? { output: exit.value } : {}),
+            ...(Exit.isFailure(exit) ? { error: errorText(Cause.squash(exit.cause)) } : {}),
+          },
+        }
+        if (status !== "cancelled") yield* persistBackground(next)
+        return [{ info: snapshot(next), done: job.done, scope: job.scope }, new Map(jobs).set(id, next)]
+      }),
+    )
     if (result.info && result.done) yield* Deferred.succeed(result.done, result.info).pipe(Effect.ignore)
     if (result.scope) {
       yield* Scope.close(result.scope, Exit.void).pipe(Effect.forkIn(state.scope, { startImmediately: true }))
@@ -170,22 +220,6 @@ export const make = Effect.gen(function* () {
     return result.info
   })
 
-  const fork = Effect.fnUntraced(function* (
-    scope: Scope.Scope,
-    id: string,
-    token: object,
-    run: Effect.Effect<string, unknown>,
-  ) {
-    return yield* run.pipe(
-      Effect.matchCauseEffect({
-        onSuccess: (output) => settle(id, token, Exit.succeed(output)),
-        onFailure: (cause) => settle(id, token, Exit.failCause(cause)),
-      }),
-      Effect.asVoid,
-      Effect.forkIn(scope, { startImmediately: true }),
-    )
-  })
-
   const get: Interface["get"] = Effect.fn("Job.get")(function* (id) {
     const job = (yield* SynchronizedRef.get(state.jobs)).get(id)
     if (!job) return undefined
@@ -201,10 +235,10 @@ export const make = Effect.gen(function* () {
         const backgrounded = yield* Deferred.make<Info>()
         const result = yield* SynchronizedRef.modifyEffect(
           state.jobs,
-          Effect.fnUntraced(function* (jobs) {
+          Effect.fnUntraced(function* (jobs): Effect.fn.Return<readonly [StartResult, Map<string, Active>]> {
             const existing = jobs.get(id)
             if (existing?.info.status === "running") {
-              return [{ info: snapshot(existing) }, jobs] as readonly [StartResult, Map<string, Active>]
+              return [{ info: snapshot(existing) }, jobs]
             }
             const scope = yield* Scope.fork(state.scope, "parallel")
             const token = {}
@@ -216,6 +250,7 @@ export const make = Effect.gen(function* () {
                 status: "running" as const,
                 started_at,
                 metadata: input.metadata,
+                ...(input.notificationID ? { notificationID: input.notificationID } : {}),
               },
               done,
               backgrounded,
@@ -223,14 +258,18 @@ export const make = Effect.gen(function* () {
               token,
               blockingSessions: new Map<SessionSchema.ID, number>(),
               isBackgrounded: false,
+              recovery: input.recovery,
             }
-            return [{ info: snapshot(job), scope, token }, new Map(jobs).set(id, job)] as readonly [
-              StartResult,
-              Map<string, Active>,
-            ]
+            return [{ info: snapshot(job), scope, token }, new Map(jobs).set(id, job)]
           }),
         )
-        if ("scope" in result) yield* fork(result.scope, id, result.token, restore(input.run))
+        if ("scope" in result)
+          yield* restore(input.run).pipe(
+            Effect.exit,
+            Effect.flatMap((exit) => settle(id, result.token, exit)),
+            Effect.asVoid,
+            Effect.forkIn(result.scope, { startImmediately: true }),
+          )
         return result.info
       }),
     )
@@ -281,20 +320,31 @@ export const make = Effect.gen(function* () {
     ).pipe(Effect.ensuring(removeBlock(input)))
   })
 
-  const background: Interface["background"] = Effect.fn("Job.background")(function* (id) {
-    const result = yield* SynchronizedRef.modify(
-      state.jobs,
-      (jobs): readonly [BackgroundResult, Map<string, Active>] => {
-        const job = jobs.get(id)
-        if (!job || job.info.status !== "running") return [{}, jobs]
-        if (job.isBackgrounded) return [{ info: snapshot(job) }, jobs]
-        const next = {
-          ...job,
-          isBackgrounded: true,
-          blockingSessions: new Map<SessionSchema.ID, number>(),
-        }
-        return [{ info: snapshot(next), backgrounded: job.backgrounded }, new Map(jobs).set(id, next)]
+  const markBackground = Effect.fnUntraced(function* (job: Active) {
+    const next = {
+      ...job,
+      isBackgrounded: true,
+      blockingSessions: new Map<SessionSchema.ID, number>(),
+      info: {
+        ...job.info,
+        ...(job.recovery ? { notificationID: job.info.notificationID ?? SessionMessage.ID.create() } : {}),
       },
+    }
+    yield* persistBackground(next)
+    return next
+  })
+
+  const background: Interface["background"] = Effect.fn("Job.background")(function* (id) {
+    const result = yield* SynchronizedRef.modifyEffect(
+      state.jobs,
+      Effect.fnUntraced(function* (jobs): Effect.fn.Return<readonly [BackgroundResult, Map<string, Active>]> {
+        const job = jobs.get(id)
+        // Recoverable work may finish before the caller backgrounds it.
+        if (!job || (job.info.status !== "running" && !job.recovery)) return [{}, jobs]
+        if (job.isBackgrounded) return [{ info: snapshot(job) }, jobs]
+        const next = yield* markBackground(job)
+        return [{ info: snapshot(next), backgrounded: job.backgrounded }, new Map(jobs).set(id, next)]
+      }),
     )
     if (result.info && result.backgrounded)
       yield* Deferred.succeed(result.backgrounded, result.info).pipe(Effect.ignore)
@@ -302,60 +352,83 @@ export const make = Effect.gen(function* () {
   })
 
   const backgroundAll: Interface["backgroundAll"] = Effect.fn("Job.backgroundAll")(function* (input) {
-    const result = yield* SynchronizedRef.modify(
+    const result = yield* SynchronizedRef.modifyEffect(
       state.jobs,
-      (jobs): readonly [BackgroundResult[], Map<string, Active>] => {
-        const results: BackgroundResult[] = []
+      Effect.fnUntraced(function* (jobs): Effect.fn.Return<
+        readonly [Required<BackgroundResult>[], Map<string, Active>]
+      > {
+        const results: Required<BackgroundResult>[] = []
         const next = new Map(jobs)
         for (const [id, job] of jobs) {
           if (job.info.status !== "running") continue
           if (job.isBackgrounded) continue
           if (input.type !== undefined && job.info.type !== input.type) continue
           if (!job.blockingSessions.has(input.sessionID)) continue
-          const updated = {
-            ...job,
-            isBackgrounded: true,
-            blockingSessions: new Map<SessionSchema.ID, number>(),
-          }
+          const updated = yield* markBackground(job)
           results.push({ info: snapshot(updated), backgrounded: job.backgrounded })
           next.set(id, updated)
         }
         return [results, next]
-      },
+      }),
     )
-    yield* Effect.forEach(
-      result,
-      (item) => (item.info && item.backgrounded ? Deferred.succeed(item.backgrounded, item.info) : Effect.void),
-      { discard: true },
-    )
-    return result.flatMap((item) => (item.info ? [item.info] : []))
+    yield* Effect.forEach(result, (item) => Deferred.succeed(item.backgrounded, item.info), { discard: true })
+    return result.map((item) => item.info)
   })
 
   const cancel: Interface["cancel"] = Effect.fn("Job.cancel")(function* (id) {
     const completed_at = yield* Clock.currentTimeMillis
-    const result = yield* SynchronizedRef.modify(state.jobs, (jobs): readonly [FinishResult, Map<string, Active>] => {
-      const job = jobs.get(id)
-      if (!job) return [{}, jobs]
-      if (job.info.status !== "running") return [{ info: snapshot(job) }, jobs]
-      const next = {
-        ...job,
-        blockingSessions: new Map<SessionSchema.ID, number>(),
-        info: {
-          ...job.info,
-          status: "cancelled" as const,
-          completed_at,
-        },
-      }
-      return [{ info: snapshot(next), done: job.done, scope: job.scope }, new Map(jobs).set(id, next)]
-    })
+    const result = yield* SynchronizedRef.modifyEffect(
+      state.jobs,
+      Effect.fnUntraced(function* (jobs): Effect.fn.Return<readonly [FinishResult, Map<string, Active>]> {
+        const job = jobs.get(id)
+        if (!job) return [{}, jobs]
+        if (job.info.status !== "running") return [{ info: snapshot(job) }, jobs]
+        const next = {
+          ...job,
+          blockingSessions: new Map<SessionSchema.ID, number>(),
+          info: {
+            ...job.info,
+            status: "cancelled" as const,
+            completed_at,
+          },
+        }
+        yield* persistBackground(next)
+        return [{ info: snapshot(next), done: job.done, scope: job.scope }, new Map(jobs).set(id, next)]
+      }),
+    )
     if (result.info && result.done) yield* Deferred.succeed(result.done, result.info).pipe(Effect.ignore)
     if (result.scope) yield* Scope.close(result.scope, Exit.void)
     return result.info
   })
 
-  return Service.of({ get, start, wait, block, background, backgroundAll, cancel })
+  const pendingBackground: Interface["pendingBackground"] = Effect.gen(function* () {
+    const recovered: Background[] = []
+    let after: string | undefined
+    do {
+      const page = yield* kv.scan({ prefix: backgroundPrefix, after })
+      recovered.push(...Array.filterMap(page.entries, (entry) => decodeBackground(entry.value)))
+      after = page.next
+    } while (after)
+    return recovered
+  }).pipe(Effect.withSpan("Job.pendingBackground"))
+
+  const completeBackground: Interface["completeBackground"] = Effect.fn("Job.completeBackground")((notificationID) =>
+    kv.remove(`${backgroundPrefix}${notificationID}`),
+  )
+
+  return Service.of({
+    get,
+    start,
+    wait,
+    block,
+    background,
+    backgroundAll,
+    cancel,
+    pendingBackground,
+    completeBackground,
+  })
 })
 
 const layer = Layer.effect(Service, make)
 
-export const node = makeGlobalNode({ service: Service, layer, deps: [] })
+export const node = makeGlobalNode({ service: Service, layer, deps: [KV.node] })

packages/core/src/mcp/client.ts

@@ -207,7 +207,10 @@ export const connect = Effect.fnUntraced(function* (
     }
     if (!URL.canParse(config.url))
       return yield* new ConnectError({ server, message: `Invalid MCP URL for "${server}"` })
-    return new StreamableHTTPClientTransport(new URL(config.url), {
+    // Prefer raw tools for our Code Mode without changing the configured URL used for OAuth identity.
+    const url = new URL(config.url)
+    if (config.codemode !== false && !url.searchParams.has("codemode")) url.searchParams.set("codemode", "false")
+    return new StreamableHTTPClientTransport(url, {
       requestInit: config.headers ? { headers: config.headers } : undefined,
       authProvider,
     })

packages/core/src/plugin/host.ts

@@ -377,14 +377,8 @@ export const make = Effect.fn("PluginHost.make")(function* (plugin: Interface, p
       hook: (name, callback) => hooks.register("shell", name, callback),
     },
     tool: {
-      transform: (callback) =>
-        tools
-          .transform((draft) =>
-            callback({
-              add: (tool) => draft.add(tool),
-            }),
-          )
-          .pipe(Effect.orDie, Effect.as({ dispose: Effect.void })),
+      transform: tools.transform,
+      reload: tools.reload,
       hook: (name, callback) => hooks.register("tool", name, callback),
     },
     vcs: {

packages/core/src/plugin/mcp-codemode-exclusion.ts

@@ -4,7 +4,7 @@ import { define } from "@opencode-ai/plugin/effect/plugin"
 import { Effect } from "effect"
 
 // These servers provide Code Mode, so expose them directly instead of nesting them inside OpenCode Code Mode.
-const urls = [/^https:\/\/mcp\.cloudflare\.com\/mcp$/, /^https:\/\/executor\.sh\/[^/]+\/mcp$/]
+const urls = [/^https:\/\/executor\.sh\/[^/]+\/mcp$/]
 
 export const Plugin = define({
   id: "opencode.mcp.codemode.exclusion",

packages/core/src/plugin/runtime.ts

@@ -29,7 +29,7 @@ export interface Interface {
     | "wait"
     | "context"
   >
-  readonly job: Pick<Job.Interface, "start" | "wait" | "block" | "background" | "cancel">
+  readonly job: Pick<Job.Interface, "start" | "wait" | "block" | "background" | "cancel" | "completeBackground">
   readonly location: {
     readonly agent: {
       readonly list: (
@@ -92,6 +92,8 @@ export const layerWithCell = (cell: Cell) =>
         block: (input) => require(cell, (runtime) => runtime.job.block(input)),
         background: (id) => require(cell, (runtime) => runtime.job.background(id)),
         cancel: (id) => require(cell, (runtime) => runtime.job.cancel(id)),
+        completeBackground: (notificationID) =>
+          require(cell, (runtime) => runtime.job.completeBackground(notificationID)),
       },
       location: {
         agent: {

packages/core/src/session/execution.ts

@@ -3,6 +3,7 @@ export * as SessionExecution from "./execution.js"
 import { Cause, Context, Effect, Exit, Layer } from "effect"
 import { Bus } from "../bus.js"
 import { Database } from "../database/database.js"
+import { Job } from "../job.js"
 import { LocationServiceMap } from "../location-service-map.js"
 import { makeGlobalNode } from "@opencode-ai/util/effect/app-node"
 import { SessionEvent } from "./event.js"
@@ -52,6 +53,7 @@ export const layer = Layer.effect(
     const store = yield* SessionStore.Service
     const locations = yield* LocationServiceMap.Service
     const bus = yield* Bus.Service
+    const jobs = yield* Job.Service
     const db = (yield* Database.Service).db
     const reportLifecycle = <A>(sessionID: SessionSchema.ID, effect: Effect.Effect<A>) =>
       effect.pipe(
@@ -118,6 +120,7 @@ export const layer = Layer.effect(
             if (outcome.type === "interrupted") {
               // A user cancel releases the claim: the turn must not resurrect at the next
               // boot. Shutdown interruption keeps it for restart continuity.
+              if (outcome.reason === "user") yield* jobs.cancel(sessionID)
               yield* bus.publish(
                 SessionEvent.Execution.Interrupted,
                 { sessionID, reason: outcome.reason },
@@ -167,7 +170,7 @@ export const layer = Layer.effect(
 export const node = makeGlobalNode({
   service: Service,
   layer,
-  deps: [SessionStore.node, LocationServiceMap.node, Bus.node, Database.node],
+  deps: [SessionStore.node, LocationServiceMap.node, Bus.node, Database.node, Job.node],
 })
 
 /** Low-level compatibility layer for callers that only need durable Session recording. */

packages/core/src/session/execution/restart.ts

@@ -3,6 +3,8 @@ export * as SessionRestart from "./restart.js"
 import { Context, Effect, Layer } from "effect"
 import { makeGlobalNode } from "@opencode-ai/util/effect/app-node"
 import { Bus } from "../../bus.js"
+import { Job } from "../../job.js"
+import { Session } from "../../session.js"
 import { SessionEvent } from "../event.js"
 import { SessionExecution } from "../execution.js"
 import { SessionSchema } from "../schema.js"
@@ -45,6 +47,9 @@ export interface Interface {
  * process: crash, SIGKILL, isolate eviction, and graceful restart all leave
  * the same durable signature.
  *
+ * Recovery is at-least-once: local coordination prevents concurrent drains,
+ * not repeated external side effects after a crash.
+ *
  * The sweep assumes every orphaned claim's owner is dead. The managed-server
  * protocol guarantees this: a successor is only spawned after the previous
  * process is confirmed dead (client service `kill`/`evict` poll the PID), the
@@ -62,14 +67,16 @@ export const layer = (options?: Options) =>
       const store = yield* SessionStore.Service
       const execution = yield* SessionExecution.Service
       const bus = yield* Bus.Service
+      const jobs = yield* Job.Service
+      const sessions = yield* Session.Service
       const scope = yield* Effect.scope
       const maxAttempts = options?.maxAttempts ?? DEFAULT_MAX_ATTEMPTS
 
-      const resumeOne = Effect.fnUntraced(function* (sessionID: SessionSchema.ID) {
+      const prepareResume = Effect.fnUntraced(function* (sessionID: SessionSchema.ID) {
         // Durable before the resume runs, so a crash inside the resumed turn is
         // counted by the next sweep and the budget cannot be dodged.
         const attempts = yield* store.countResume(sessionID)
-        if (attempts === undefined) return // the Session was deleted since listing
+        if (attempts === undefined) return false
         if (attempts > maxAttempts) {
           // Terminalize instead: the release hook clears the claim and resets the
           // counter atomically with the terminal event.
@@ -78,31 +85,166 @@ export const layer = (options?: Options) =>
             { sessionID, error: RESUME_EXHAUSTED },
             { commit: () => store.release(sessionID) },
           )
-          return
+          return false
         }
         yield* bus.publish(SessionEvent.Synthetic, {
           sessionID,
           text: CONTINUE_AFTER_SERVER_RESTART,
           description: "Continuing after restart",
         })
-        // Forked into the service scope so boot never waits on resumed turns;
-        // resuming an already-live Session joins its execution. Drain failures
-        // are logged and durably recorded by the execution layer.
-        yield* execution.resume(sessionID).pipe(Effect.ignore, Effect.forkIn(scope))
+        return true
+      })
+
+      const recoverShell = Effect.fnUntraced(function* (
+        background: Job.Background,
+        recovery: Extract<Job.Recovery, { kind: "shell" }>,
+      ) {
+        const state = background.status === "running" ? "cancelled" : background.status
+        const text =
+          background.status === "running"
+            ? "Command cancelled because the server restarted"
+            : state === "completed"
+              ? (background.output ?? "Command completed")
+              : state === "error"
+                ? (background.error ?? "Command failed")
+                : "Command cancelled"
+
+        yield* sessions
+          .synthetic({
+            id: background.notificationID,
+            sessionID: recovery.sessionID,
+            description: recovery.command,
+            text: `<shell id="${background.id}" state="${state}" command="${recovery.command}">\n${text}\n</shell>`,
+            metadata: {
+              source: "shell",
+              jobID: background.id,
+              shellID: recovery.shellID,
+              state,
+            },
+            resume: false,
+          })
+          .pipe(
+            Effect.catchTag("Session.NotFoundError", () => Effect.void),
+            Effect.orDie,
+          )
+        yield* jobs.completeBackground(background.notificationID)
+      })
+
+      const recoverSubagent = Effect.fnUntraced(function* (
+        background: Job.Background,
+        recovery: Extract<Job.Recovery, { kind: "subagent" }>,
+        suspended: ReadonlySet<SessionSchema.ID>,
+      ) {
+        const child = yield* store.get(recovery.childSessionID)
+        if (!child || child.parentID !== recovery.parentSessionID || !(yield* store.get(recovery.parentSessionID))) {
+          yield* jobs.completeBackground(background.notificationID)
+          return
+        }
+
+        const notify = Effect.fnUntraced(function* (result: Pick<Job.Background, "status" | "output" | "error">) {
+          if (result.status === "running") return
+          const text =
+            result.status === "completed"
+              ? (result.output ?? "Subagent completed without a text response.")
+              : result.status === "error"
+                ? (result.error ?? "Subagent failed")
+                : "Subagent cancelled"
+          yield* sessions
+            .synthetic({
+              id: background.notificationID,
+              sessionID: recovery.parentSessionID,
+              ...(suspended.has(recovery.parentSessionID) ? { resume: false } : {}),
+              description: recovery.description,
+              text: `<subagent sessionID="${recovery.childSessionID}" state="${result.status}" description="${recovery.description}">\n${text}\n</subagent>`,
+              metadata: {
+                source: "subagent",
+                childID: recovery.childSessionID,
+                agent: recovery.agent,
+                state: result.status,
+              },
+            })
+            .pipe(Effect.orDie)
+          yield* jobs.completeBackground(background.notificationID)
+        })
+
+        if (background.status !== "running") {
+          yield* notify(background)
+          return
+        }
+        if ((yield* execution.active).has(recovery.childSessionID)) return
+        if (!(yield* prepareResume(recovery.childSessionID))) {
+          yield* notify({ status: "error", error: RESUME_EXHAUSTED.message })
+          return
+        }
+
+        yield* jobs.start({
+          id: background.id,
+          type: "subagent",
+          title: recovery.description,
+          notificationID: background.notificationID,
+          recovery,
+          run: execution.resume(recovery.childSessionID).pipe(
+            Effect.andThen(store.context(recovery.childSessionID)),
+            Effect.map((messages) => {
+              const assistant = messages.findLast(
+                (message) =>
+                  message.type === "assistant" && message.time.completed !== undefined && message.error === undefined,
+              )
+              if (assistant?.type !== "assistant") return "Subagent completed without a text response."
+              return (
+                assistant.content
+                  .filter((part) => part.type === "text")
+                  .map((part) => part.text)
+                  .join("") || "Subagent completed without a text response."
+              )
+            }),
+          ),
+        })
+        yield* jobs.background(background.id)
+        yield* jobs.wait({ id: background.id }).pipe(
+          Effect.flatMap((result) => (result.info ? notify(result.info) : Effect.void)),
+          Effect.ignore,
+          Effect.forkIn(scope),
+        )
       })
 
       return Service.of({
         resumeSuspendedSessions: Effect.gen(function* () {
-          // Child claims never drive recovery (children are not resumed), so a
-          // dead child's claim is noise no terminal will ever release. Clearing
-          // is safe even against a live child: claims are recovery markers, not
-          // locks, and children are excluded from that recovery.
-          yield* store.releaseChildClaims
           const active = yield* execution.active
-          // Sessions already draining in this process keep their claim; resuming
-          // them would only inject a stray continuation into a live turn.
-          const orphaned = (yield* store.listSuspended()).filter((sessionID) => !active.has(sessionID))
-          yield* Effect.forEach(orphaned, resumeOne, { concurrency: "unbounded", discard: true })
+          // Early notices wait for root recovery's accounting, including roots that exhaust their budget.
+          const suspended = new Set((yield* store.listSuspended()).filter((sessionID) => !active.has(sessionID)))
+          const pending = yield* jobs.pendingBackground
+          yield* store.releaseChildClaims(
+            pending.flatMap((background) =>
+              background.status === "running" && background.recovery.kind === "subagent"
+                ? [background.recovery.childSessionID]
+                : [],
+            ),
+          )
+          yield* Effect.forEach(
+            pending,
+            Effect.fnUntraced(function* (background) {
+              if ((yield* jobs.get(background.id))?.status === "running") return
+              const recovery = background.recovery
+              yield* recovery.kind === "shell"
+                ? recoverShell(background, recovery)
+                : recoverSubagent(background, recovery, suspended)
+            }),
+            { discard: true },
+          )
+
+          // Background completion can wake a parent, so inspect local ownership only after recovery.
+          const resumed = yield* execution.active
+          yield* Effect.forEach(
+            (yield* store.listSuspended()).filter((sessionID) => !resumed.has(sessionID)),
+            (sessionID) =>
+              execution
+                .resume(sessionID)
+                .pipe(Effect.ignore, Effect.forkIn(scope), Effect.when(prepareResume(sessionID))),
+            { concurrency: "unbounded", discard: true },
+          )
+          // Async observers consult this set at delivery; later completions wake parents normally.
+          suspended.clear()
         }),
       })
     }),
@@ -111,5 +253,5 @@ export const layer = (options?: Options) =>
 export const node = makeGlobalNode({
   service: Service,
   layer: layer(),
-  deps: [SessionStore.node, SessionExecution.node, Bus.node],
+  deps: [SessionStore.node, SessionExecution.node, Bus.node, Job.node, Session.node],
 })

packages/core/src/session/runner/llm.ts

@@ -78,83 +78,143 @@ const layer = Layer.effect(
       readonly continuation?: Continuation
       readonly promotable?: SessionInbox.Promotable
     }) {
+      const sessionID = input.sessionID
       let force = input.force
-      let continuation = input.continuation
+      let continuing = input.continuation !== undefined
+      let step = input.continuation?.step ?? 1
+      let entering = true
       const promotable = input.promotable ?? "input"
-      if (!force && !continuation && !(yield* eligible(input.sessionID, promotable))) return DrainResult.Complete()
-      yield* plugins.flush
-      yield* settleStaleToolCalls(input.sessionID)
-      while (true) {
-        // Scope gates input promotion, not a between-step control that is next in line.
-        if (yield* runPendingCompaction(input.sessionID, "input")) {
-          force = false
-          continue
-        }
-        if (yield* runPendingMove(input.sessionID, "input")) return DrainResult.Moved({})
-        if (!force && !continuation && !(yield* SessionInbox.has(db, input.sessionID, promotable)))
+      if (!force && !continuing) {
+        const pending = yield* SessionInbox.nextPromotable(db, sessionID, "input")
+        if (
+          !pending ||
+          (pending.delivery === "queue" &&
+            promotable === "steer" &&
+            pending.type !== "compaction" &&
+            pending.type !== "move")
+        )
           return DrainResult.Complete()
-        const result = yield* runSteps(input.sessionID, continuation, promotable)
-        if (result._tag === "Moved") return result
-        force = false
-        continuation = undefined
       }
-    })
+      yield* plugins.flush
+      yield* settleStaleToolCalls(sessionID)
 
-    const eligible = Effect.fnUntraced(function* (sessionID: SessionSchema.ID, promotable: SessionInbox.Promotable) {
-      if (yield* SessionInbox.has(db, sessionID, promotable)) return true
-      if (promotable === "input") return false
-      const next = yield* SessionInbox.nextPromotable(db, sessionID, "input")
-      return next?.type === "compaction" || next?.type === "move"
-    })
+      const advanceToStep = Effect.fn("SessionRunner.advanceToStep")(() =>
+        Effect.uninterruptibleMask((restore) =>
+          Effect.gen(function* () {
+            while (true) {
+              // Location entry and idle boundaries allow queued controls, not necessarily queued prompts.
+              const pending = yield* SessionInbox.serialized(
+                sessionID,
+                Effect.gen(function* () {
+                  const next = yield* SessionInbox.nextPromotable(
+                    db,
+                    sessionID,
+                    entering || !continuing ? "input" : "steer",
+                  )
+                  if (next?.type === "compaction")
+                    yield* bus.publishAll([
+                      [SessionEvent.InboxDelivered, { sessionID, inboxID: next.id }],
+                      [SessionEvent.Compaction.Started, { sessionID, reason: "manual", recent: "", inputID: next.id }],
+                    ])
+                  if (next?.type === "move")
+                    yield* restore(
+                      Effect.gen(function* () {
+                        yield* modelTransport.close(sessionID)
+                        yield* bus.publishAll([
+                          [SessionEvent.InboxDelivered, { sessionID, inboxID: next.id }],
+                          [SessionEvent.Moved, { sessionID, ...next.payload }],
+                        ])
+                      }),
+                    )
+                  return next
+                }),
+              )
+              if (!continuing && pending?.delivery !== "steer") {
+                entering = true
+                step = 1
+              }
+              if (pending?.type === "move")
+                return DrainResult.Moved({ continuation: !entering && continuing ? { step } : undefined })
+              if (pending?.type === "compaction") {
+                const session = yield* store.get(sessionID)
+                if (!session) return yield* Effect.die(new Error(`Session not found: ${sessionID}`))
+                const compacted = yield* restore(
+                  Effect.gen(function* () {
+                    return yield* compaction.compactManual({
+                      session,
+                      messages: yield* store.context(sessionID),
+                      inputID: pending.id,
+                      started: true,
+                    })
+                  }),
+                ).pipe(Effect.exit)
+                if (Exit.isFailure(compacted)) {
+                  yield* bus.publish(SessionEvent.Compaction.Failed, {
+                    sessionID,
+                    reason: "manual",
+                    error: Cause.hasInterruptsOnly(compacted.cause)
+                      ? { type: "aborted", message: "Compaction cancelled" }
+                      : { type: "compaction.failed", message: Cause.pretty(compacted.cause) },
+                    inputID: pending.id,
+                  })
+                  return yield* Effect.failCause(compacted.cause)
+                }
+                force = false
+                continue
+              }
+              if (!force && !continuing && (!pending || (pending.delivery === "queue" && promotable === "steer")))
+                return DrainResult.Complete()
+              return yield* restore(
+                Effect.gen(function* () {
+                  const selected = yield* prepareContext(sessionID)
+                  const promoted = yield* SessionInbox.promote(
+                    db,
+                    bus,
+                    sessionID,
+                    entering && !continuing ? promotable : "steer",
+                  )
+                  if (promoted > 0 && !selected.session.parentID && SessionTitle.isUntitled(selected.session))
+                    yield* FiberMap.run(titles, sessionID, title.generate(sessionID).pipe(Effect.ignore), {
+                      onlyIfMissing: true,
+                    })
+                  if (promoted > 0) step = 1
+                  return { _tag: "Ready" as const, context: yield* context.load(selected) }
+                }),
+              )
+            }
+          }),
+        ),
+      )
 
-    /** Queued inputs wait until the current model work reaches idle; later Steps absorb only steers. */
-    const runSteps = Effect.fn("SessionRunner.runSteps")(function* (
-      sessionID: SessionSchema.ID,
-      continuation: Continuation | undefined,
-      drainPromotable: SessionInbox.Promotable,
-    ) {
-      let promotable: SessionInbox.Promotable = continuation ? "steer" : drainPromotable
-      let step = continuation?.step ?? 1
-      let next = continuation
-      let first = true
       while (true) {
-        if (yield* runPendingCompaction(sessionID, "steer")) continue
-        if (yield* runPendingMove(sessionID, "steer")) return DrainResult.Moved({ continuation: next })
-        if (!first && !next && !(yield* SessionInbox.has(db, sessionID, "steer"))) return DrainResult.Complete()
-        const result = yield* runStep(sessionID, promotable, step)
-        first = false
-        promotable = "steer"
-        step = result.step + 1
-        next = result.needsContinuation ? { step } : undefined
+        const next = yield* advanceToStep()
+        if (next._tag !== "Ready") return next
+        continuing = yield* runStep(next.context, step)
+        step++
+        force = false
+        entering = false
       }
     })
 
+    const prepareContext = Effect.fn("SessionRunner.prepareContext")(function* (sessionID: SessionSchema.ID) {
+      const selected = yield* context.select(sessionID)
+      // A blocked initial instruction baseline must leave admitted input pending.
+      yield* InstructionState.prepare(db, bus, selected.instructions, sessionID)
+      return selected
+    })
+
     /** Owns logical Step policy; each attempt owns its streaming, tools, and durable settlement. */
-    const runStep = Effect.fn("SessionRunner.runStep")(function* (
-      sessionID: SessionSchema.ID,
-      promotable: SessionInbox.Promotable,
-      step: number,
-    ) {
+    const runStep = Effect.fn("SessionRunner.runStep")(function* (first: SessionContext.Loaded, step: number) {
+      const sessionID = first.session.id
       let assistantMessageID = SessionMessage.ID.create()
       const retry = yield* Schedule.toStepWithSleep(SessionRunnerRetry.schedule(bus, sessionID))
-      let currentPromotable: SessionInbox.Promotable | undefined = promotable
-      let currentStep = step
+      let initial: SessionContext.Loaded | undefined = first
       let recoverOverflow = true
       let recoverContinuation = true
       while (true) {
-        const selected = yield* context.select(sessionID)
-        // A blocked initial instruction baseline must leave admitted input pending.
-        yield* InstructionState.prepare(db, bus, selected.instructions, selected.session.id)
-        const promoted = currentPromotable
-          ? yield* SessionInbox.promote(db, bus, selected.session.id, currentPromotable)
-          : 0
-        if (promoted > 0 && !selected.session.parentID && SessionTitle.isUntitled(selected.session))
-          yield* FiberMap.run(titles, sessionID, title.generate(sessionID).pipe(Effect.ignore), {
-            onlyIfMissing: true,
-          })
-        currentStep = promoted > 0 ? 1 : currentStep
-        currentPromotable = undefined
-        const loaded = yield* context.load(selected)
+        // Reuse boundary preparation once; retries refresh context without delivering more input.
+        const loaded = initial ?? (yield* prepareContext(sessionID).pipe(Effect.flatMap(context.load)))
+        initial = undefined
         const compactionInput = { session: loaded.session, messages: loaded.messages, resolved: loaded.model }
         if (compaction.required(compactionInput)) {
           const compacted = yield* compaction.compact(compactionInput)
@@ -162,7 +222,7 @@ const layer = Layer.effect(
           assistantMessageID = SessionMessage.ID.create()
           continue
         }
-        const stepLimitReached = loaded.agent.info.steps !== undefined && currentStep >= loaded.agent.info.steps
+        const stepLimitReached = loaded.agent.info.steps !== undefined && step >= loaded.agent.info.steps
         const transcript = SessionModelRequest.baseTranscript({
           agent: loaded.agent.info,
           model: loaded.model,
@@ -197,7 +257,7 @@ const layer = Layer.effect(
               : Effect.succeed(false),
           ),
         })
-        if (outcome._tag === "Completed") return { needsContinuation: outcome.needsContinuation, step: currentStep }
+        if (outcome._tag === "Completed") return outcome.needsContinuation
         if (outcome._tag === "Retry" || outcome._tag === "Continue") {
           yield* retry({ cause: outcome.cause, error: outcome.error, assistantMessageID }).pipe(
             Pull.catchDone(() =>
@@ -223,77 +283,6 @@ const layer = Layer.effect(
       }
     })
 
-    const runPendingCompaction = Effect.fn("SessionRunner.runPendingCompaction")(function* (
-      sessionID: SessionSchema.ID,
-      promotable: SessionInbox.Promotable,
-    ) {
-      return yield* Effect.uninterruptibleMask((restore) =>
-        Effect.gen(function* () {
-          const pending = yield* SessionInbox.serialized(
-            sessionID,
-            Effect.gen(function* () {
-              const selected = yield* SessionInbox.nextPromotable(db, sessionID, promotable)
-              if (selected?.type !== "compaction") return
-              yield* bus.publishAll([
-                [SessionEvent.InboxDelivered, { sessionID, inboxID: selected.id }],
-                [SessionEvent.Compaction.Started, { sessionID, reason: "manual", recent: "", inputID: selected.id }],
-              ])
-              return selected
-            }),
-          )
-          if (pending?.type !== "compaction") return false
-          const session = yield* getSession(sessionID)
-          const compacted = yield* restore(
-            Effect.gen(function* () {
-              return yield* compaction.compactManual({
-                session,
-                messages: yield* store.context(sessionID),
-                inputID: pending.id,
-                started: true,
-              })
-            }),
-          ).pipe(Effect.exit)
-          if (Exit.isSuccess(compacted)) return true
-          yield* bus.publish(SessionEvent.Compaction.Failed, {
-            sessionID,
-            reason: "manual",
-            error: Cause.hasInterruptsOnly(compacted.cause)
-              ? { type: "aborted", message: "Compaction cancelled" }
-              : { type: "compaction.failed", message: Cause.pretty(compacted.cause) },
-            inputID: pending.id,
-          })
-          return yield* Effect.failCause(compacted.cause)
-        }),
-      )
-    })
-
-    const runPendingMove = Effect.fn("SessionRunner.runPendingMove")(function* (
-      sessionID: SessionSchema.ID,
-      promotable: SessionInbox.Promotable,
-    ) {
-      return yield* SessionInbox.serialized(
-        sessionID,
-        Effect.gen(function* () {
-          const pending = yield* SessionInbox.nextPromotable(db, sessionID, promotable)
-          if (pending?.type !== "move") return false
-          yield* modelTransport.close(sessionID)
-          yield* bus.publishAll([
-            [SessionEvent.InboxDelivered, { sessionID, inboxID: pending.id }],
-            [
-              SessionEvent.Moved,
-              {
-                sessionID,
-                location: pending.payload.location,
-                projectID: pending.payload.projectID,
-                subpath: pending.payload.subpath,
-              },
-            ],
-          ])
-          return true
-        }),
-      )
-    })
-
     const settleStaleToolCalls = Effect.fn("SessionRunner.settleStaleToolCalls")(function* (
       sessionID: SessionSchema.ID,
     ) {
@@ -301,23 +290,24 @@ const layer = Layer.effect(
         if (message.type !== "assistant") continue
         for (const tool of message.content) {
           if (tool.type !== "tool" || (tool.state.status !== "streaming" && tool.state.status !== "running")) continue
+          const metadata = tool.state.status === "running" ? tool.state.metadata : undefined
+          const childID =
+            tool.name === "subagent" && typeof metadata?.sessionID === "string" ? metadata.sessionID : undefined
           yield* bus.publish(SessionEvent.Tool.Failed, {
             sessionID,
             assistantMessageID: message.id,
             id: tool.id,
-            error: { type: "aborted", message: `Tool execution interrupted: ${tool.name}` },
+            error: {
+              type: "aborted",
+              message: `Tool execution interrupted: ${tool.name}${childID ? ` (sessionID: ${childID})` : ""}`,
+            },
+            ...(metadata && Object.keys(metadata).length > 0 ? { metadata } : {}),
             executed: tool.executed === true,
           })
         }
       }
     })
 
-    const getSession = Effect.fn("SessionRunner.getSession")(function* (sessionID: SessionSchema.ID) {
-      const session = yield* store.get(sessionID)
-      if (!session) return yield* Effect.die(new Error(`Session not found: ${sessionID}`))
-      return session
-    })
-
     return Service.of({ drain })
   }),
 )

packages/core/src/session/runner/publish-llm-event.ts

@@ -327,7 +327,10 @@ export const createLLMEventPublisher = (bus: Pick<Bus.Interface, "publish">, inp
       sessionID: input.sessionID,
       assistantMessageID,
       id,
-      error,
+      error:
+        tool.name === "subagent" && error.type === "aborted" && typeof tool.progress?.sessionID === "string"
+          ? { ...error, message: `${error.message} (sessionID: ${tool.progress.sessionID})` }
+          : error,
       ...failureSnapshot(tool, metadata),
       executed: tool.providerExecuted,
     })

packages/core/src/session/store.ts

@@ -1,6 +1,6 @@
 export * as SessionStore from "./store.js"
 
-import { and, eq, isNotNull, isNull, sql } from "drizzle-orm"
+import { and, eq, isNotNull, isNull, notInArray, sql } from "drizzle-orm"
 import { Context, Effect, Layer } from "effect"
 import { Database } from "../database/database.js"
 import { makeGlobalNode } from "@opencode-ai/util/effect/app-node"
@@ -18,9 +18,8 @@ export interface Interface {
     messageID: SessionMessage.ID,
   ) => Effect.Effect<{ readonly sessionID: Session.ID; readonly message: SessionMessage.Info } | undefined>
   /**
-   * Top-level Sessions holding an execution claim. Child (subagent) Sessions
-   * are excluded: a resumed parent re-runs its tool call and spawns fresh
-   * children, so resuming orphaned children would duplicate their work.
+   * Top-level Sessions holding an execution claim. Recoverable background
+   * children are resumed separately through their durable Job records.
    */
   readonly listSuspended: () => Effect.Effect<ReadonlyArray<Session.ID>>
   /**
@@ -33,11 +32,10 @@ export interface Interface {
   /** Releases the claim and resets resume accounting. Terminal events call this on commit. */
   readonly release: (sessionID: Session.ID) => Effect.Effect<void>
   /**
-   * Clears orphaned child (subagent) claims. Children are never resumed
-   * independently, so a dead child's claim is noise no terminal will ever
-   * release.
+   * Clears orphaned child claims except children owned by recoverable
+   * background subagent jobs.
    */
-  readonly releaseChildClaims: Effect.Effect<void>
+  readonly releaseChildClaims: (recoverable: ReadonlyArray<Session.ID>) => Effect.Effect<void>
   /**
    * Durably counts one more resume of an orphaned claim, returning the new
    * total — or undefined when the Session no longer exists.
@@ -103,12 +101,20 @@ const layer = Layer.effect(
           .run()
           .pipe(Effect.orDie)
       }),
-      releaseChildClaims: db
-        .update(SessionTable)
-        .set({ time_suspended: null, resume_attempts: 0, time_updated: sql`${SessionTable.time_updated}` })
-        .where(and(isNotNull(SessionTable.time_suspended), isNotNull(SessionTable.parent_id)))
-        .run()
-        .pipe(Effect.orDie, Effect.asVoid, Effect.withSpan("SessionStore.releaseChildClaims")),
+      releaseChildClaims: Effect.fn("SessionStore.releaseChildClaims")((recoverable) =>
+        db
+          .update(SessionTable)
+          .set({ time_suspended: null, resume_attempts: 0, time_updated: sql`${SessionTable.time_updated}` })
+          .where(
+            and(
+              isNotNull(SessionTable.time_suspended),
+              isNotNull(SessionTable.parent_id),
+              recoverable.length > 0 ? notInArray(SessionTable.id, Array.from(recoverable)) : undefined,
+            ),
+          )
+          .run()
+          .pipe(Effect.orDie, Effect.asVoid),
+      ),
       countResume: Effect.fn("SessionStore.countResume")(function* (sessionID) {
         const row = yield* db
           .update(SessionTable)

packages/core/src/shell.ts

@@ -134,8 +134,8 @@ const layer = () =>
         Effect.gen(function* () {
           for (const session of sessions.values()) {
             if (session.timeoutFiber) yield* Fiber.interrupt(session.timeoutFiber)
-            // Unblock waiters still pending at teardown; succeed is a no-op once already resolved.
-            yield* Deferred.fail(session.done, new NotFoundError({ id: Shell.ID.make(session.info.id) }))
+            // Teardown interrupts pending commands; it is not a terminal command failure.
+            yield* Deferred.interrupt(session.done)
           }
           sessions.clear()
           exitOrder.length = 0

packages/core/src/shell/parse.ts

@@ -7,12 +7,17 @@ import path from "path"
 import type { Node } from "web-tree-sitter"
 import { shellParserWasm } from "#shell-parser-wasm"
 import { ShellSelect } from "./select.js"
+import { Wildcard } from "../util/wildcard.js"
 
 type Part = { type: string; text: string }
-type SourceToken = { raw: string; value: string }
 const CWD = new Set(["cd", "chdir", "popd", "pushd", "push-location", "set-location"])
 const POWERSHELL_PATH_FLAGS = new Set(["-literalpath", "-path"])
 
+export type Result = {
+  commands: Array<{ resource: string; save: string }>
+  directories: string[]
+}
+
 const ARITY: Record<string, number> = {
   cat: 1,
   cd: 1,
@@ -152,6 +157,7 @@ const ARITY: Record<string, number> = {
   "yarn dlx": 3,
   "yarn run": 3,
 }
+const PREFIX_LENGTH = Math.max(...Object.values(ARITY))
 
 export const scan = Effect.fnUntraced(function* (
   command: string,
@@ -159,7 +165,7 @@ export const scan = Effect.fnUntraced(function* (
   cwd: string,
   options?: { portable?: boolean },
 ) {
-  if (options?.portable) return yield* Effect.promise(() => scanPortable(command, shell, cwd))
+  if (options?.portable) return yield* scanPortable(command, shell, cwd)
   return yield* scanLegacy(command, shell, cwd)
 })
 
@@ -197,416 +203,65 @@ const scanLegacy = Effect.fnUntraced(function* (command: string, shell: string,
   )
 })
 
-async function scanPortable(command: string, shell: string, cwd: string) {
-  const { ShellScan } = await import("./scan.js")
+export const scanPortable = Effect.fnUntraced(function* (command: string, shell: string, cwd: string) {
+  const { ShellScan } = yield* Effect.tryPromise({
+    try: () => import("./scan.js"),
+    catch: (cause) => new Error(`Portable shell scanner failed to load: ${cause}`, { cause }),
+  })
   const powershell = ShellSelect.ps(shell)
   const result = powershell ? ShellScan.scanPowerShell(command) : ShellScan.scan(command)
-  if (result.kind === "opaque") return { commands: [{ resource: command, save: command }], directories: [] }
-  const carriage = powershell ? command.search(/\r(?!\n)/) : -1
-  if (carriage >= 0) return { commands: [], directories: [] }
+  if (result.kind === "opaque")
+    return yield* Effect.fail(new Error(`Portable shell scanner cannot analyze command: ${result.reason}`))
 
-  const parsed = result.commands.reduce(
-    (output, item) => {
-      const index = item[ShellScan.Nested] ? -1 : command.indexOf(item.resource, output.cursor)
-      const offset = item[ShellScan.Nested]
-        ? command.lastIndexOf(item.resource, output.cursor - 1)
-        : index < 0
-          ? command.indexOf(item.resource)
-          : index
-      if (index >= 0) output.cursor = index + item.resource.length
-      const before = command.slice(0, Math.max(0, offset))
-      const name = powershell ? item.words[0]?.toLowerCase() : item.words[0]
-      if (!name) return output
-      if (powershell && name === "<") return output
-      if (
-        powershell &&
-        name === "foreach-object" &&
-        item.words.some((word) => word.startsWith("{")) &&
-        !/\|\s*$/.test(before)
+  const output: Result = { commands: [], directories: [] }
+  for (const item of result.commands) {
+    // The legacy command walk skips declarations, not the substitutions within them.
+    if (item.declaration) continue
+    const words = item.redirectWordCount === undefined ? item.rawWords : item.rawWords.slice(0, item.redirectWordCount)
+    // The shipped PowerShell grammar treats bare statement-head foreach prefixes as control flow.
+    if (powershell && item.statementHead && /^foreach(?:-|$)/i.test(words[0] ?? "")) continue
+    const name = powershell ? words[0]?.toLowerCase() : words[0]
+    if (CWD.has(name)) {
+      output.directories.push(
+        ...directoryArgs(
+          words.flatMap((text): Part[] => {
+            const parameter = powershell ? /^(-(?:literalpath|path)):(.*)$/i.exec(text) : undefined
+            if (parameter)
+              return [
+                { type: "command_parameter", text: parameter[1] },
+                { type: "word", text: parameter[2] },
+              ]
+            return [{ type: powershell && text.startsWith("-") ? "command_parameter" : "word", text }]
+          }),
+          powershell,
+          cwd,
+          shell,
+        ),
       )
-        return output
-      const tokens = powershell ? powerShellSourceTokens(item.resource) : sourceTokens(item.resource)
-      const sourceHead = powershell ? item.words[0] : tokens.find((token) => token.value === item.words[0])?.raw
-      if (CWD.has(name) && (powershell || sourceHead === item.words[0])) {
-        output.directories.push(...portableDirectoryArgs(item.words, tokens, powershell, cwd, shell))
-        return output
-      }
-      const save = powershell ? powerShellSourcePrefix(tokens, item.words) : bashSourcePrefix(tokens, item.words)
-      output.commands.push({
-        resource: powershell ? item.resource : bashResource(item.resource, before),
-        save: `${save} *`,
-      })
-      return output
-    },
-    {
-      commands: [] as Array<{ resource: string; save: string }>,
-      directories: [] as string[],
-      cursor: 0,
-    },
-  )
-  return { commands: parsed.commands, directories: parsed.directories }
-}
-
-function bashResource(resource: string, before: string) {
-  if (!/(?:&&|\|\||\|&)\s*$|\|\s*$/.test(before)) return resource
-  const redirect = bashRedirect(resource)
-  return redirect < 0 ? resource : resource.slice(0, redirect).replace(/\d+$/, "").trim()
-}
-
-function bashRedirect(resource: string) {
-  let quote: "single" | "double" | undefined
-  for (let index = 0; index < resource.length; index++) {
-    const char = resource[index]
-    if (quote === "single") {
-      if (char === "'") quote = undefined
       continue
     }
-    if (char === "\\") {
-      index++
-      continue
-    }
-    if (char === '"') {
-      quote = quote === "double" ? undefined : "double"
-      continue
-    }
-    if (quote === "double") {
-      if (char === "$" && resource[index + 1] === "(") index = bashParenthesizedEnd(resource, index + 1)
-      else if (char === "`") index = bashBacktickEnd(resource, index)
-      continue
-    }
-    if (char === "'") {
-      quote = "single"
-      continue
-    }
-    if ((char === "$" || char === "<" || char === ">") && resource[index + 1] === "(") {
-      index = bashParenthesizedEnd(resource, index + 1)
-      continue
-    }
-    if (char === "`") {
-      index = bashBacktickEnd(resource, index)
-      continue
-    }
-    if (char === "<" || char === ">" || (char === "&" && resource[index + 1] === ">")) return index
+    const selected = prefix(words.slice(0, PREFIX_LENGTH))
+    const conventional = `${selected.join(" ")} *`
+    const end = item.wordEnds?.[selected.length - 1]
+    // Keep existing grants stable unless normalized spacing loses the original source boundary.
+    const save =
+      !powershell || end === undefined || Wildcard.match(item.resource, conventional)
+        ? conventional
+        : (() => {
+            const boundary =
+              item.wordEnds?.find(
+                (value) => value >= end && (value >= item.resource.length || /\s/.test(item.resource[value])),
+              ) ?? end
+            const separator = /^\s+(?:`(?:\r\n|\r|\n)\s*)?/.exec(item.resource.slice(boundary))?.[0]
+            return `${item.resource.slice(0, boundary)}${separator ?? " "}*`
+          })()
+    output.commands.push({
+      resource: item.resource,
+      save,
+    })
   }
-  return -1
-}
-
-function bashParenthesizedEnd(resource: string, start: number) {
-  let level = 1
-  let quote: "single" | "double" | undefined
-  for (let index = start + 1; index < resource.length; index++) {
-    const char = resource[index]
-    if (quote === "single") {
-      if (char === "'") quote = undefined
-      continue
-    }
-    if (char === "\\") {
-      index++
-      continue
-    }
-    if (char === '"') {
-      quote = quote === "double" ? undefined : "double"
-      continue
-    }
-    if (quote === "double") continue
-    if (char === "'") {
-      quote = "single"
-      continue
-    }
-    if (char === "(") level++
-    if (char === ")" && --level === 0) return index
-  }
-  return resource.length - 1
-}
-
-function bashBacktickEnd(resource: string, start: number) {
-  for (let index = start + 1; index < resource.length; index++) {
-    if (resource[index] === "\\") index++
-    else if (resource[index] === "`") return index
-  }
-  return resource.length - 1
-}
-
-function portableDirectoryArgs(
-  command: string[],
-  tokens: SourceToken[],
-  powershell: boolean,
-  cwd: string,
-  shell: string,
-) {
-  if (!powershell) {
-    const start = tokens.findIndex((token) => token.value === command[0])
-    if (start < 0) return []
-    return directoryArgs(
-      tokens.slice(start).map((token) => ({ type: "word", text: token.raw })),
-      false,
-      cwd,
-      shell,
-    )
-  }
-
-  const start = tokens.findIndex((token) => token.value.toLowerCase() === command[0]?.toLowerCase())
-  if (start < 0) return []
-  const directories: string[] = []
-  let expectsPath = false
-  for (const part of tokens.slice(start + 1).map((token) => token.raw)) {
-    if (expectsPath) {
-      const value = directoryArgument(part, true, cwd, shell)
-      if (value) directories.push(value)
-      expectsPath = false
-      continue
-    }
-    if (part.startsWith("-")) {
-      expectsPath = POWERSHELL_PATH_FLAGS.has(part.toLowerCase())
-      continue
-    }
-    const value = directoryArgument(part, true, cwd, shell)
-    if (value) directories.push(value)
-  }
-  return directories
-}
-
-function sourceTokens(resource: string) {
-  const tokens: SourceToken[] = []
-  let raw = ""
-  let value = ""
-  let quote: "single" | "double" | "backtick" | undefined
-  let substitution = 0
-  let redirect = false
-
-  const finish = () => {
-    if (!raw) return
-    if (!redirect) tokens.push({ raw, value })
-    raw = ""
-    value = ""
-    redirect = false
-  }
-
-  for (let index = 0; index < resource.length; index++) {
-    const char = resource[index]
-    if (quote === "single") {
-      raw += char
-      if (char === "'") quote = undefined
-      else value += char
-      continue
-    }
-    if (quote === "double") {
-      raw += char
-      if (char === '"') quote = undefined
-      else if (char === "\\" && index + 1 < resource.length) {
-        const next = resource[index + 1]
-        if ('$`"\\\n'.includes(next)) {
-          raw += resource[++index]
-          if (next !== "\n") value += next
-        } else value += char
-      } else value += char
-      continue
-    }
-    if (quote === "backtick") {
-      raw += char
-      value += char
-      if (char === "`" && resource[index - 1] !== "\\") quote = undefined
-      continue
-    }
-    if (char === "'") {
-      raw += char
-      quote = "single"
-      continue
-    }
-    if (char === '"') {
-      raw += char
-      quote = "double"
-      continue
-    }
-    if (char === "`") {
-      raw += char
-      value += char
-      quote = "backtick"
-      continue
-    }
-    if (char === "\\" && index + 1 < resource.length) {
-      if (resource[index + 1] === "\n") {
-        finish()
-        index++
-        continue
-      }
-      if (!raw && /\s/.test(resource[index + 1])) {
-        index++
-        continue
-      }
-      raw += char + resource[++index]
-      value += resource[index]
-      continue
-    }
-    if ((char === "<" || char === ">") && resource[index + 1] === "(") {
-      const end = bashParenthesizedEnd(resource, index + 1)
-      if (raw) {
-        raw += resource.slice(index, end + 1)
-        value += resource.slice(index, end + 1)
-      }
-      index = end
-      continue
-    }
-    if (char === "$" && resource[index + 1] === "(") substitution++
-    if (char === ")" && substitution > 0) substitution--
-    if (substitution === 0 && /\s/.test(char)) {
-      finish()
-      continue
-    }
-    if (substitution === 0 && (char === "<" || char === ">" || (char === "&" && resource[index + 1] === ">"))) {
-      if (/^\d+$/.test(value)) {
-        raw = ""
-        value = ""
-      } else finish()
-      redirect = true
-      if (char === "&") index++
-      while (/[<>&|]/.test(resource[index + 1] ?? "")) index++
-      continue
-    }
-    raw += char
-    value += char
-  }
-  finish()
-
-  return tokens
-}
-
-function bashSourcePrefix(tokens: SourceToken[], words: string[]) {
-  const start = tokens.findIndex((token) => token.value === words[0])
-  if (start < 0) {
-    const command = tokens.findIndex((token) => !/^[A-Za-z_][A-Za-z0-9_]*\+?=/.test(token.raw))
-    return prefix(tokens.slice(Math.max(0, command)).map((token) => token.raw)).join(" ")
-  }
-  const source = tokens
-    .slice(start)
-    .map((token) => token.raw)
-    .filter((token) => !/^\$\([\s\S]*\)$/.test(token) && !/^`[\s\S]*`$/.test(token))
-  return prefix(source).join(" ")
-}
-
-function powerShellSourcePrefix(tokens: SourceToken[], words: string[]) {
-  const start = tokens.findIndex((token) => token.value.toLowerCase() === words[0]?.toLowerCase())
-  if (start < 0) return prefix(words).join(" ")
-  return prefix(tokens.slice(start).map((token) => token.raw)).join(" ")
-}
-
-function powerShellSourceTokens(resource: string) {
-  const tokens: SourceToken[] = []
-  let raw = ""
-  let value = ""
-  let quote: "single" | "double" | undefined
-  let redirect = false
-
-  const finish = () => {
-    if (!raw) return
-    if (!redirect) tokens.push({ raw, value })
-    raw = ""
-    value = ""
-    redirect = false
-  }
-
-  for (let index = 0; index < resource.length; index++) {
-    const char = resource[index]
-    if (quote === "single") {
-      raw += char
-      if (char === "'" && resource[index + 1] === "'") {
-        raw += resource[++index]
-        value += "'"
-      } else if (char === "'") quote = undefined
-      else value += char
-      continue
-    }
-    if (quote === "double") {
-      raw += char
-      if (char === '"') quote = undefined
-      else if (char === "`" && index + 1 < resource.length) {
-        raw += resource[++index]
-        value += resource[index]
-      } else value += char
-      continue
-    }
-    if (char === "'") {
-      raw += char
-      quote = "single"
-      continue
-    }
-    if (char === '"') {
-      raw += char
-      quote = "double"
-      continue
-    }
-    if (char === "`" && index + 1 < resource.length) {
-      raw += char + resource[++index]
-      if (resource[index] !== "\n" && resource[index] !== "\r") value += resource[index]
-      continue
-    }
-    if (char === "{" && !raw) {
-      const end = powerShellBracedEnd(resource, index)
-      raw = resource.slice(index, end + 1)
-      value = raw
-      index = end
-      continue
-    }
-    if (/\s/.test(char)) {
-      finish()
-      continue
-    }
-    if (char === ">") {
-      if (resource[index + 1] && !/[\s>&]/.test(resource[index + 1])) {
-        raw += char
-        value += char
-        continue
-      }
-      if (/^\d+$/.test(value)) {
-        raw = ""
-        value = ""
-      } else if (raw === "*") {
-        raw = ""
-        value = ""
-      } else finish()
-      redirect = true
-      while (/[>&\d]/.test(resource[index + 1] ?? "")) index++
-      continue
-    }
-    if ((char === "&" || char === ".") && !raw && tokens.length === 0) continue
-    raw += char
-    value += char
-  }
-  finish()
-  return tokens
-}
-
-function powerShellBracedEnd(resource: string, start: number) {
-  let level = 1
-  let quote: "single" | "double" | undefined
-  for (let index = start + 1; index < resource.length; index++) {
-    const char = resource[index]
-    if (char === "`" && quote !== "single") {
-      index++
-      continue
-    }
-    if (quote === "single") {
-      if (char === "'" && resource[index + 1] === "'") index++
-      else if (char === "'") quote = undefined
-      continue
-    }
-    if (quote === "double") {
-      if (char === '"') quote = undefined
-      continue
-    }
-    if (char === "'") {
-      quote = "single"
-      continue
-    }
-    if (char === '"') {
-      quote = "double"
-      continue
-    }
-    if (char === "{") level++
-    if (char === "}" && --level === 0) return index
-  }
-  return resource.length - 1
-}
+  return output
+})
 
 function parts(node: Node) {
   return Array.from({ length: node.childCount }).flatMap((_, index): Part[] => {

packages/core/src/tool.ts

@@ -4,7 +4,8 @@ export type { Context, Metadata, Options, Result } from "@opencode-ai/schema/too
 
 import { ToolDefinition, type ToolCall } from "@opencode-ai/ai"
 import { Tool } from "@opencode-ai/schema/tool"
-import { Context, Effect, Layer, Schema, SchemaIssue, Scope, Semaphore } from "effect"
+import type { ToolDraft } from "@opencode-ai/plugin/effect/tool"
+import { Context, Effect, Layer, Result, Schema, SchemaIssue, Types } from "effect"
 import { makeLocationNode } from "@opencode-ai/util/effect/app-node"
 import type { Agent } from "./agent.js"
 import { CodeModeCatalog } from "./codemode/catalog.js"
@@ -14,6 +15,7 @@ import { Permission } from "./permission.js"
 import { PluginHooks } from "./plugin/hooks.js"
 import { SessionMessage } from "./session/message.js"
 import { SessionSchema } from "./session/schema.js"
+import { State } from "./state.js"
 import { definition, execute, normalizeContent } from "./tool/runtime.js"
 import { Wildcard } from "./util/wildcard.js"
 
@@ -22,10 +24,7 @@ export class RegistrationError extends Schema.TaggedError<RegistrationError>()("
   message: Schema.String,
 }) {}
 
-export interface Interface {
-  readonly transform: (
-    callback: (draft: { readonly add: (tool: Tool.Info) => void }) => void,
-  ) => Effect.Effect<void, RegistrationError, Scope.Scope>
+export interface Interface extends State.Transformable<ToolDraft> {
   readonly snapshot: (permissions?: Permission.Ruleset) => Effect.Effect<Snapshot>
 }
 
@@ -79,9 +78,6 @@ const layer = Layer.effect(
       ]
     })
 
-    const local = new Map<string, Array<{ readonly token: object; readonly tool: Tool.Info }>>()
-    const lock = Semaphore.makeUnsafe(1)
-
     const executeTool = Effect.fn("Tool.execute")(function* (
       tool: Tool.Info,
       name: string,
@@ -137,122 +133,123 @@ const layer = Layer.effect(
       }
     })
 
-    const transform: Interface["transform"] = Effect.fn("Tool.transform")(function* (callback) {
-      const tools: Array<Tool.Info> = []
-      yield* Effect.sync(() => callback({ add: (tool) => tools.push(tool) }))
-      yield* Effect.forEach(
-        tools.flatMap((tool) => (tool.options?.namespace === undefined ? [] : [tool.options.namespace])),
-        validateNamespace,
-        { discard: true },
-      )
-      const entries = normalizedEntries(tools)
-      yield* Effect.forEach(entries, (entry) => validateName(normalizedName(entry.tool)), { discard: true })
-      const collision = entries.find(
-        (entry, index) => entries.findIndex((candidate) => candidate.key === entry.key) !== index,
-      )
-      if (collision)
-        return yield* Effect.fail(
-          new RegistrationError({
-            name: collision.key,
-            message: `Duplicate normalized tool name: ${collision.key}`,
-          }),
-        )
-      const reserved = entries.find((entry) => entry.tool.options?.codemode === false && entry.key === "execute")
-      if (reserved)
-        return yield* Effect.fail(
-          new RegistrationError({
-            name: reserved.key,
-            message: 'Tool name "execute" is reserved for CodeMode',
-          }),
-        )
-      if (entries.length === 0) return
-      yield* Effect.forEach(
-        entries,
-        (entry) =>
-          Effect.try({
-            try: () => ToolDefinition.make(definition(entry.tool)),
-            catch: (error) =>
-              new RegistrationError({
-                name: entry.key,
-                message: `Invalid tool definition ${entry.key}: ${schemaMakeError(error)}`,
-              }),
-          }),
-        { discard: true },
-      )
-      yield* Effect.uninterruptible(
-        lock.withPermit(
-          Effect.gen(function* () {
-            const token = {}
-            for (const entry of entries)
-              local.set(entry.key, [...(local.get(entry.key) ?? []), { token, tool: entry.tool }])
-            yield* Effect.addFinalizer(() =>
-              lock.withPermit(
-                Effect.sync(() => {
-                  for (const entry of entries) {
-                    const remaining = local.get(entry.key)?.filter((item) => item.token !== token) ?? []
-                    if (remaining.length > 0) local.set(entry.key, remaining)
-                    else local.delete(entry.key)
-                  }
-                }),
-              ),
-            )
-          }),
+    const state = State.create({
+      name: "tool",
+      initial: () => ({
+        tools: new Map<string, Types.Mutable<Tool.Info>>(),
+        errors: new Array<{ tool: Tool.Info; error: RegistrationError }>(),
+      }),
+      draft: (data) => data,
+      finalize: (draft) =>
+        Effect.forEach(
+          draft.errors,
+          (entry) =>
+            Effect.logError("Skipping invalid tool registration", {
+              name: entry.tool.name,
+              namespace: entry.tool.options?.namespace,
+              error: entry.error.message,
+            }),
+          { discard: true },
         ),
-      )
     })
 
     return Service.of({
-      transform,
-      snapshot: Effect.fn("Tool.snapshot")((permissions) =>
-        lock.withPermit(
-          Effect.gen(function* () {
-            const active = new Map<string, Tool.Info>()
-            const rules = permissions ?? []
-            for (const [name, entries] of local) {
-              const tool = entries.at(-1)?.tool
-              if (!tool) continue
-              if (whollyDisabled(tool.options?.permission ?? name, rules)) continue
-              active.set(name, tool)
-            }
-            const direct = new Map(Array.from(active).filter(([, tool]) => tool.options?.codemode === false))
-            const codemode = new Map(Array.from(active).filter(([, tool]) => tool.options?.codemode !== false))
-            const executeRule = rules.findLast((rule) => Wildcard.match("execute", rule.action))
-            const codemodeEnabled = executeRule?.resource !== "*" || executeRule.effect !== "deny"
-            const codemodeTool = codemodeEnabled
-              ? CodeModeTool.create(codemode, (name, tool, input, context) => executeTool(tool, name, input, context))
-              : undefined
-            const codeModeCatalog = codemodeEnabled ? CodeModeTool.catalog(codemode) : undefined
-            return {
-              ...(codeModeCatalog === undefined ? {} : { codeModeCatalog }),
-              definitions: [
-                ...Array.from(direct)
-                  .sort(([left], [right]) => (left < right ? -1 : left > right ? 1 : 0))
-                  .map(([, tool]) => definition(tool)),
-                ...(codemodeTool ? [definition(codemodeTool)] : []),
-              ],
-              execute: (input: {
-                readonly sessionID: SessionSchema.ID
-                readonly agent: Agent.ID
-                readonly messageID: SessionMessage.ID
-                readonly call: ToolCall
-                readonly progress?: (update: Tool.Metadata) => Effect.Effect<void>
-              }) => {
-                const context: Tool.Context = {
-                  sessionID: input.sessionID,
-                  agent: input.agent,
-                  messageID: input.messageID,
-                  id: Tool.CallID.make(input.call.id),
-                  progress: input.progress ?? (() => Effect.void),
+      transform: (callback) =>
+        state.transform((draft) => {
+          // Preserve rejection of ambiguous adds within one transform, without rejecting later overrides.
+          const added = new Map<string, Tool.Info | undefined>()
+          callback({
+            add: (tool) => {
+              const error = registrationError(tool)
+              if (error) {
+                draft.errors.push({ tool, error })
+                return
+              }
+              const id = effectiveName(tool)
+              if (added.has(id)) {
+                draft.errors.push({
+                  tool,
+                  error: new RegistrationError({ name: id, message: `Duplicate normalized tool name: ${id}` }),
+                })
+                const previous = added.get(id)
+                if (previous) {
+                  draft.tools.set(id, previous)
+                  return
                 }
-                if (input.call.name === "execute" && codemodeTool)
-                  return executeTool(codemodeTool, input.call.name, input.call.input, context)
-                const tool = direct.get(input.call.name)
-                if (tool) return executeTool(tool, input.call.name, input.call.input, context)
-                return new Tool.Error({ message: `Unknown tool: ${input.call.name}` })
-              },
-            }
-          }),
-        ),
+                draft.tools.delete(id)
+                return
+              }
+              added.set(id, draft.tools.get(id))
+              draft.tools.set(id, { ...tool })
+            },
+            update: (id, update) => {
+              const current = draft.tools.get(id)
+              if (!current) return
+              const tool = { ...current }
+              update(tool)
+              tool.name = current.name
+              if (tool.options?.namespace !== current.options?.namespace)
+                tool.options = { ...tool.options, namespace: current.options?.namespace }
+              const error = registrationError(tool)
+              if (error) {
+                draft.errors.push({ tool, error })
+                return
+              }
+              draft.tools.set(id, tool)
+            },
+            remove: (id) => {
+              draft.tools.delete(id)
+              added.delete(id)
+            },
+          })
+        }),
+      reload: state.reload,
+      snapshot: Effect.fn("Tool.snapshot")((permissions) =>
+        Effect.sync(() => {
+          const active = new Map<string, Tool.Info>()
+          const rules = permissions ?? []
+          for (const [name, tool] of state.get().tools) {
+            if (whollyDisabled(tool.options?.permission ?? name, rules)) continue
+            active.set(name, tool)
+          }
+          const direct = new Map(Array.from(active).filter(([, tool]) => tool.options?.codemode === false))
+          const codemode = new Map(Array.from(active).filter(([, tool]) => tool.options?.codemode !== false))
+          const executeRule = rules.findLast((rule) => Wildcard.match("execute", rule.action))
+          const codemodeEnabled = executeRule?.resource !== "*" || executeRule.effect !== "deny"
+          const codemodeTool = codemodeEnabled
+            ? CodeModeTool.create(codemode, (name, tool, input, context) => executeTool(tool, name, input, context))
+            : undefined
+          const codeModeCatalog = codemodeEnabled ? CodeModeTool.catalog(codemode) : undefined
+          return {
+            ...(codeModeCatalog === undefined ? {} : { codeModeCatalog }),
+            definitions: [
+              ...Array.from(direct)
+                .sort(([left], [right]) => (left < right ? -1 : left > right ? 1 : 0))
+                .map(([, tool]) => definition(tool)),
+              ...(codemodeTool ? [definition(codemodeTool)] : []),
+            ],
+            execute: (input: {
+              readonly sessionID: SessionSchema.ID
+              readonly agent: Agent.ID
+              readonly messageID: SessionMessage.ID
+              readonly call: ToolCall
+              readonly progress?: (update: Tool.Metadata) => Effect.Effect<void>
+            }) => {
+              const context: Tool.Context = {
+                sessionID: input.sessionID,
+                agent: input.agent,
+                messageID: input.messageID,
+                id: Tool.CallID.make(input.call.id),
+                progress: input.progress ?? (() => Effect.void),
+              }
+              if (input.call.name === "execute" && codemodeTool)
+                return executeTool(codemodeTool, input.call.name, input.call.input, context)
+              const tool = direct.get(input.call.name)
+              if (tool) return executeTool(tool, input.call.name, input.call.input, context)
+              return new Tool.Error({ message: `Unknown tool: ${input.call.name}` })
+            },
+          }
+        }),
       ),
     })
   }),
@@ -270,20 +267,22 @@ function schemaMakeError(error: unknown) {
   return error instanceof Error ? error.message : String(error)
 }
 
-const validateName = (name: string) =>
-  /^[A-Za-z0-9_-]{1,64}$/.test(name)
-    ? Effect.void
-    : Effect.fail(new RegistrationError({ name, message: `Invalid tool name: ${name}` }))
-
-const validateNamespace = (namespace: string) =>
-  namespace.split(".").every((segment) => /^[A-Za-z0-9_-]{1,64}$/.test(segment))
-    ? Effect.void
-    : Effect.fail(
-        new RegistrationError({
-          name: namespace,
-          message: `Invalid tool namespace: ${JSON.stringify(namespace)}`,
-        }),
-      )
+function registrationError(tool: Tool.Info) {
+  const namespace = tool.options?.namespace
+  if (namespace !== undefined && !namespace.split(".").every((segment) => /^[A-Za-z0-9_-]{1,64}$/.test(segment)))
+    return new RegistrationError({ name: namespace, message: `Invalid tool namespace: ${JSON.stringify(namespace)}` })
+  const name = normalizedName(tool)
+  if (!/^[A-Za-z0-9_-]{1,64}$/.test(name)) return new RegistrationError({ name, message: `Invalid tool name: ${name}` })
+  const id = effectiveName(tool)
+  if (tool.options?.codemode === false && id === "execute")
+    return new RegistrationError({ name: id, message: 'Tool name "execute" is reserved for CodeMode' })
+  const result = Result.try({
+    try: () => ToolDefinition.make(definition(tool)),
+    catch: (error) =>
+      new RegistrationError({ name: id, message: `Invalid tool definition ${id}: ${schemaMakeError(error)}` }),
+  })
+  return Result.isFailure(result) ? result.failure : undefined
+}
 
 const normalizedName = (tool: Tool.Info) => tool.name.replace(/[^a-zA-Z0-9_-]/g, "_")
 
@@ -292,12 +291,6 @@ const effectiveName = (tool: Tool.Info) =>
     ? normalizedName(tool)
     : `${tool.options.namespace.replaceAll(".", "_")}_${normalizedName(tool)}`
 
-const normalizedEntries = (tools: ReadonlyArray<Tool.Info>) =>
-  tools.map((tool) => ({
-    key: effectiveName(tool),
-    tool,
-  }))
-
 export const node = makeLocationNode({
   service: Service,
   layer,

packages/core/src/tool/AGENTS.md

@@ -30,17 +30,18 @@ Leaves own resolution, permission, and side-effect ordering. Translate only expe
 
 ## Registration
 
-Built-ins, plugins, and MCP install tools through `ToolRegistry.Service.transform`, adding complete tool objects to the draft. A tool may provide a namespace, which flattens direct model names to `<namespace>_<tool>`, and defaults into CodeMode (`codemode` defaults true; `codemode: false` keeps the tool on the provider's native tool list).
+Built-ins, plugins, and MCP install tools through `Tool.Service.transform`, adding complete tool objects to the draft. A tool may provide a namespace, which flattens direct model names to `<namespace>_<tool>`, and defaults into CodeMode (`codemode` defaults true; `codemode: false` keeps the tool on the provider's native tool list).
 
 Registrations are scoped:
 
-- The latest active same-placement registration wins.
-- Closing any registration removes only that registration and reveals the next active one.
+- Tool transforms use the shared `State.create` lifecycle, like agents and skills: `add`, `update`, and `remove` replay in registration order when state is rebuilt.
+- `update` and `remove` do nothing for missing tools. `add` requires a complete tool definition.
+- Disposing a registration or closing its scope removes its transform and rebuilds the remaining state. `reload` replays transforms after their external inputs change.
 - Each model request captures the effective tools it advertises; later registration changes affect later requests.
 
 Type safety ends at registration. The registry validates model input and declared output at runtime and should not carry producer schema generics through storage or execution.
 
-`ToolRegistry.Service` is Location-scoped. Do not make the registry process-global or construct a separate application-tool service for each Location.
+`Tool.Service` is Location-scoped. Do not make the registry process-global or construct a separate application-tool service for each Location.
 
 ## Permissions
 
@@ -56,4 +57,4 @@ Producer capture limits remain local to producers. For example, Bash keeps `AppP
 
 ## Current Gaps
 
-- MCP and future Session-scoped registrations still need an explicit canonical registration design.
+- Future Session-scoped registrations still need an explicit canonical registration design.

packages/core/src/tool/mcp.ts

@@ -2,7 +2,7 @@ export * as McpTool from "./mcp.js"
 
 import { ToolFailure } from "@opencode-ai/ai"
 import { McpEvent } from "@opencode-ai/schema/mcp-event"
-import { Context, Effect, Exit, Fiber, type JsonSchema, Layer, Scope, Semaphore, Stream } from "effect"
+import { Context, Effect, Fiber, type JsonSchema, Layer, Semaphore, Stream } from "effect"
 import { makeLocationNode } from "@opencode-ai/util/effect/app-node"
 import { Bus } from "../bus.js"
 
@@ -30,94 +30,88 @@ export const layer = Layer.effect(
     const tools = yield* Tool.Service
     const bus = yield* Bus.Service
     const permission = yield* Permission.Service
-    const scope = yield* Scope.Scope
     const lock = Semaphore.makeUnsafe(1)
-    let current: Scope.Closeable | undefined
+    let discovered: MCP.Tool[] = []
 
-    // Register the current tool set under a fresh child scope, then close the previous one so the
-    // registry never has a gap where MCP tools disappear mid-swap.
+    // Keep the source's position so later plugin transforms also apply after MCP refreshes.
+    yield* tools.transform((draft) => {
+      for (const tool of discovered) {
+        const schema = (tool.inputSchema ?? {}) as JsonSchema.JsonSchema
+        draft.add({
+          name: tool.name,
+          options: { namespace: namespace(tool.server), codemode: tool.codemode !== false },
+          description: tool.description ?? "",
+          input: {
+            ...schema,
+            type: "object",
+            properties: schema.properties ?? {},
+            additionalProperties: false,
+          },
+          output: (tool.outputSchema ?? {}) as JsonSchema.JsonSchema,
+          execute: (input, context) =>
+            Effect.gen(function* () {
+              yield* permission.assert({
+                action: name(tool.server, tool.name),
+                resources: ["*"],
+                save: ["*"],
+                metadata: {},
+                sessionID: context.sessionID,
+                agent: context.agent,
+                source: {
+                  type: "tool",
+                  messageID: context.messageID,
+                  id: context.id,
+                },
+              })
+              const result = yield* mcp
+                .callTool({
+                  server: tool.server,
+                  name: tool.name,
+                  args: (input ?? {}) as Record<string, unknown>,
+                })
+                .pipe(
+                  Effect.catchTags({
+                    "MCP.NotFoundError": (error) =>
+                      new ToolFailure({ message: `MCP server "${error.server}" is not available` }),
+                    "MCP.ToolCallError": (error) => new ToolFailure({ message: error.message }),
+                  }),
+                )
+              if (result.isError)
+                return yield* new ToolFailure({
+                  message:
+                    result.content
+                      .flatMap((part) => (part.type === "text" ? [part.text] : []))
+                      .join("\n")
+                      .trim() || "MCP tool returned an error",
+                })
+              const content = result.content.map((part) =>
+                part.type === "text"
+                  ? { type: "text" as const, text: part.text }
+                  : {
+                      type: "file" as const,
+                      uri: `data:${part.mimeType};base64,${part.data}`,
+                      mime: part.mimeType,
+                    },
+              )
+              const text = content.flatMap((part) => (part.type === "text" ? [part.text] : [])).join("\n")
+              return {
+                output: result.structured ?? (text === "" ? null : text),
+                ...(content.length === 0 ? {} : { content }),
+              }
+            }).pipe(
+              Effect.mapError((error) =>
+                error instanceof ToolFailure
+                  ? error
+                  : new ToolFailure({ message: `Unable to execute ${name(tool.server, tool.name)}` }),
+              ),
+            ),
+        })
+      }
+    })
     const reconcile = lock.withPermit(
       Effect.gen(function* () {
-        const discovered = yield* mcp.tools()
-        const next = yield* Scope.fork(scope)
-        yield* tools
-          .transform((draft) => {
-            for (const tool of discovered) {
-              const schema = (tool.inputSchema ?? {}) as JsonSchema.JsonSchema
-              draft.add({
-                name: tool.name,
-                options: { namespace: namespace(tool.server), codemode: tool.codemode !== false },
-                description: tool.description ?? "",
-                input: {
-                  ...schema,
-                  type: "object",
-                  properties: schema.properties ?? {},
-                  additionalProperties: false,
-                },
-                output: (tool.outputSchema ?? {}) as JsonSchema.JsonSchema,
-                execute: (input, context) =>
-                  Effect.gen(function* () {
-                    yield* permission.assert({
-                      action: name(tool.server, tool.name),
-                      resources: ["*"],
-                      save: ["*"],
-                      metadata: {},
-                      sessionID: context.sessionID,
-                      agent: context.agent,
-                      source: {
-                        type: "tool",
-                        messageID: context.messageID,
-                        id: context.id,
-                      },
-                    })
-                    const result = yield* mcp
-                      .callTool({
-                        server: tool.server,
-                        name: tool.name,
-                        args: (input ?? {}) as Record<string, unknown>,
-                      })
-                      .pipe(
-                        Effect.catchTags({
-                          "MCP.NotFoundError": (error) =>
-                            new ToolFailure({ message: `MCP server "${error.server}" is not available` }),
-                          "MCP.ToolCallError": (error) => new ToolFailure({ message: error.message }),
-                        }),
-                      )
-                    if (result.isError)
-                      return yield* new ToolFailure({
-                        message:
-                          result.content
-                            .flatMap((part) => (part.type === "text" ? [part.text] : []))
-                            .join("\n")
-                            .trim() || "MCP tool returned an error",
-                      })
-                    const content = result.content.map((part) =>
-                      part.type === "text"
-                        ? { type: "text" as const, text: part.text }
-                        : {
-                            type: "file" as const,
-                            uri: `data:${part.mimeType};base64,${part.data}`,
-                            mime: part.mimeType,
-                          },
-                    )
-                    const text = content.flatMap((part) => (part.type === "text" ? [part.text] : [])).join("\n")
-                    return {
-                      output: result.structured ?? (text === "" ? null : text),
-                      ...(content.length === 0 ? {} : { content }),
-                    }
-                  }).pipe(
-                    Effect.mapError((error) =>
-                      error instanceof ToolFailure
-                        ? error
-                        : new ToolFailure({ message: `Unable to execute ${name(tool.server, tool.name)}` }),
-                    ),
-                  ),
-              })
-            }
-          })
-          .pipe(Scope.provide(next), Effect.orDie)
-        if (current) yield* Scope.close(current, Exit.void)
-        current = next
+        discovered = yield* mcp.tools()
+        yield* tools.reload()
       }),
     )
 

packages/core/src/tool/plugin/shell.ts

@@ -115,56 +115,45 @@ export const Plugin = {
     const permission = yield* Permission.Service
     const config = yield* Config.Service
 
-    const notifyWhenDone = Effect.fn("ShellTool.notifyWhenDone")(function* (
-      sessionID: SessionSchema.ID,
-      id: string,
-      shellID: string,
-      command: string,
-      settled: Deferred.Deferred<Output>,
-    ) {
-      yield* runtime.job.wait({ id: id }).pipe(
-        Effect.flatMap((result) =>
-          Effect.gen(function* () {
-            const info = result.info
-            if (!info) return
-            const state =
-              info.status === "completed"
-                ? "completed"
-                : info.status === "error"
-                  ? "error"
-                  : info.status === "cancelled"
-                    ? "cancelled"
-                    : undefined
-            if (state === undefined) return
-            const output = state === "completed" ? yield* Deferred.await(settled) : undefined
-            const text = output
-              ? resultMessages(output).join("\n\n")
-              : state === "error"
-                ? (info.error ?? "Command failed")
-                : "Command cancelled"
-            yield* runtime.session.synthetic({
-              sessionID,
-              text: `<shell id="${id}" state="${state}" command="${command}">\n${text}\n</shell>`,
-              description: command,
-              metadata: {
-                source: "shell",
-                jobID: id,
-                shellID,
-                state,
-                ...(output
-                  ? {
-                      truncated: output.truncated,
-                      ...(output.exit !== undefined ? { exit: output.exit } : {}),
-                      ...(output.timeout !== undefined ? { timeout: output.timeout } : {}),
-                    }
-                  : {}),
-              },
-            })
-          }),
-        ),
-        Effect.forkIn(scope, { startImmediately: true }),
-      )
-    })
+    const notifyWhenDone = Effect.fn("ShellTool.notifyWhenDone")(
+      function* (
+        sessionID: SessionSchema.ID,
+        id: string,
+        shellID: string,
+        command: string,
+        settled: Deferred.Deferred<Output>,
+      ) {
+        const info = (yield* runtime.job.wait({ id })).info
+        if (!info || info.status === "running") return
+        const output = info.status === "completed" ? yield* Deferred.await(settled) : undefined
+        const text = output
+          ? resultMessages(output).join("\n\n")
+          : info.status === "error"
+            ? (info.error ?? "Command failed")
+            : "Command cancelled"
+        yield* runtime.session.synthetic({
+          ...(info.notificationID ? { id: info.notificationID } : {}),
+          sessionID,
+          text: `<shell id="${id}" state="${info.status}" command="${command}">\n${text}\n</shell>`,
+          description: command,
+          metadata: {
+            source: "shell",
+            jobID: id,
+            shellID,
+            state: info.status,
+            ...(output
+              ? {
+                  truncated: output.truncated,
+                  ...(output.exit !== undefined ? { exit: output.exit } : {}),
+                  ...(output.timeout !== undefined ? { timeout: output.timeout } : {}),
+                }
+              : {}),
+          },
+        })
+        if (info.notificationID) yield* runtime.job.completeBackground(info.notificationID)
+      },
+      Effect.forkIn(scope, { startImmediately: true }),
+    )
 
     yield* ctx.tool
       .transform((draft) =>
@@ -286,7 +275,7 @@ export const Plugin = {
               const settled = yield* Deferred.make<Output>()
               const run = settleShell().pipe(
                 Effect.tap((output) => Deferred.succeed(settled, output)),
-                Effect.map((output) => output.output),
+                Effect.map((output) => resultMessages(output).join("\n\n")),
                 Effect.onInterrupt(() => shell.remove(info.id).pipe(Effect.ignore)),
               )
               const job = yield* runtime.job.start({
@@ -294,6 +283,12 @@ export const Plugin = {
                 type: name,
                 title: info.command,
                 metadata: { sessionID: context.sessionID, shellID: info.id },
+                recovery: {
+                  kind: "shell",
+                  sessionID: context.sessionID,
+                  shellID: info.id,
+                  command: info.command,
+                },
                 run,
               })
 

packages/core/src/tool/plugin/subagent.ts

@@ -78,22 +78,6 @@ export const Plugin = {
       return text.length > 0 ? text : NO_TEXT
     })
 
-    const injectCompletion = Effect.fn("SubagentTool.injectCompletion")(function* (
-      parentID: SessionSchema.ID,
-      childID: SessionSchema.ID,
-      agent: string,
-      description: string,
-      state: "completed" | "error" | "cancelled",
-      text: string,
-    ) {
-      yield* runtime.session.synthetic({
-        sessionID: parentID,
-        text: `<subagent sessionID="${childID}" state="${state}" description="${description}">\n${text}\n</subagent>`,
-        description,
-        metadata: { source: "subagent", childID, agent, state },
-      })
-    })
-
     const notifyWhenDone = Effect.fn("SubagentTool.notifyWhenDone")(function* (
       parentID: SessionSchema.ID,
       childID: SessionSchema.ID,
@@ -104,23 +88,24 @@ export const Plugin = {
       const key = `${childID}:${startedAt}`
       if (notifications.has(key)) return
       notifications.add(key)
-      yield* runtime.job.wait({ id: childID }).pipe(
-        Effect.flatMap((result) => {
-          if (result.info?.status === "completed")
-            return injectCompletion(parentID, childID, agent, description, "completed", result.info.output ?? NO_TEXT)
-          if (result.info?.status === "error")
-            return injectCompletion(
-              parentID,
-              childID,
-              agent,
-              description,
-              "error",
-              result.info.error ?? "Subagent failed",
-            )
-          if (result.info?.status === "cancelled")
-            return injectCompletion(parentID, childID, agent, description, "cancelled", "Subagent cancelled")
-          return Effect.void
-        }),
+      yield* Effect.gen(function* () {
+        const info = (yield* runtime.job.wait({ id: childID })).info
+        if (!info || info.status === "running") return
+        const text =
+          info.status === "completed"
+            ? (info.output ?? NO_TEXT)
+            : info.status === "error"
+              ? (info.error ?? "Subagent failed")
+              : "Subagent cancelled"
+        yield* runtime.session.synthetic({
+          ...(info.notificationID ? { id: info.notificationID } : {}),
+          sessionID: parentID,
+          text: `<subagent sessionID="${childID}" state="${info.status}" description="${description}">\n${text}\n</subagent>`,
+          description,
+          metadata: { source: "subagent", childID, agent, state: info.status },
+        })
+        if (info.notificationID) yield* runtime.job.completeBackground(info.notificationID)
+      }).pipe(
         Effect.ensuring(Effect.sync(() => notifications.delete(key))),
         Effect.forkIn(scope, { startImmediately: true }),
       )
@@ -239,6 +224,7 @@ export const Plugin = {
                     existing === undefined
                       ? ["You are a subagent spawned by another session.", input.prompt].join("\n")
                       : input.prompt,
+                  ...(background && existing === undefined ? { resume: false } : {}),
                 })
                 .pipe(
                   Effect.mapError(
@@ -246,17 +232,19 @@ export const Plugin = {
                   ),
                 )
 
-              const run = Effect.gen(function* () {
-                yield* runtime.session.resume(child.id)
-                return yield* latestAssistantText(child.id)
-              }).pipe(Effect.onInterrupt(() => runtime.session.interrupt(child.id)))
-
               const info = yield* runtime.job.start({
                 id: child.id,
                 type: name,
                 title: input.description,
                 metadata: {},
-                run,
+                recovery: {
+                  kind: "subagent",
+                  parentSessionID: context.sessionID,
+                  childSessionID: child.id,
+                  agent: agent.name,
+                  description: input.description,
+                },
+                run: runtime.session.resume(child.id).pipe(Effect.andThen(latestAssistantText(child.id))),
               })
 
               if (background) {

packages/core/test/aisdk-native.test.ts

@@ -222,9 +222,11 @@ describe("AISDKNative", () => {
       },
       headers: { "x-test": "value" },
     })
-    expect(map("@ai-sdk/amazon-bedrock/mantle", settings, "openai.gpt-oss-safeguard-20b")?.package).toBe(
-      "@opencode-ai/ai/providers/amazon-bedrock/mantle/chat",
-    )
+    for (const modelID of ["openai.gpt-oss-safeguard-20b", "openai.gpt-oss-safeguard-120b"]) {
+      expect(map("@ai-sdk/amazon-bedrock/mantle", settings, modelID)?.package).toBe(
+        "@opencode-ai/ai/providers/amazon-bedrock/mantle/chat",
+      )
+    }
     expect(
       map(
         "@ai-sdk/amazon-bedrock/mantle",

packages/core/test/database-migration.test.ts

@@ -385,6 +385,9 @@ describe("DatabaseMigration", () => {
     const content = JSON.stringify({
       openai: { type: "oauth", refresh: "refresh", access: "access", expires: 123, accountId: "account" },
       anthropic: { type: "api", key: "legacy-key", metadata: { region: "us" } },
+      google: { type: "api", key: "google-key", metadata: { region: "us" } },
+      "github-copilot": { type: "oauth", refresh: "refresh", access: "access", expires: 123 },
+      "custom-provider": { type: "api", key: "custom-key" },
       "https://example.com/": { type: "wellknown", key: "TOKEN", token: "wellknown-key" },
       invalid: { type: "unknown" },
     })
@@ -402,6 +405,7 @@ describe("DatabaseMigration", () => {
 
         yield* db.run(sql`DELETE FROM migration WHERE id = ${legacyCredentialsMigration.id}`)
         yield* DatabaseMigration.applyOnly(db, [legacyCredentialsMigration])
+        yield* DatabaseMigration.applyOnly(db, [legacyCredentialsMigration])
 
         expect(yield* db.all(sql`SELECT integration_id, label, value FROM credential ORDER BY integration_id`)).toEqual(
           [
@@ -410,14 +414,35 @@ describe("DatabaseMigration", () => {
               label: "Existing",
               value: JSON.stringify({ type: "key", key: "current-key" }),
             },
+            {
+              integration_id: "custom-provider",
+              label: "API key",
+              value: JSON.stringify({ type: "key", key: "custom-key" }),
+            },
+            {
+              integration_id: "github-copilot",
+              label: "OAuth",
+              value: JSON.stringify({
+                type: "oauth",
+                methodID: "device",
+                refresh: "refresh",
+                access: "access",
+                expires: 123,
+              }),
+            },
+            {
+              integration_id: "google",
+              label: "API key",
+              value: JSON.stringify({ type: "key", key: "google-key", metadata: { region: "us" } }),
+            },
             {
               integration_id: "https://example.com",
-              label: "default",
+              label: "API key",
               value: JSON.stringify({ type: "key", key: "wellknown-key" }),
             },
             {
               integration_id: "openai",
-              label: "default",
+              label: "OAuth",
               value: JSON.stringify({
                 type: "oauth",
                 methodID: "chatgpt-browser",

packages/core/test/job.test.ts

@@ -1,11 +1,13 @@
 import { describe, expect } from "bun:test"
 import { Job } from "@opencode-ai/core/job"
+import { KV } from "@opencode-ai/core/kv"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
+import { LayerNode } from "@opencode-ai/util/effect/layer-node"
 import { Deferred, Effect, Exit, Fiber, Scope } from "effect"
 import { SessionSchema } from "@opencode-ai/core/session/schema"
 import { testEffect } from "./lib/effect"
 
-const it = testEffect(AppNodeBuilder.build(Job.node))
+const it = testEffect(AppNodeBuilder.build(LayerNode.group([Job.node, KV.node])))
 
 describe("Job", () => {
   it.live("tracks process-local work through explicit observation", () =>
@@ -145,6 +147,177 @@ describe("Job", () => {
     }),
   )
 
+  it.live("retains background ownership and terminal output until notification acknowledgment", () =>
+    Effect.gen(function* () {
+      const jobs = yield* Job.Service
+      const latch = yield* Deferred.make<void>()
+      const recovery = {
+        kind: "shell" as const,
+        sessionID: SessionSchema.ID.make("ses_background_shell"),
+        shellID: "shell_background",
+        command: "echo done",
+      }
+      const job = yield* jobs.start({ type: "shell", recovery, run: Deferred.await(latch).pipe(Effect.as("done")) })
+
+      expect((yield* jobs.pendingBackground).find((item) => item.id === job.id)).toBeUndefined()
+      const background = yield* jobs.background(job.id)
+
+      const running = (yield* jobs.pendingBackground).find((item) => item.id === job.id)
+      expect(running).toMatchObject({ id: job.id, recovery, status: "running" })
+      expect(running?.notificationID).toStartWith("msg_")
+      expect(background?.notificationID).toBe(running?.notificationID)
+
+      yield* Deferred.succeed(latch, undefined)
+      yield* jobs.wait({ id: job.id })
+
+      const completed = (yield* jobs.pendingBackground).find((item) => item.id === job.id)
+      expect(completed).toMatchObject({
+        id: job.id,
+        notificationID: running?.notificationID,
+        recovery,
+        status: "completed",
+        output: "done",
+      })
+      if (!completed) return yield* Effect.die("background marker missing")
+
+      yield* jobs.completeBackground(completed.notificationID)
+      expect((yield* jobs.pendingBackground).find((item) => item.id === job.id)).toBeUndefined()
+    }),
+  )
+
+  it.live("persists backgroundAll ownership before releasing a blocked subagent", () =>
+    Effect.gen(function* () {
+      const jobs = yield* Job.Service
+      const parentSessionID = SessionSchema.ID.make("ses_background_parent")
+      const latch = yield* Deferred.make<void>()
+      const recovery = {
+        kind: "subagent" as const,
+        parentSessionID,
+        childSessionID: SessionSchema.ID.make("ses_background_child"),
+        agent: "explore",
+        description: "Explore background recovery",
+      }
+      const job = yield* jobs.start({ type: "subagent", recovery, run: Deferred.await(latch).pipe(Effect.as("done")) })
+      const waiting = yield* jobs
+        .block({ id: job.id, sessionID: parentSessionID })
+        .pipe(Effect.forkIn(yield* Scope.Scope, { startImmediately: true }))
+
+      yield* jobs.backgroundAll({ sessionID: parentSessionID })
+      expect(yield* Fiber.join(waiting)).toMatchObject({ type: "backgrounded", info: { id: job.id } })
+
+      const marker = (yield* jobs.pendingBackground).find((item) => item.id === job.id)
+      expect(marker).toMatchObject({ id: job.id, recovery, status: "running" })
+      if (!marker) return yield* Effect.die("background marker missing")
+
+      yield* jobs.cancel(job.id)
+      expect((yield* jobs.pendingBackground).find((item) => item.id === job.id)).toMatchObject({
+        notificationID: marker.notificationID,
+        status: "cancelled",
+      })
+      yield* jobs.completeBackground(marker.notificationID)
+    }),
+  )
+
+  it.live("retains terminal errors for recovery until notification acknowledgment", () =>
+    Effect.gen(function* () {
+      const jobs = yield* Job.Service
+      const latch = yield* Deferred.make<void>()
+      const job = yield* jobs.start({
+        type: "shell",
+        recovery: {
+          kind: "shell",
+          sessionID: SessionSchema.ID.make("ses_background_error"),
+          shellID: "shell_error",
+          command: "exit 1",
+        },
+        run: Deferred.await(latch).pipe(Effect.andThen(Effect.fail(new Error("shell failed")))),
+      })
+
+      yield* jobs.background(job.id)
+      yield* Deferred.succeed(latch, undefined)
+      yield* jobs.wait({ id: job.id })
+
+      const marker = (yield* jobs.pendingBackground).find((item) => item.id === job.id)
+      expect(marker).toMatchObject({ id: job.id, status: "error", error: "shell failed" })
+      if (!marker) return yield* Effect.die("background marker missing")
+      yield* jobs.completeBackground(marker.notificationID)
+    }),
+  )
+
+  it.live("durably backgrounds recoverable work that has already failed", () =>
+    Effect.gen(function* () {
+      const jobs = yield* Job.Service
+      const job = yield* jobs.start({
+        type: "shell",
+        recovery: {
+          kind: "shell",
+          sessionID: SessionSchema.ID.make("ses_immediate_error"),
+          shellID: "shell_immediate_error",
+          command: "exit 1",
+        },
+        run: Effect.fail(new Error("shell failed")),
+      })
+      expect((yield* jobs.wait({ id: job.id })).info?.status).toBe("error")
+
+      const background = yield* jobs.background(job.id)
+      expect(background?.notificationID).toStartWith("msg_")
+      expect(yield* jobs.pendingBackground).toMatchObject([
+        { id: job.id, notificationID: background?.notificationID, status: "error", error: "shell failed" },
+      ])
+    }),
+  )
+
+  it.live("recovers a background marker after its process-local registry closes", () =>
+    Effect.gen(function* () {
+      const scope = yield* Scope.make()
+      const previous = yield* Job.make.pipe(Scope.provide(scope))
+      const job = yield* previous.start({
+        type: "shell",
+        recovery: {
+          kind: "shell",
+          sessionID: SessionSchema.ID.make("ses_background_restart"),
+          shellID: "shell_restart",
+          command: "sleep 60",
+        },
+        run: Effect.never,
+      })
+      yield* previous.background(job.id)
+      yield* Scope.close(scope, Exit.void)
+
+      const current = yield* Job.make
+      const marker = (yield* current.pendingBackground).find((item) => item.id === job.id)
+      expect(marker).toMatchObject({ id: job.id, status: "running" })
+      if (!marker) return yield* Effect.die("background marker missing")
+      yield* current.completeBackground(marker.notificationID)
+    }),
+  )
+
+  it.live("preserves running background ownership when its work is interrupted", () =>
+    Effect.gen(function* () {
+      const jobs = yield* Job.Service
+      const interrupted = yield* Deferred.make<void>()
+      const job = yield* jobs.start({
+        type: "subagent",
+        recovery: {
+          kind: "subagent",
+          parentSessionID: SessionSchema.ID.make("ses_interrupted_parent"),
+          childSessionID: SessionSchema.ID.make("ses_interrupted_child"),
+          agent: "explore",
+          description: "Continue after shutdown",
+        },
+        run: Deferred.await(interrupted).pipe(Effect.andThen(Effect.interrupt)),
+      })
+      yield* jobs.background(job.id)
+      yield* Deferred.succeed(interrupted, undefined)
+      yield* jobs.wait({ id: job.id })
+
+      const marker = (yield* jobs.pendingBackground).find((item) => item.id === job.id)
+      expect(marker).toMatchObject({ id: job.id, status: "running" })
+      if (!marker) return yield* Effect.die("background marker missing")
+      yield* jobs.completeBackground(marker.notificationID)
+    }),
+  )
+
   it.live("interrupts live work without promising settlement after the owning process-local scope closes", () =>
     Effect.gen(function* () {
       const scope = yield* Scope.make()

packages/core/test/lib/tool.ts

@@ -64,10 +64,8 @@ export const registerToolPlugin = <R>(
         hook: () => Effect.succeed({ dispose: Effect.void }),
       },
       tool: {
-        transform: (callback) =>
-          tools
-            .transform((draft) => callback({ add: (tool) => draft.add(tool) }))
-            .pipe(Effect.orDie, Effect.as({ dispose: Effect.void })),
+        transform: tools.transform,
+        reload: tools.reload,
         hook: () => Effect.die("registerToolPlugin does not support tool hooks"),
       },
     })

packages/core/test/mcp.test.ts

@@ -35,7 +35,7 @@ import { AbsolutePath } from "@opencode-ai/core/schema"
 import { Session } from "@opencode-ai/core/session"
 import { McpTool } from "@opencode-ai/core/tool/mcp"
 import { Tool } from "@opencode-ai/core/tool"
-import { DateTime, Deferred, Effect, Exit, Fiber, Layer, PubSub, Schedule, Schema, Sink, Stream } from "effect"
+import { DateTime, Deferred, Effect, Exit, Fiber, Layer, PubSub, Ref, Schedule, Schema, Sink, Stream } from "effect"
 import { ChildProcess, ChildProcessSpawner } from "effect/unstable/process"
 import { ExitCode, makeHandle, ProcessId } from "effect/unstable/process/ChildProcessSpawner"
 import { Image } from "@opencode-ai/core/image"
@@ -77,6 +77,7 @@ function resourceServer(
         templateLists: 0,
         toolLists: 0,
         initializations: 0,
+        urls: [] as string[],
       }
       const protocol = new Server(
         { name: "mcp-resources", version: "1.0.0" },
@@ -145,6 +146,7 @@ function resourceServer(
       const http = Bun.serve({
         port: 0,
         fetch: async (request) => {
+          state.urls.push(request.url)
           const body: unknown = request.method === "POST" ? await request.clone().json() : undefined
           if (typeof body === "object" && body !== null && "method" in body && body.method === "initialize") {
             state.initializations += 1
@@ -718,6 +720,40 @@ test("applies configured MCP timeouts to resource operations", async () => {
   await expect(read).rejects.toThrow("Request timed out")
 })
 
+for (const entry of [
+  { name: "default", query: "", codemode: undefined, expected: "?codemode=false" },
+  { name: "explicit local code mode", query: "", codemode: true, expected: "?codemode=false" },
+  { name: "direct tools", query: "", codemode: false, expected: "" },
+  {
+    name: "existing query",
+    query: "?source=opencode",
+    codemode: undefined,
+    expected: "?source=opencode&codemode=false",
+  },
+  { name: "explicit remote code mode", query: "?codemode=true", codemode: undefined, expected: "?codemode=true" },
+  { name: "explicit remote opt-out", query: "?codemode=false", codemode: undefined, expected: "?codemode=false" },
+  { name: "portal opt-out", query: "?codemode=off", codemode: undefined, expected: "?codemode=off" },
+]) {
+  testEffect(Layer.empty).live(`remote MCP code mode preference: ${entry.name}`, () =>
+    Effect.gen(function* () {
+      const server = yield* resourceServer()
+      const config = new ConfigMCP.Remote({
+        type: "remote",
+        url: server.url + entry.query,
+        oauth: false,
+        codemode: entry.codemode,
+      })
+      const connection = yield* connect("resources", config, import.meta.dir)
+      yield* connection.tools()
+      expect(server.state.initializations).toBe(1)
+      expect(server.state.toolLists).toBe(1)
+      expect(server.state.urls.length).toBeGreaterThanOrEqual(3)
+      expect(new Set(server.state.urls)).toEqual(new Set([server.url + entry.expected]))
+      expect(config.url).toBe(server.url + entry.query)
+    }),
+  )
+}
+
 test("lists, reads, and reports MCP resource changes", async () => {
   await Effect.runPromise(
     Effect.scoped(
@@ -1193,7 +1229,102 @@ test("serializes concurrent MCP lifecycle operations", async () => {
   )
 })
 
-it.effect("advertises MCP output schemas to Code Mode", () =>
+testEffect(Layer.empty).live("isolates invalid MCP tools and reapplies plugin mutations on catalog updates", () =>
+  Effect.gen(function* () {
+    const tool = (server: string, name: string) =>
+      new MCP.Tool({
+        server: MCP.ServerName.make(server),
+        name,
+        codemode: false,
+        inputSchema: { type: "object", properties: {} },
+      })
+    const healthy = [tool("demo", "search"), tool("other", "lookup")]
+    const namespace = tool("x".repeat(65), "lookup")
+    const catalog = yield* Ref.make([tool("demo", "x".repeat(65)), ...healthy, namespace])
+
+    yield* Effect.gen(function* () {
+      const registry = yield* Tool.Service
+      const registration = yield* McpTool.Service
+      const bus = yield* Bus.Service
+      const policy = yield* registry.transform((draft) => {
+        draft.update("demo_search", (tool) => {
+          tool.description = "Updated search"
+        })
+        draft.remove("other_lookup")
+      })
+      yield* registration.flush
+      expect((yield* toolDefinitions(registry)).map((tool) => tool.name)).toEqual(["demo_search", "execute"])
+
+      yield* Ref.set(catalog, [tool("demo", "y".repeat(65)), ...healthy, tool("demo", "added"), namespace])
+      yield* bus.publish(McpEvent.ToolsChanged, { server: "demo" })
+      yield* waitForTool(registry, "demo_added")
+      expect((yield* toolDefinitions(registry)).map((tool) => tool.name)).toEqual([
+        "demo_added",
+        "demo_search",
+        "execute",
+      ])
+      expect((yield* toolDefinitions(registry)).find((tool) => tool.name === "demo_search")?.description).toBe(
+        "Updated search",
+      )
+      expect(
+        yield* executeTool(registry, {
+          sessionID: Session.ID.make("ses_mcp_invalid_catalog"),
+          ...toolIdentity,
+          call: { type: "tool-call", id: "call_demo_search", name: "demo_search", input: {} },
+        }),
+      ).toMatchObject({ status: "completed" })
+
+      yield* Ref.set(catalog, [
+        tool("demo", "status"),
+        tool("other", "lookup"),
+        tool("demo", "added"),
+        tool("repaired", "lookup"),
+      ])
+      yield* bus.publish(McpEvent.ToolsChanged, { server: "demo" })
+      yield* waitForTool(registry, "demo_status")
+      expect((yield* toolDefinitions(registry)).map((tool) => tool.name)).toEqual([
+        "demo_added",
+        "demo_status",
+        "repaired_lookup",
+        "execute",
+      ])
+      yield* policy.dispose
+      expect((yield* toolDefinitions(registry)).map((tool) => tool.name)).toEqual([
+        "demo_added",
+        "demo_status",
+        "other_lookup",
+        "repaired_lookup",
+        "execute",
+      ])
+    }).pipe(
+      Effect.provide(
+        Layer.fresh(
+          AppNodeBuilder.build(LayerNode.group([Tool.node, McpTool.node, Bus.node]), [
+            [
+              MCP.node,
+              Layer.mock(MCP.Service, {
+                tools: () => Ref.get(catalog),
+                callTool: (input) =>
+                  Effect.succeed(
+                    new MCP.ToolResult({
+                      server: MCP.ServerName.make(input.server),
+                      tool: input.name,
+                      isError: false,
+                      content: [{ type: "text", text: "healthy" }],
+                    }),
+                  ),
+              }),
+            ],
+            [Permission.node, Layer.mock(Permission.Service, { assert: () => Effect.void })],
+            [Image.node, imagePassthrough],
+          ]),
+        ),
+      ),
+    )
+  }),
+)
+
+it.live("advertises MCP output schemas to Code Mode", () =>
   Effect.gen(function* () {
     const registry = yield* Tool.Service
     const toolSet = yield* waitForCodeModeTool(registry, "demo.search")
@@ -1210,7 +1341,7 @@ it.effect("advertises MCP output schemas to Code Mode", () =>
   }),
 )
 
-it.effect("returns content-only MCP results through Code Mode", () =>
+it.live("returns content-only MCP results through Code Mode", () =>
   Effect.gen(function* () {
     assertion = yield* Deferred.make<Permission.AssertInput>()
     decision = Effect.void
@@ -1235,7 +1366,7 @@ it.effect("returns content-only MCP results through Code Mode", () =>
   }),
 )
 
-it.effect("advertises MCP tools directly when Code Mode is disabled for the server", () =>
+it.live("advertises MCP tools directly when Code Mode is disabled for the server", () =>
   Effect.gen(function* () {
     const registry = yield* Tool.Service
     yield* waitForTool(registry, "direct_lookup")
@@ -1249,7 +1380,7 @@ it.effect("advertises MCP tools directly when Code Mode is disabled for the serv
 
 // Baseline (PLAN.md step 1): MCP isError must become one failed tool call, not a
 // success whose text happens to describe an error.
-it.effect("fails the call when MCP reports isError", () =>
+it.live("fails the call when MCP reports isError", () =>
   Effect.gen(function* () {
     assertion = yield* Deferred.make<Permission.AssertInput>()
     decision = Effect.void
@@ -1267,7 +1398,7 @@ it.effect("fails the call when MCP reports isError", () =>
 )
 
 // Baseline (PLAN.md step 1): mixed MCP text and media content must reach the model intact.
-it.effect("preserves MCP text and media content for the model", () =>
+it.live("preserves MCP text and media content for the model", () =>
   Effect.gen(function* () {
     assertion = yield* Deferred.make<Permission.AssertInput>()
     decision = Effect.void
@@ -1288,7 +1419,7 @@ it.effect("preserves MCP text and media content for the model", () =>
   }),
 )
 
-it.effect("waits for permission before calling an MCP tool", () =>
+it.live("waits for permission before calling an MCP tool", () =>
   Effect.gen(function* () {
     calls = 0
     assertion = yield* Deferred.make<Permission.AssertInput>()
@@ -1330,7 +1461,7 @@ it.effect("waits for permission before calling an MCP tool", () =>
   }),
 )
 
-it.effect("does not call MCP when permission is blocked", () =>
+it.live("does not call MCP when permission is blocked", () =>
   Effect.gen(function* () {
     calls = 0
     assertion = yield* Deferred.make<Permission.AssertInput>()

packages/core/test/model-resolver.test.ts

@@ -1126,7 +1126,7 @@ describe("ModelResolver", () => {
       const mantle = yield* ModelResolver.fromCatalogModel(
         model(Provider.aisdk("@ai-sdk/amazon-bedrock/mantle"), {
           modelID: "openai.gpt-oss-120b",
-          settings: { region: "us-east-1" },
+          settings: { region: "us-east-1", topP: 0.6 },
         }),
       )
 
@@ -1158,6 +1158,7 @@ describe("ModelResolver", () => {
       expect(bedrock.route.defaults.generation).toEqual({ topP: 0.8 })
       expect(bedrock.route.defaults.http?.body).toEqual({ serviceTier: { type: "priority" } })
       expect(mantle.route.id).toBe("bedrock-mantle-responses")
+      expect(mantle.route.defaults.generation).toEqual({ topP: 0.6 })
     }),
   )
 

packages/core/test/npm.test.ts

@@ -223,6 +223,7 @@ describe("Npm.add", () => {
     ).toBeTruthy()
   })
 
+  // Several real Git installs and refreshes exceed Bun's default timeout on Windows.
   test("refreshes mutable Git packages once per service lifetime and preserves pinned or cached installs", async () => {
     await using tmp = await tmpdir()
     const fixture = await createGitFixture(tmp.path)
@@ -262,7 +263,7 @@ describe("Npm.add", () => {
       return yield* npm.add(mutable, { refresh: true })
     }).pipe(Effect.scoped, Effect.provide(npmLayer(cache)), Effect.runPromise)
     expect(await Bun.file(path.join(offline.directory, "index.js")).text()).toContain('root: "second"')
-  })
+  }, 30_000)
 })
 
 describe("Npm.resolve", () => {

packages/core/test/permission.test.ts

@@ -17,6 +17,7 @@ import { AbsolutePath } from "@opencode-ai/core/schema"
 import { Session } from "@opencode-ai/core/session"
 import { SessionTable } from "@opencode-ai/core/session/sql"
 import { SessionStore } from "@opencode-ai/core/session/store"
+import { ShellParse } from "@opencode-ai/core/shell/parse"
 import { eq } from "drizzle-orm"
 import { location } from "./fixture/location"
 import { testEffect } from "./lib/effect"
@@ -384,3 +385,334 @@ describe("Permission", () => {
     }),
   )
 })
+
+describe("shell scanner permission impact", () => {
+  // Fixed cases require matching outcomes; remaining differences are investigation snapshots, not contracts.
+  // These service-level cases all have command resources; tool tests cover skipped checks and directories.
+  // Outcome pairs are [legacy, native].
+  for (const fixture of [
+    {
+      name: "timed command preserves wrapper approvals",
+      shell: "bash",
+      command: "time -p git status",
+      approved: ["time *"],
+      exact: ["time -p git status"],
+      denied: "time -p git status",
+      savedEffect: ["allow", "allow"],
+      exactEffect: ["allow", "allow"],
+      deniedEffect: ["deny", "deny"],
+    },
+    {
+      name: "coprocess command preserves wrapper approvals",
+      shell: "bash",
+      command: "coproc git status",
+      approved: ["coproc *"],
+      exact: ["coproc git status"],
+      denied: "coproc git status",
+      savedEffect: ["allow", "allow"],
+      exactEffect: ["allow", "allow"],
+      deniedEffect: ["deny", "deny"],
+    },
+    {
+      name: "declarations and unset",
+      shell: "bash",
+      command: "export X=value; unset X; git status",
+      approved: ["git status *"],
+      exact: ["git status"],
+      denied: "export *",
+      savedEffect: ["allow", "allow"],
+      exactEffect: ["allow", "allow"],
+      deniedEffect: ["allow", "allow"],
+    },
+    {
+      name: "export with an approved command substitution",
+      shell: "bash",
+      command: "export VERSION=$(git describe --tags); npm run build",
+      approved: ["git describe *", "npm run build *"],
+      exact: ["git describe --tags", "npm run build"],
+      denied: "export *",
+      savedEffect: ["allow", "allow"],
+      exactEffect: ["allow", "allow"],
+      deniedEffect: ["allow", "allow"],
+    },
+    {
+      name: "export retains checks on the command substitution",
+      shell: "bash",
+      command: "export VERSION=$(git describe --tags); npm run build",
+      approved: ["git describe *", "npm run build *"],
+      exact: ["git describe --tags", "npm run build"],
+      denied: "git describe *",
+      savedEffect: ["allow", "allow"],
+      exactEffect: ["allow", "allow"],
+      deniedEffect: ["deny", "deny"],
+    },
+    {
+      name: "redirect after a conditional list",
+      shell: "bash",
+      command: "printf ok && git status > output",
+      approved: ["printf *", "git status *"],
+      exact: ["printf ok", "git status"],
+      denied: "git status",
+      savedEffect: ["allow", "allow"],
+      exactEffect: ["allow", "allow"],
+      deniedEffect: ["deny", "deny"],
+    },
+    {
+      name: "redirect after a pipeline",
+      shell: "bash",
+      command: "printf ok | cat < input > output",
+      approved: ["printf *", "cat *"],
+      exact: ["printf ok", "cat"],
+      denied: "cat",
+      savedEffect: ["allow", "allow"],
+      exactEffect: ["allow", "allow"],
+      deniedEffect: ["deny", "deny"],
+    },
+    {
+      name: "assignment redirect followed by a command",
+      shell: "bash",
+      command: "FOO=bar > output; printf done",
+      approved: ["printf *"],
+      exact: ["printf done"],
+      denied: "FOO=bar > output; printf done",
+      savedEffect: ["ask", "allow"],
+      exactEffect: ["ask", "allow"],
+      deniedEffect: ["deny", "allow"],
+    },
+    {
+      name: "assignment redirect with an approved command substitution",
+      shell: "bash",
+      command: "VERSION=$(git describe --tags) > build/version.txt",
+      approved: ["git describe *"],
+      exact: ["git describe --tags"],
+      denied: "VERSION=$(git describe --tags) > build/version.txt",
+      savedEffect: ["ask", "allow"],
+      exactEffect: ["ask", "allow"],
+      deniedEffect: ["deny", "allow"],
+    },
+    {
+      name: "substitution in a saved prefix",
+      shell: "bash",
+      command: "git $(printf diff) --stat",
+      approved: ["git *", "printf *"],
+      exact: ["git $(printf diff) --stat", "printf diff"],
+      denied: "git $(printf diff) --stat",
+      savedEffect: ["allow", "allow"],
+      exactEffect: ["allow", "allow"],
+      deniedEffect: ["deny", "deny"],
+    },
+    {
+      name: "standalone PowerShell scriptblock caller",
+      shell: "pwsh",
+      command: "ForEach-Object { Write-Output value }",
+      approved: ["Write-Output *"],
+      exact: ["Write-Output value"],
+      denied: "ForEach-Object *",
+      savedEffect: ["allow", "allow"],
+      exactEffect: ["allow", "allow"],
+      deniedEffect: ["allow", "allow"],
+    },
+    {
+      name: "tab-separated PowerShell command",
+      shell: "pwsh",
+      command: "git\tstatus; Write-Output done",
+      approved: ["git status *", "Write-Output *"],
+      exact: ["git status", "Write-Output done"],
+      denied: "git\tstatus",
+      savedEffect: ["allow", "ask"],
+      exactEffect: ["allow", "ask"],
+      deniedEffect: ["allow", "deny"],
+    },
+    {
+      name: "PowerShell equals-joined argument",
+      shell: "pwsh",
+      command: "git --work-tree=src status",
+      approved: ["git --work-tree *"],
+      exact: ["git --work-tree"],
+      denied: "git --work-tree",
+      savedEffect: ["allow", "ask"],
+      exactEffect: ["allow", "ask"],
+      deniedEffect: ["deny", "allow"],
+    },
+  ] as const) {
+    for (const scenario of [
+      { name: "no approval", saved: [], rules: [], expected: ["ask", "ask"] },
+      { name: "saved wildcard", saved: ["*"], rules: [], expected: ["allow", "allow"] },
+      { name: "saved command approvals", saved: fixture.approved, rules: [], expected: fixture.savedEffect },
+      {
+        name: "exact configured approvals",
+        saved: [],
+        rules: fixture.exact.map((resource): Permission.Rule => ({ action: "shell", resource, effect: "allow" })),
+        expected: fixture.exactEffect,
+      },
+      {
+        name: "exact saved approvals",
+        saved: fixture.exact,
+        rules: [],
+        expected: fixture.exactEffect,
+      },
+      {
+        name: "configured deny despite saved wildcard",
+        saved: ["*"],
+        rules: [{ action: "shell", resource: fixture.denied, effect: "deny" }] satisfies Permission.Ruleset,
+        expected: fixture.deniedEffect,
+      },
+    ] as const) {
+      it.live(`${fixture.name}: ${scenario.name}`, () =>
+        Effect.gen(function* () {
+          yield* setup(scenario.rules)
+          const saved = yield* PermissionSaved.Service
+          yield* saved.add({ projectID: Project.ID.global, action: "shell", resources: scenario.saved })
+          const service = yield* Permission.Service
+
+          for (const [index, portable] of [false, true].entries()) {
+            const parsed = yield* ShellParse.scan(fixture.command, fixture.shell, "/project", { portable })
+            expect(parsed.commands.length).toBeGreaterThan(0)
+            expect(parsed.directories).toEqual([])
+            const result = yield* service.ask(
+              assertion({
+                action: "shell",
+                resources: parsed.commands.map((command) => command.resource),
+                save: parsed.commands.map((command) => command.save),
+              }),
+            )
+            expect(result.effect, portable ? "native" : "legacy").toBe(scenario.expected[index])
+            const pending = yield* service.list()
+            expect(pending).toHaveLength(result.effect === "ask" ? 1 : 0)
+            if (result.effect !== "ask") continue
+            expect(pending[0]?.resources).toEqual(parsed.commands.map((command) => command.resource))
+            expect(pending[0]?.save).toEqual(parsed.commands.map((command) => command.save))
+            yield* service.reply({ requestID: result.id, reply: "once" })
+            expect(yield* service.list()).toEqual([])
+          }
+        }),
+      )
+    }
+  }
+
+  // Grant/repeat rows select the granting parser; repeat columns select the parser used afterwards.
+  for (const fixture of [
+    {
+      name: "numeric npm script prefix",
+      shell: "bash",
+      command: "npm run 123",
+      grants: [["npm run *"], ["npm run 123 *"]],
+      repeat: [
+        ["allow", "allow"],
+        ["allow", "allow"],
+      ],
+      next: "npm run build",
+      nextEffect: ["allow", "ask"],
+    },
+    {
+      name: "numeric AWS option prefix",
+      shell: "bash",
+      command: "aws --cli-read-timeout 60 s3 ls",
+      grants: [["aws --cli-read-timeout s3 *"], ["aws --cli-read-timeout 60 *"]],
+      repeat: [
+        ["ask", "ask"],
+        ["allow", "allow"],
+      ],
+      next: "aws --cli-read-timeout 60 ec2 describe-instances",
+      nextEffect: ["ask", "allow"],
+    },
+    {
+      name: "substitution prefix",
+      shell: "bash",
+      command: "git $(printf diff) --stat",
+      grants: [
+        ["git --stat *", "printf *"],
+        ["git $(printf diff) *", "printf *"],
+      ],
+      repeat: [
+        ["ask", "ask"],
+        ["allow", "allow"],
+      ],
+      next: "git --stat",
+      nextEffect: ["allow", "ask"],
+    },
+    {
+      name: "redirect prefix",
+      shell: "bash",
+      command: "printf ok && git status > output",
+      grants: [
+        ["printf *", "git status *"],
+        ["printf *", "git status *"],
+      ],
+      repeat: [
+        ["allow", "allow"],
+        ["allow", "allow"],
+      ],
+      next: "git status --short",
+      nextEffect: ["allow", "allow"],
+    },
+    {
+      name: "assignment redirect prefix",
+      shell: "bash",
+      command: "FOO=bar > output; printf done",
+      grants: [["printf *"], ["printf *"]],
+      // Identical saved rules cover only the native resource, regardless of which parser saved them.
+      repeat: [
+        ["ask", "allow"],
+        ["ask", "allow"],
+      ],
+      next: "printf next",
+      nextEffect: ["allow", "allow"],
+    },
+    {
+      name: "PowerShell tab prefix",
+      shell: "pwsh",
+      command: "git\tstatus; Write-Output done",
+      grants: [["Write-Output *"], ["git\tstatus *", "Write-Output *"]],
+      repeat: [
+        ["allow", "ask"],
+        ["allow", "allow"],
+      ],
+      next: "git status",
+      nextEffect: ["ask", "ask"],
+    },
+  ] as const) {
+    for (const [origin, portable] of [false, true].entries()) {
+      it.live(`${fixture.name}: always allow from ${portable ? "native" : "legacy"}, then use either parser`, () =>
+        Effect.gen(function* () {
+          yield* setup()
+          const service = yield* Permission.Service
+          const saved = yield* PermissionSaved.Service
+          const parsed = yield* ShellParse.scan(fixture.command, fixture.shell, "/project", { portable })
+          const first = yield* service.ask(
+            assertion({
+              action: "shell",
+              resources: parsed.commands.map((command) => command.resource),
+              save: parsed.commands.map((command) => command.save),
+            }),
+          )
+          expect(first.effect).toBe("ask")
+          expect(yield* service.list()).toHaveLength(1)
+          yield* service.reply({ requestID: first.id, reply: "always" })
+          expect(yield* service.list()).toEqual([])
+          expect((yield* saved.list({ projectID: Project.ID.global })).map((rule) => rule.resource).sort()).toEqual(
+            [...fixture.grants[portable ? 1 : 0]].sort(),
+          )
+
+          for (const [index, target] of [false, true].entries()) {
+            for (const command of [fixture.command, fixture.next]) {
+              const parsed = yield* ShellParse.scan(command, fixture.shell, "/project", { portable: target })
+              const result = yield* service.ask(
+                assertion({
+                  action: "shell",
+                  resources: parsed.commands.map((command) => command.resource),
+                  save: parsed.commands.map((command) => command.save),
+                }),
+              )
+              expect(result.effect, `${target ? "native" : "legacy"}: ${command}`).toBe(
+                command === fixture.next ? fixture.nextEffect[origin] : fixture.repeat[origin]?.[index],
+              )
+              if (result.effect === "ask") yield* service.reply({ requestID: result.id, reply: "once" })
+              expect(yield* service.list()).toEqual([])
+            }
+          }
+        }),
+      )
+    }
+  }
+})

packages/core/test/plugin.test.ts

@@ -282,6 +282,47 @@ describe("Plugin", () => {
     }),
   )
 
+  it.effect("keeps plugins active when a tool registration is invalid", () =>
+    Effect.gen(function* () {
+      const plugins = yield* Plugin.Service
+      const tools = yield* Tool.Service
+      const agents = yield* Agent.Service
+      yield* plugins.activate([
+        {
+          id: "partial-tools",
+          version: "1",
+          effect: (ctx) =>
+            Effect.gen(function* () {
+              yield* ctx.tool.transform((draft) => {
+                const tool = {
+                  name: "healthy",
+                  description: "Healthy tool",
+                  input: Schema.Struct({}),
+                  execute: () => Effect.succeed({ content: "ok" }),
+                  options: { codemode: false },
+                }
+                draft.add({ ...tool, name: "invalid", options: { namespace: "invalid..namespace" } })
+                draft.add(tool)
+              })
+              yield* ctx.agent.transform((draft) =>
+                draft.update("configured", (agent) => {
+                  agent.description = "setup continued"
+                }),
+              )
+            }),
+        },
+      ])
+
+      expect(yield* plugins.list()).toEqual([
+        { id: Plugin.ID.make("partial-tools"), source: { type: "builtin" }, status: "active", tui: false },
+      ])
+      expect((yield* agents.get(Agent.ID.make("configured")))?.description).toBe("setup continued")
+      expect((yield* tools.snapshot()).definitions.map((tool) => tool.name)).toEqual(["healthy", "execute"])
+      yield* plugins.activate([])
+      expect((yield* tools.snapshot()).definitions.map((tool) => tool.name)).toEqual(["execute"])
+    }),
+  )
+
   it.effect("restores the previous plugin when its replacement fails", () =>
     Effect.gen(function* () {
       const plugins = yield* Plugin.Service

packages/core/test/plugin/host.ts

@@ -114,6 +114,7 @@ export function host(overrides: Overrides = {}): Plugin.Context {
     },
     tool: overrides.tool ?? {
       transform: () => Effect.die("unused tool.transform"),
+      reload: () => Effect.die("unused tool.reload"),
       hook: () => Effect.die("unused tool.hook"),
     },
     vcs: overrides.vcs ?? {

packages/core/test/plugin/mcp-codemode-exclusion.test.ts

@@ -21,7 +21,12 @@ it.effect("defaults only known Code Mode MCP servers to direct tools", () =>
       {
         name: "cloudflare code mode",
         server: { type: "remote", url: "https://mcp.cloudflare.com/mcp/" },
-        codemode: false,
+        codemode: undefined,
+      },
+      {
+        name: "cloudflare raw tools",
+        server: { type: "remote", url: "https://mcp.cloudflare.com/mcp?codemode=false" },
+        codemode: undefined,
       },
       {
         name: "cloudflare docs",

packages/core/test/plugin/plan.test.ts

@@ -72,6 +72,7 @@ const run = Effect.fnUntraced(function* (events: ReadonlyArray<SessionEvent.Agen
       },
       tool: {
         transform: () => Effect.die("unused tool.transform"),
+        reload: () => Effect.die("unused tool.reload"),
         hook: (name, callback) => {
           if (name === "execute.after") {
             // Hook names and callbacks are correlated, but TypeScript does not narrow this generic registration API.

packages/core/test/plugin/promise.test.ts

@@ -636,4 +636,77 @@ describe("fromPromise", () => {
       })
     }),
   )
+
+  it.live("adapts tool mutation, replay, and disposal through the Promise API", () =>
+    Effect.gen(function* () {
+      const plugins = yield* Plugin.Service
+      const registry = yield* Tool.Service
+      const host = yield* PluginHost.make(plugins)
+      const progress: Tool.Metadata[] = []
+      let greeting = "Hello"
+      let registration: { dispose(): Promise<void> } | undefined
+      yield* host.tool.transform((draft) => {
+        const text = greeting
+        draft.add({
+          name: "hello",
+          description: "Hello",
+          options: { namespace: "acme", codemode: false },
+          input: Schema.Struct({ name: Schema.String }),
+          output: Schema.String,
+          execute: ({ name }, context) =>
+            context.progress({ phase: "original" }).pipe(Effect.as({ output: `${text}, ${name}!` })),
+        })
+        draft.add({
+          name: "temporary",
+          description: "Temporary",
+          input: Schema.Struct({}),
+          options: { codemode: false },
+          execute: () => Effect.succeed({ content: "temporary" }),
+        })
+      })
+      yield* PluginPromise.fromPromise(
+        define({
+          id: "promise-update",
+          setup: async (ctx) => {
+            registration = await ctx.tool.transform((draft) => {
+              draft.update("missing", () => {
+                throw new Error("must not create a tool")
+              })
+              draft.update("acme_hello", (tool) => {
+                const execute = tool.execute
+                tool.description = "Wrapped"
+                tool.execute = async (input, context) => {
+                  const result = await execute(input, context)
+                  return { ...result, output: `${result.output} Wrapped.` }
+                }
+              })
+              draft.remove("temporary")
+            })
+            greeting = "Hi"
+            await ctx.tool.reload()
+          },
+        }),
+      ).effect(host)
+      const snapshot = yield* registry.snapshot()
+      expect(snapshot.definitions.map((tool) => tool.name)).toEqual(["acme_hello", "execute"])
+      expect(snapshot.definitions[0]?.description).toBe("Wrapped")
+      expect(
+        yield* snapshot.execute({
+          sessionID: Session.ID.make("ses_promise_update"),
+          agent: Agent.ID.make("build"),
+          messageID: SessionMessage.ID.make("msg_promise_update"),
+          progress: (value) => Effect.sync(() => progress.push(value)),
+          call: { type: "tool-call", id: "call_promise_update", name: "acme_hello", input: { name: "world" } },
+        }),
+      ).toMatchObject({ output: "Hi, world! Wrapped." })
+      expect(progress).toEqual([{ phase: "original" }])
+      const registered = registration
+      if (!registered) throw new Error("Expected registration")
+      yield* Effect.promise(() => registered.dispose())
+      yield* Effect.promise(() => registered.dispose())
+      const restored = yield* registry.snapshot()
+      expect(restored.definitions.map((tool) => tool.name)).toEqual(["acme_hello", "temporary", "execute"])
+      expect(restored.definitions[0]?.description).toBe("Hello")
+    }),
+  )
 })

packages/core/test/session-execution.test.ts

@@ -4,6 +4,8 @@ import { Database } from "@opencode-ai/core/database/database"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { LayerNode } from "@opencode-ai/util/effect/layer-node"
 import { Bus } from "@opencode-ai/core/bus"
+import { Job } from "@opencode-ai/core/job"
+import { KV } from "@opencode-ai/core/kv"
 import { LocationServiceMap } from "@opencode-ai/core/location-service-map"
 import type { LocationServices } from "@opencode-ai/core/location-services"
 import { Project } from "@opencode-ai/core/project"
@@ -23,7 +25,9 @@ import { Context, Deferred, Effect, Exit, Fiber, Layer, LayerMap, Scope } from "
 import { eq } from "drizzle-orm"
 import { testEffect } from "./lib/effect"
 
-const it = testEffect(AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SessionStore.node])))
+const it = testEffect(
+  AppNodeBuilder.build(LayerNode.group([Database.node, Bus.node, SessionStore.node, Job.node, KV.node, Session.node])),
+)
 
 describe("SessionExecution lifecycle", () => {
   test("classifies success and typed failure terminals", () => {
@@ -60,14 +64,13 @@ describe("SessionExecution lifecycle", () => {
       const idle = Session.ID.make("ses_recover_idle")
       yield* seedSessions(database, [parent], { time_suspended: Date.now() })
       yield* seedSessions(database, [idle])
-      // An orphaned child is never resumed: the resumed parent re-runs its
-      // tool call and spawns a fresh child instead.
+      // Children recover through background Job records, never through the root claim sweep.
       yield* seedSessions(database, [child], { time_suspended: Date.now(), parent_id: parent })
 
       expect(yield* store.listSuspended()).toEqual([parent])
 
       // The sweep clears orphaned child claims outright; parents keep theirs.
-      yield* store.releaseChildClaims
+      yield* store.releaseChildClaims([])
       expect(yield* claims(database)).toEqual({ [parent]: true, [child]: false, [idle]: false })
     }),
   )
@@ -147,6 +150,66 @@ describe("SessionExecution lifecycle", () => {
     }),
   )
 
+  it.effect("does not resume a user-cancelled background child whose notification was not admitted", () =>
+    Effect.gen(function* () {
+      const database = yield* Database.Service
+      const parent = Session.ID.make("ses_cancelled_background_parent")
+      const child = Session.ID.make("ses_cancelled_background_child")
+      yield* seedSessions(database, [parent])
+      yield* seedSessions(database, [child], { parent_id: parent })
+
+      const running = yield* Deferred.make<void>()
+      const scope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+      const jobs = yield* Job.make.pipe(Scope.provide(scope))
+      const context = yield* buildExecution(
+        scope,
+        () => Deferred.succeed(running, undefined).pipe(Effect.andThen(Effect.never)),
+        undefined,
+        jobs,
+      )
+      const execution = Context.get(context, SessionExecution.Service)
+      yield* jobs.start({
+        id: child,
+        type: "subagent",
+        recovery: {
+          kind: "subagent",
+          parentSessionID: parent,
+          childSessionID: child,
+          agent: "general",
+          description: "Cancelled inspection",
+        },
+        run: execution.resume(child).pipe(Effect.as("unused")),
+      })
+      yield* jobs.background(child)
+      yield* Deferred.await(running)
+      expect(yield* execution.interrupt(child)).toBeTrue()
+      yield* execution.awaitIdle(child)
+      expect((yield* jobs.wait({ id: child })).info?.status).toBe("cancelled")
+      expect(yield* jobs.pendingBackground).toMatchObject([{ id: child, status: "cancelled" }])
+      expect((yield* claims(database))[child]).toBe(false)
+      yield* Scope.close(scope, Exit.void)
+
+      const restartedScope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(restartedScope, Exit.void))
+      const restartedJobs = yield* Job.make.pipe(Scope.provide(restartedScope))
+      const drained: Session.ID[] = []
+      const restarted = yield* buildExecution(
+        restartedScope,
+        ({ sessionID }) => Effect.sync(() => void drained.push(sessionID)),
+        undefined,
+        restartedJobs,
+      )
+      yield* Context.get(restarted, SessionRestart.Service).resumeSuspendedSessions
+      yield* Context.get(restarted, SessionExecution.Service).awaitIdle(parent)
+      expect(drained).toEqual([parent])
+      expect(yield* SessionInbox.list(database.db, parent)).toMatchObject([
+        { payload: { text: expect.stringContaining("Subagent cancelled"), metadata: { state: "cancelled" } } },
+      ])
+      expect(yield* restartedJobs.pendingBackground).toEqual([])
+    }),
+  )
+
   it.effect("starts every claimed execution without waiting for earlier drains to finish", () =>
     Effect.gen(function* () {
       const database = yield* Database.Service
@@ -304,6 +367,518 @@ describe("SessionExecution lifecycle", () => {
   )
 })
 
+describe("SessionRestart background recovery", () => {
+  it.effect("admits orphaned shell notices without waking and delivers them once on the next run", () =>
+    Effect.gen(function* () {
+      const database = yield* Database.Service
+      const store = yield* SessionStore.Service
+      const jobs = yield* Job.Service
+      const bus = yield* Bus.Service
+      const parent = Session.ID.make("ses_background_recovery_parent")
+      const child = Session.ID.make("ses_background_recovery_child")
+      yield* seedSessions(database, [parent])
+      yield* seedSessions(database, [child], { parent_id: parent, time_suspended: Date.now() })
+      yield* seedBackground(jobs, parent, [
+        { id: "call-background-shell", shellID: "sh_background_orphan", command: "sleep 60" },
+      ])
+      yield* seedBackground(jobs, child, [{ id: "call-child-shell", shellID: "sh_child_orphan", command: "sleep 30" }])
+
+      expect(yield* store.listSuspended()).toEqual([])
+      expect(yield* jobs.pendingBackground).toHaveLength(2)
+
+      const drained: Session.ID[] = []
+      const scope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+      const restarted = yield* Job.make.pipe(Effect.provideService(Scope.Scope, scope))
+      const context = yield* buildExecution(
+        scope,
+        ({ sessionID }) =>
+          Effect.sync(() => void drained.push(sessionID)).pipe(
+            Effect.andThen(SessionInbox.promote(database.db, bus, sessionID, "steer")),
+            Effect.asVoid,
+          ),
+        undefined,
+        restarted,
+      )
+      const restart = Context.get(context, SessionRestart.Service)
+      yield* restart.resumeSuspendedSessions
+
+      expect((yield* store.context(parent)).filter((message) => message.type === "synthetic")).toEqual([])
+      expect(yield* SessionInbox.list(database.db, parent)).toMatchObject([
+        {
+          type: "synthetic",
+          payload: {
+            description: "sleep 60",
+            text: expect.stringContaining("server restarted"),
+            metadata: {
+              source: "shell",
+              jobID: "call-background-shell",
+              shellID: "sh_background_orphan",
+              state: "cancelled",
+            },
+          },
+        },
+      ])
+      expect(yield* SessionInbox.list(database.db, child)).toMatchObject([
+        {
+          type: "synthetic",
+          payload: {
+            metadata: {
+              source: "shell",
+              jobID: "call-child-shell",
+              shellID: "sh_child_orphan",
+              state: "cancelled",
+            },
+          },
+        },
+      ])
+      expect(drained).toEqual([])
+      expect(yield* claims(database)).toEqual({ [parent]: false, [child]: false })
+      expect(yield* restarted.pendingBackground).toEqual([])
+
+      yield* restart.resumeSuspendedSessions
+      expect(yield* SessionInbox.list(database.db, parent)).toHaveLength(1)
+      expect(drained).toEqual([])
+      const execution = Context.get(context, SessionExecution.Service)
+      yield* execution.resume(parent)
+      expect(yield* SessionInbox.list(database.db, parent)).toEqual([])
+      expect((yield* store.context(parent)).filter((message) => message.type === "synthetic")).toHaveLength(1)
+      yield* execution.resume(parent)
+      expect((yield* store.context(parent)).filter((message) => message.type === "synthetic")).toHaveLength(1)
+    }),
+  )
+
+  it.effect("preserves locally running background work", () =>
+    Effect.gen(function* () {
+      const database = yield* Database.Service
+      const store = yield* SessionStore.Service
+      const jobs = yield* Job.Service
+      const parent = Session.ID.make("ses_background_existing_parent")
+      yield* seedSessions(database, [parent])
+      yield* seedBackground(jobs, parent, [{ id: "call-running-shell", shellID: "sh_running", command: "sleep 60" }])
+
+      const scope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+      const context = yield* buildExecution(scope, () => Effect.void)
+      const restart = Context.get(context, SessionRestart.Service)
+      yield* restart.resumeSuspendedSessions
+
+      expect((yield* store.context(parent)).filter((message) => message.type === "synthetic")).toEqual([])
+      expect(yield* jobs.get("call-running-shell")).toMatchObject({ status: "running" })
+      expect(yield* jobs.pendingBackground).toHaveLength(1)
+    }),
+  )
+
+  it.effect("preserves a silent shell failure persisted before its completion notification", () =>
+    Effect.gen(function* () {
+      const database = yield* Database.Service
+      const jobs = yield* Job.Service
+      const sessionID = Session.ID.make("ses_background_completed_shell")
+      yield* seedSessions(database, [sessionID])
+      const complete = yield* Deferred.make<string>()
+      yield* jobs.start({
+        id: "call-completed-shell",
+        type: "shell",
+        recovery: {
+          kind: "shell",
+          sessionID,
+          shellID: "sh_completed",
+          command: "exit 7",
+        },
+        run: Deferred.await(complete),
+      })
+      yield* jobs.background("call-completed-shell")
+      yield* Deferred.succeed(complete, "(no output)\n\nCommand exited with code 7.")
+      yield* jobs.wait({ id: "call-completed-shell" })
+
+      const scope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+      const restarted = yield* Job.make.pipe(Effect.provideService(Scope.Scope, scope))
+      const context = yield* buildExecution(scope, () => Effect.void, undefined, restarted)
+      yield* Context.get(context, SessionRestart.Service).resumeSuspendedSessions
+
+      expect(yield* SessionInbox.list(database.db, sessionID)).toMatchObject([
+        {
+          type: "synthetic",
+          payload: {
+            text: expect.stringContaining("(no output)\n\nCommand exited with code 7."),
+            metadata: { source: "shell", shellID: "sh_completed", state: "completed" },
+          },
+        },
+      ])
+      expect(yield* restarted.pendingBackground).toEqual([])
+    }),
+  )
+
+  for (const delivered of [false, true]) {
+    it.effect(`does not duplicate a shell notification already ${delivered ? "delivered" : "admitted"}`, () =>
+      Effect.gen(function* () {
+        const database = yield* Database.Service
+        const bus = yield* Bus.Service
+        const jobs = yield* Job.Service
+        const sessions = yield* Session.Service
+        const sessionID = Session.ID.make("ses_shell_notification_retry")
+        yield* seedSessions(database, [sessionID])
+        yield* seedBackground(jobs, sessionID, [
+          { id: "call-shell-notified", shellID: "sh_notified", command: "echo done" },
+        ])
+        const background = (yield* jobs.pendingBackground)[0]
+        if (!background) return yield* Effect.die("background record missing")
+        yield* sessions.synthetic({
+          id: background.notificationID,
+          sessionID,
+          text: "Command already completed",
+          metadata: { source: "shell", shellID: "sh_notified", state: "completed" },
+          resume: false,
+        })
+        if (delivered) yield* SessionInbox.promote(database.db, bus, sessionID, "steer")
+
+        const scope = yield* Scope.make()
+        yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+        const restarted = yield* Job.make.pipe(Effect.provideService(Scope.Scope, scope))
+        const context = yield* buildExecution(scope, () => Effect.void, undefined, restarted)
+        yield* Context.get(context, SessionRestart.Service).resumeSuspendedSessions
+
+        expect(yield* restarted.pendingBackground).toEqual([])
+        expect(yield* SessionInbox.list(database.db, sessionID)).toHaveLength(delivered ? 0 : 1)
+        yield* SessionInbox.promote(database.db, bus, sessionID, "steer")
+        expect(yield* sessions.messages({ sessionID })).toMatchObject([
+          {
+            id: background.notificationID,
+            type: "synthetic",
+            text: "Command already completed",
+            metadata: { state: "completed" },
+          },
+        ])
+        expect(yield* sessions.messages({ sessionID })).toHaveLength(1)
+      }),
+    )
+  }
+
+  it.effect("acknowledges recovery markers when their owning session is deleted", () =>
+    Effect.gen(function* () {
+      const database = yield* Database.Service
+      const jobs = yield* Job.Service
+      const sessionID = Session.ID.make("ses_background_deleted")
+      yield* seedSessions(database, [sessionID])
+      yield* seedBackground(jobs, sessionID, [{ id: "call-deleted-shell", shellID: "sh_deleted", command: "sleep 60" }])
+      yield* database.db.delete(SessionTable).where(eq(SessionTable.id, sessionID)).run().pipe(Effect.orDie)
+
+      const scope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+      const restarted = yield* Job.make.pipe(Effect.provideService(Scope.Scope, scope))
+      const context = yield* buildExecution(scope, () => Effect.void, undefined, restarted)
+      yield* Context.get(context, SessionRestart.Service).resumeSuspendedSessions
+
+      expect(yield* restarted.pendingBackground).toEqual([])
+    }),
+  )
+
+  it.effect("delivers cancellation at the resumed parent's next step", () =>
+    Effect.gen(function* () {
+      const database = yield* Database.Service
+      const jobs = yield* Job.Service
+      const store = yield* SessionStore.Service
+      const bus = yield* Bus.Service
+      const parent = Session.ID.make("ses_background_claimed_parent")
+      yield* seedSessions(database, [parent], { time_suspended: Date.now() })
+      yield* seedBackground(jobs, parent, [{ id: "call-claimed-shell", shellID: "sh_claimed", command: "sleep 60" }])
+
+      const observed = yield* Deferred.make<string[]>()
+      const scope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+      const restarted = yield* Job.make.pipe(Effect.provideService(Scope.Scope, scope))
+      const context = yield* buildExecution(
+        scope,
+        ({ sessionID }) =>
+          SessionInbox.promote(database.db, bus, sessionID, "steer").pipe(
+            Effect.andThen(store.context(sessionID)),
+            Effect.orDie,
+            Effect.flatMap((messages) =>
+              Deferred.succeed(
+                observed,
+                messages.filter((message) => message.type === "synthetic").map((message) => message.text),
+              ),
+            ),
+            Effect.asVoid,
+          ),
+        undefined,
+        restarted,
+      )
+      const execution = Context.get(context, SessionExecution.Service)
+      yield* Context.get(context, SessionRestart.Service).resumeSuspendedSessions
+      expect(yield* Deferred.await(observed)).toEqual([
+        "The server restarted while you were working. Continue from where you left off without repeating completed work.",
+        expect.stringContaining("Command cancelled because the server restarted"),
+      ])
+      yield* execution.awaitIdle(parent)
+      expect(yield* SessionInbox.list(database.db, parent)).toEqual([])
+      expect((yield* claims(database))[parent]).toBe(false)
+    }),
+  )
+
+  it.effect("resumes a background subagent and notifies its parent exactly once", () =>
+    Effect.gen(function* () {
+      const database = yield* Database.Service
+      const jobs = yield* Job.Service
+      const parent = Session.ID.make("ses_subagent_recovery_parent")
+      const child = Session.ID.make("ses_subagent_recovery_child")
+      const unrelated = Session.ID.make("ses_subagent_unrelated_child")
+      yield* seedSessions(database, [parent], { time_suspended: Date.now(), resume_attempts: 1 })
+      yield* seedSessions(database, [child, unrelated], { parent_id: parent, time_suspended: Date.now() })
+      yield* jobs.start({
+        id: child,
+        type: "subagent",
+        recovery: {
+          kind: "subagent",
+          parentSessionID: parent,
+          childSessionID: child,
+          agent: "explore",
+          description: "Inspect recovery",
+        },
+        run: Effect.never,
+      })
+      yield* jobs.background(child)
+
+      const resumed = yield* Deferred.make<void>()
+      const release = yield* Deferred.make<void>()
+      const parentResumed = yield* Deferred.make<void>()
+      const parentWoken = yield* Deferred.make<void>()
+      const drained: Session.ID[] = []
+      const scope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+      const restarted = yield* Job.make.pipe(Effect.provideService(Scope.Scope, scope))
+      const context = yield* buildExecution(
+        scope,
+        ({ sessionID }) =>
+          Effect.gen(function* () {
+            drained.push(sessionID)
+            if (sessionID === child) {
+              yield* Deferred.succeed(resumed, undefined)
+              yield* Deferred.await(release)
+              return
+            }
+            yield* Deferred.succeed(
+              drained.filter((id) => id === parent).length === 1 ? parentResumed : parentWoken,
+              undefined,
+            )
+          }),
+        undefined,
+        restarted,
+      )
+      const restart = Context.get(context, SessionRestart.Service)
+      const execution = Context.get(context, SessionExecution.Service)
+      yield* restart.resumeSuspendedSessions
+      yield* Deferred.await(resumed)
+      yield* Deferred.await(parentResumed)
+      yield* execution.awaitIdle(parent)
+
+      yield* restart.resumeSuspendedSessions
+      expect(drained.toSorted()).toEqual([child, parent].toSorted())
+      expect(yield* claims(database)).toEqual({ [parent]: false, [child]: true, [unrelated]: false })
+      expect(yield* attempts(database, child)).toBe(1)
+      expect(yield* restarted.get(child)).toMatchObject({ status: "running" })
+
+      yield* Deferred.succeed(release, undefined)
+      yield* Deferred.await(parentWoken)
+      expect(drained.filter((id) => id === child)).toHaveLength(1)
+      expect(drained.filter((id) => id === parent)).toHaveLength(2)
+      expect(yield* SessionInbox.list(database.db, parent)).toMatchObject([
+        {
+          payload: {
+            description: "Inspect recovery",
+            metadata: { source: "subagent", childID: child, agent: "explore", state: "completed" },
+          },
+        },
+      ])
+      expect(yield* restarted.pendingBackground).toEqual([])
+      yield* restart.resumeSuspendedSessions
+      expect(yield* SessionInbox.list(database.db, parent)).toHaveLength(1)
+    }),
+  )
+
+  it.effect("delivers a subagent result persisted before restart without rerunning the child", () =>
+    Effect.gen(function* () {
+      const database = yield* Database.Service
+      const jobs = yield* Job.Service
+      const parent = Session.ID.make("ses_subagent_completed_parent")
+      const child = Session.ID.make("ses_subagent_completed_child")
+      yield* seedSessions(database, [parent])
+      yield* seedSessions(database, [child], { parent_id: parent })
+      const complete = yield* Deferred.make<string>()
+      yield* jobs.start({
+        id: child,
+        type: "subagent",
+        recovery: {
+          kind: "subagent",
+          parentSessionID: parent,
+          childSessionID: child,
+          agent: "explore",
+          description: "Completed inspection",
+        },
+        run: Deferred.await(complete),
+      })
+      yield* jobs.background(child)
+      yield* Deferred.succeed(complete, "Recovered result")
+      yield* jobs.wait({ id: child })
+
+      const parentWoken = yield* Deferred.make<void>()
+      const drained: Session.ID[] = []
+      const scope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+      const restarted = yield* Job.make.pipe(Effect.provideService(Scope.Scope, scope))
+      const context = yield* buildExecution(
+        scope,
+        ({ sessionID }) =>
+          Effect.sync(() => void drained.push(sessionID)).pipe(
+            Effect.andThen(Deferred.succeed(parentWoken, undefined)),
+          ),
+        undefined,
+        restarted,
+      )
+      yield* Context.get(context, SessionRestart.Service).resumeSuspendedSessions
+      yield* Deferred.await(parentWoken)
+
+      expect(drained).toEqual([parent])
+      expect(yield* SessionInbox.list(database.db, parent)).toMatchObject([
+        { payload: { text: expect.stringContaining("Recovered result"), metadata: { state: "completed" } } },
+      ])
+      expect(yield* restarted.pendingBackground).toEqual([])
+    }),
+  )
+
+  for (const resumeAttempts of [1, 2]) {
+    it.effect(`honors a suspended parent's restart budget after ${resumeAttempts} attempts before notifying it`, () =>
+      Effect.gen(function* () {
+        const database = yield* Database.Service
+        const bus = yield* Bus.Service
+        const jobs = yield* Job.Service
+        const parent = Session.ID.make("ses_subagent_budget_parent")
+        const children = [
+          Session.ID.make("ses_subagent_budget_child_1"),
+          Session.ID.make("ses_subagent_budget_child_2"),
+        ]
+        yield* seedSessions(database, [parent], { time_suspended: Date.now(), resume_attempts: resumeAttempts })
+        yield* seedSessions(database, children, { parent_id: parent })
+        const complete = yield* Deferred.make<string>()
+        for (const child of children) {
+          yield* jobs.start({
+            id: child,
+            type: "subagent",
+            recovery: {
+              kind: "subagent",
+              parentSessionID: parent,
+              childSessionID: child,
+              agent: "explore",
+              description: "Completed inspection",
+            },
+            run: Deferred.await(complete),
+          })
+          yield* jobs.background(child)
+        }
+        yield* Deferred.succeed(complete, "Recovered result")
+        yield* Effect.forEach(children, (id) => jobs.wait({ id }), { discard: true })
+
+        const draining = yield* Deferred.make<number | undefined>()
+        const release = yield* Deferred.make<void>()
+        const drained: Session.ID[] = []
+        const continued: Session.ID[] = []
+        yield* bus.project(SessionEvent.Synthetic, (event) =>
+          Effect.sync(() => void continued.push(event.data.sessionID)),
+        )
+        const scope = yield* Scope.make()
+        yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+        const restarted = yield* Job.make.pipe(Scope.provide(scope))
+        const context = yield* buildExecution(
+          scope,
+          ({ sessionID }) =>
+            Effect.gen(function* () {
+              drained.push(sessionID)
+              yield* Deferred.succeed(draining, yield* attempts(database, sessionID))
+              yield* Deferred.await(release)
+            }),
+          { maxAttempts: 2 },
+          restarted,
+        )
+        const restart = Context.get(context, SessionRestart.Service)
+        const execution = Context.get(context, SessionExecution.Service)
+        yield* restart.resumeSuspendedSessions
+
+        if (resumeAttempts < 2) {
+          expect(yield* Deferred.await(draining)).toBe(2)
+          expect(drained).toEqual([parent])
+          expect(continued).toEqual([parent])
+          yield* Deferred.succeed(release, undefined)
+          yield* execution.awaitIdle(parent)
+        }
+        if (resumeAttempts === 2) {
+          expect(drained).toEqual([])
+          expect(continued).toEqual([])
+        }
+        expect((yield* claims(database))[parent]).toBe(false)
+        expect(yield* attempts(database, parent)).toBe(0)
+        expect(yield* SessionInbox.list(database.db, parent)).toHaveLength(2)
+        expect(yield* restarted.pendingBackground).toEqual([])
+        yield* restart.resumeSuspendedSessions
+        expect(drained).toHaveLength(resumeAttempts < 2 ? 1 : 0)
+      }),
+    )
+  }
+
+  it.effect("terminalizes a recovered subagent that exhausts its resume budget", () =>
+    Effect.gen(function* () {
+      const database = yield* Database.Service
+      const jobs = yield* Job.Service
+      const parent = Session.ID.make("ses_subagent_exhausted_parent")
+      const child = Session.ID.make("ses_subagent_exhausted_child")
+      yield* seedSessions(database, [parent])
+      yield* seedSessions(database, [child], { parent_id: parent, time_suspended: Date.now(), resume_attempts: 2 })
+      yield* jobs.start({
+        id: child,
+        type: "subagent",
+        recovery: {
+          kind: "subagent",
+          parentSessionID: parent,
+          childSessionID: child,
+          agent: "explore",
+          description: "Exhausted inspection",
+        },
+        run: Effect.never,
+      })
+      yield* jobs.background(child)
+
+      const parentWoken = yield* Deferred.make<void>()
+      const drained: Session.ID[] = []
+      const scope = yield* Scope.make()
+      yield* Effect.addFinalizer(() => Scope.close(scope, Exit.void))
+      const restarted = yield* Job.make.pipe(Effect.provideService(Scope.Scope, scope))
+      const context = yield* buildExecution(
+        scope,
+        ({ sessionID }) =>
+          Effect.sync(() => void drained.push(sessionID)).pipe(
+            Effect.andThen(Deferred.succeed(parentWoken, undefined)),
+          ),
+        { maxAttempts: 2 },
+        restarted,
+      )
+      yield* Context.get(context, SessionRestart.Service).resumeSuspendedSessions
+      yield* Deferred.await(parentWoken)
+
+      expect(drained).toEqual([parent])
+      expect((yield* claims(database))[child]).toBe(false)
+      expect(yield* SessionInbox.list(database.db, parent)).toMatchObject([
+        {
+          payload: {
+            text: expect.stringContaining("will not be resumed automatically"),
+            metadata: { source: "subagent", childID: child, state: "error" },
+          },
+        },
+      ])
+      expect(yield* restarted.pendingBackground).toEqual([])
+    }),
+  )
+})
+
 describe("SessionExecution interrupt continuation", () => {
   it.effect("resumes only steering input after an interrupt with continue", () =>
     Effect.gen(function* () {
@@ -446,6 +1021,27 @@ describe("SessionExecution interrupt continuation", () => {
   )
 })
 
+function seedBackground(
+  jobs: Job.Interface,
+  sessionID: Session.ID,
+  background: ReadonlyArray<{ readonly id: string; readonly shellID: string; readonly command: string }>,
+) {
+  return Effect.forEach(
+    background,
+    (job) =>
+      Effect.gen(function* () {
+        yield* jobs.start({
+          id: job.id,
+          type: "shell",
+          recovery: { kind: "shell", sessionID, shellID: job.shellID, command: job.command },
+          run: Effect.never,
+        })
+        yield* jobs.background(job.id)
+      }),
+    { discard: true },
+  )
+}
+
 /** Plain deliveries seed user prompts; objects seed control items. */
 function seedInbox(
   database: Database.Service["Service"],
@@ -531,11 +1127,27 @@ function buildExecution(
   scope: Scope.Closeable,
   drain: (input: Parameters<SessionRunner.Interface["drain"]>[0]) => Effect.Effect<void, SessionRunner.RunError>,
   options?: SessionRestart.Options,
+  overrideJobs?: Job.Interface,
 ) {
   return Effect.gen(function* () {
     const database = yield* Database.Service
     const bus = yield* Bus.Service
     const store = yield* SessionStore.Service
+    const jobs = overrideJobs ?? (yield* Job.Service)
+    const sessions = yield* Session.Service
+    const sessionLayer = Layer.effect(
+      Session.Service,
+      Effect.gen(function* () {
+        const execution = yield* SessionExecution.Service
+        return Session.Service.of({
+          ...sessions,
+          synthetic: (input) =>
+            sessions
+              .synthetic({ ...input, resume: false })
+              .pipe(Effect.tap(() => (input.resume === false ? Effect.void : execution.wake(input.sessionID)))),
+        })
+      }),
+    )
     const runner = Layer.succeed(
       SessionRunner.Service,
       SessionRunner.Service.of({
@@ -553,10 +1165,12 @@ function buildExecution(
     )
     return yield* Layer.buildWithScope(
       SessionRestart.layer(options).pipe(
-        Layer.provideMerge(SessionExecution.layer),
+        Layer.provideMerge(sessionLayer),
+        Layer.provideMerge(Layer.fresh(SessionExecution.layer)),
         Layer.provide(Layer.succeed(Database.Service, database)),
         Layer.provide(Layer.succeed(Bus.Service, bus)),
         Layer.provide(Layer.succeed(SessionStore.Service, store)),
+        Layer.provide(Layer.succeed(Job.Service, jobs)),
         Layer.provide(locations),
       ),
       scope,

packages/core/test/session-runner-tool-events.test.ts

@@ -128,6 +128,35 @@ test("interrupted progress metadata remains in the terminal failure snapshot", a
   })
 })
 
+test("interrupted subagent failures expose their existing child session to the model", async () => {
+  const { published, publisher } = capture("anthropic", { interruptProgress: true })
+  const subagent = LLMEvent.toolCall({
+    id: "call-subagent",
+    name: "subagent",
+    input: { agent: "general", description: "Recover child", prompt: "Continue working" },
+  })
+  await Effect.runPromise(publisher.publish(subagent))
+  await Effect.runPromiseExit(publisher.progress(subagent.id, { sessionID: "ses_existing_child", status: "running" }))
+  await Effect.runPromise(publisher.failUnsettledTools({ type: "aborted", message: "Tool execution interrupted" }))
+
+  expect(published.find((event) => event.type === "session.tool.failed.2")?.data).toMatchObject({
+    error: { type: "aborted", message: "Tool execution interrupted (sessionID: ses_existing_child)" },
+    metadata: { sessionID: "ses_existing_child", status: "running" },
+  })
+})
+
+test("interrupted non-subagent failures do not expose their progress session IDs", async () => {
+  const { published, publisher } = capture()
+  await Effect.runPromise(publisher.publish(call))
+  await Effect.runPromise(publisher.progress(call.id, { sessionID: "ses_private", status: "running" }))
+  await Effect.runPromise(publisher.failUnsettledTools({ type: "aborted", message: "Tool execution interrupted" }))
+
+  expect(published.find((event) => event.type === "session.tool.failed.2")?.data).toMatchObject({
+    error: { type: "aborted", message: "Tool execution interrupted" },
+    metadata: { sessionID: "ses_private", status: "running" },
+  })
+})
+
 test("local failure metadata completes the progress snapshot", async () => {
   const { published, publisher } = capture()
   await Effect.runPromise(publisher.publish(call))

packages/core/test/session-runner-tool-registry.test.ts

@@ -7,10 +7,11 @@ import { PluginHooks } from "@opencode-ai/core/plugin/hooks"
 import { Session } from "@opencode-ai/core/session"
 import { SessionMessage } from "@opencode-ai/core/session/message"
 import { Tool } from "@opencode-ai/core/tool"
+import { State } from "@opencode-ai/core/state"
 import type { Info } from "@opencode-ai/schema/tool"
 import { LayerNode } from "@opencode-ai/util/effect/layer-node"
 import { executeTool, toolDefinitions } from "./lib/tool"
-import { Cause, Deferred, Effect, Exit, Fiber, Layer, Schema, SchemaGetter, SchemaIssue, Scope } from "effect"
+import { Cause, Deferred, Effect, Exit, Fiber, Layer, Logger, Schema, SchemaGetter, SchemaIssue, Scope } from "effect"
 import { z } from "zod"
 import { testEffect } from "./lib/effect"
 
@@ -71,30 +72,181 @@ const transform = (service: Tool.Interface, tools: Readonly<Record<string, Info>
   )
 
 describe("Tool", () => {
-  it.effect("rejects invalid dotted namespaces", () =>
+  it.live("replays updates and removals on reload and restores definitions on disposal", () =>
     Effect.gen(function* () {
       const service = yield* Tool.Service
-      const error = yield* transform(service, { echo: make() }, { namespace: "slack..admin" }).pipe(Effect.flip)
+      let text = "original"
+      const source = yield* service.transform((draft) =>
+        draft.add({
+          ...constant(text),
+          name: "echo",
+          description: text,
+          options: { namespace: "acme", codemode: false },
+        }),
+      )
+      yield* transform(service, { echo: make() }, { namespace: "other", codemode: false })
+      const before = yield* service.snapshot()
+      const update = yield* service.transform((draft) => {
+        draft.update("missing", () => {
+          throw new Error("must not create a tool")
+        })
+        draft.update("acme_echo", (tool) => {
+          tool.description += " updated"
+          const execute = tool.execute
+          tool.execute = (input, context) =>
+            execute(input, context).pipe(
+              Effect.map((result) => ({ ...result, output: { text: `${result.output.text} updated` } })),
+            )
+        })
+      })
+      const removal = yield* service.transform((draft) => {
+        draft.remove("missing")
+        draft.remove("other_echo")
+      })
+      const updated = yield* service.snapshot()
+      expect(updated.definitions.map((tool) => tool.name)).toEqual(["acme_echo", "execute"])
+      expect(updated.definitions[0]?.description).toBe("original updated")
+      expect((yield* updated.execute(call("acme_echo"))).output).toEqual({ text: "original updated" })
+      text = "refreshed"
+      yield* service.reload()
+      const reloaded = yield* service.snapshot()
+      expect(reloaded.definitions.map((tool) => tool.name)).toEqual(["acme_echo", "execute"])
+      expect(reloaded.definitions[0]?.description).toBe("refreshed updated")
+      expect((yield* reloaded.execute(call("acme_echo"))).output).toEqual({ text: "refreshed updated" })
+      expect((yield* before.execute(call("acme_echo"))).output).toEqual({ text: "original" })
+      yield* removal.dispose
+      yield* removal.dispose
+      yield* update.dispose
+      const restored = yield* service.snapshot()
+      expect(restored.definitions.map((tool) => tool.name)).toEqual(["acme_echo", "other_echo", "execute"])
+      expect((yield* restored.execute(call("acme_echo"))).output).toEqual({ text: "refreshed" })
+      yield* source.dispose
+      expect((yield* service.snapshot()).definitions.map((tool) => tool.name)).toEqual(["other_echo", "execute"])
+    }),
+  )
 
-      expect(error).toBeInstanceOf(Tool.RegistrationError)
-      expect(error.message).toBe('Invalid tool namespace: "slack..admin"')
+  it.effect("does not retain an updated tool after its source scope closes", () =>
+    Effect.gen(function* () {
+      const service = yield* Tool.Service
+      const scope = yield* Scope.make()
+      yield* transform(service, { echo: make() }, { codemode: false }).pipe(Scope.provide(scope))
+      yield* service.transform((draft) =>
+        draft.update("echo", (tool) => {
+          tool.description = "Updated"
+        }),
+      )
+      yield* Scope.close(scope, Exit.void)
       expect((yield* service.snapshot()).definitions.map((tool) => tool.name)).toEqual(["execute"])
     }),
   )
 
-  it.effect("rejects invalid and colliding normalized names", () =>
+  it.effect("batches tool transforms with the shared state lifecycle", () =>
     Effect.gen(function* () {
       const service = yield* Tool.Service
-      for (const name of ["", "x".repeat(65)]) {
-        const invalid = yield* transform(service, { [name]: make() }, { codemode: false }).pipe(Effect.flip)
-        expect(invalid.message).toBe(`Invalid tool name: ${name}`)
-      }
-
-      const collision = yield* transform(service, { "echo.tool": make(), echo_tool: make() }, { codemode: false }).pipe(
-        Effect.flip,
+      let runs = 0
+      yield* State.batch(
+        Effect.gen(function* () {
+          yield* service.transform((draft) => {
+            runs++
+            draft.add({ ...make(), options: { codemode: false } })
+          })
+          yield* service.transform((draft) =>
+            draft.update("echo", (tool) => {
+              tool.description = "Batched"
+            }),
+          )
+          expect(runs).toBe(0)
+        }),
       )
-      expect(collision.message).toBe("Duplicate normalized tool name: echo_tool")
-      expect((yield* service.snapshot()).definitions.map((tool) => tool.name)).toEqual(["execute"])
+      expect(runs).toBe(1)
+      expect((yield* service.snapshot()).definitions[0]?.description).toBe("Batched")
+    }),
+  )
+
+  it.effect("skips invalid updates without dropping the existing definition", () =>
+    Effect.gen(function* () {
+      const service = yield* Tool.Service
+      yield* transform(service, { echo: make() }, { codemode: false })
+      yield* service.transform((draft) =>
+        draft.update("echo", (tool) => {
+          Object.assign(tool, { description: undefined })
+        }),
+      )
+      expect((yield* service.snapshot()).definitions[0]?.description).toBe("Echo text")
+    }),
+  )
+
+  it.effect("updates newly added tools and applies removals in order", () =>
+    Effect.gen(function* () {
+      const service = yield* Tool.Service
+      yield* service.transform((draft) => {
+        draft.add({ ...make(), options: { codemode: false } })
+        draft.update("echo", (tool) => {
+          tool.description = "Updated"
+          tool.input = Schema.Struct({ value: Schema.Number })
+          tool.output = Schema.Number
+          tool.execute = ({ value }) => Effect.succeed({ output: value * 2 })
+        })
+        draft.add({ ...make(), name: "removed" })
+        draft.remove("removed")
+        draft.add({ ...make(), name: "removed" })
+        draft.remove("removed")
+      })
+      const snapshot = yield* service.snapshot()
+      expect(snapshot.definitions.map((tool) => tool.name)).toEqual(["echo", "execute"])
+      expect(snapshot.definitions[0]?.description).toBe("Updated")
+      expect(
+        (yield* snapshot.execute({
+          ...call("echo"),
+          call: { type: "tool-call", id: "updated", name: "echo", input: { value: 3 } },
+        })).output,
+      ).toBe(6)
+    }),
+  )
+
+  it.effect("logs and skips invalid dotted namespaces", () => {
+    const output: unknown[] = []
+    const logger = Logger.map(Logger.formatStructured, (entry) => {
+      output.push(entry.message)
+    })
+    return Effect.gen(function* () {
+      const service = yield* Tool.Service
+      yield* transform(service, { echo: make() }, { namespace: "slack..admin" })
+
+      expect(output).toEqual([
+        [
+          "Skipping invalid tool registration",
+          { name: "echo", namespace: "slack..admin", error: 'Invalid tool namespace: "slack..admin"' },
+        ],
+      ])
+      const snapshot = yield* service.snapshot()
+      expect(snapshot.definitions.map((tool) => tool.name)).toEqual(["execute"])
+      expect(snapshot.codeModeCatalog).toEqual([])
+    }).pipe(Effect.provide(Logger.layer([logger])))
+  })
+
+  it.effect("skips invalid, reserved, and colliding names without dropping healthy tools", () =>
+    Effect.gen(function* () {
+      const service = yield* Tool.Service
+      yield* transform(
+        service,
+        {
+          before: make(),
+          "": make(),
+          ["x".repeat(65)]: make(),
+          "echo.tool": make(),
+          echo_tool: make(),
+          execute: make(),
+          after: make(),
+        },
+        { codemode: false },
+      )
+      const snapshot = yield* service.snapshot()
+      expect(snapshot.definitions.map((tool) => tool.name)).toEqual(["after", "before", "execute"])
+      expect((yield* snapshot.execute(call("before"))).output).toEqual({ text: "before" })
+      expect((yield* snapshot.execute(call("after"))).output).toEqual({ text: "after" })
+      expect((yield* snapshot.execute(call("echo_tool")).pipe(Effect.flip)).message).toBe("Unknown tool: echo_tool")
+      expect(snapshot.codeModeCatalog).toEqual([])
     }),
   )
 
@@ -159,40 +311,72 @@ describe("Tool", () => {
     }),
   )
 
-  it.effect("validates a registration batch before installing any tools", () =>
+  it.effect("keeps healthy tools when another namespace is invalid", () =>
     Effect.gen(function* () {
       const service = yield* Tool.Service
-      const error = yield* service
-        .transform((draft) => {
-          draft.add({ ...make(), name: "first", options: { codemode: false } })
-          draft.add({ ...make(), name: "second", options: { namespace: "invalid..namespace", codemode: false } })
-        })
-        .pipe(Effect.flip)
+      yield* service.transform((draft) => {
+        draft.add({ ...make(), name: "first", options: { codemode: false } })
+        draft.add({ ...make(), name: "second", options: { namespace: "invalid..namespace", codemode: false } })
+        draft.add({ ...make(), name: "second", options: { namespace: "invalid__namespace" } })
+      })
 
-      expect(error).toBeInstanceOf(Tool.RegistrationError)
-      expect((yield* service.snapshot()).definitions.map((tool) => tool.name)).toEqual(["execute"])
+      const snapshot = yield* service.snapshot()
+      expect(snapshot.definitions.map((tool) => tool.name)).toEqual(["first", "execute"])
+      expect(snapshot.codeModeCatalog?.map((tool) => tool.path)).toEqual(["invalid__namespace.second"])
     }),
   )
 
-  it.effect("rejects invalid tool definitions before installing any tools", () =>
+  it.effect("logs invalid tool definitions without dropping healthy tools", () => {
+    const output: unknown[] = []
+    const logger = Logger.map(Logger.formatStructured, (entry) => {
+      output.push(entry.message)
+    })
+    return Effect.gen(function* () {
+      const service = yield* Tool.Service
+      yield* service.transform((draft) => {
+        draft.add({ ...make(), name: "healthy", options: { codemode: false } })
+        draft.add({
+          name: "phone_type",
+          input: Schema.Struct({}),
+          execute: () => Effect.succeed({ content: "ok" }),
+          options: { codemode: false },
+        } as unknown as Info)
+        draft.add({ ...make(), name: "codemode" })
+      })
+
+      expect(output).toEqual([
+        [
+          "Skipping invalid tool registration",
+          {
+            name: "phone_type",
+            namespace: undefined,
+            error: expect.stringContaining('Expected string\n  at ["description"]'),
+          },
+        ],
+      ])
+      const snapshot = yield* service.snapshot()
+      expect(snapshot.definitions.map((tool) => tool.name)).toEqual(["healthy", "execute"])
+      expect(snapshot.codeModeCatalog?.map((tool) => tool.path)).toEqual(["codemode"])
+      expect((yield* snapshot.execute(call("phone_type")).pipe(Effect.flip)).message).toBe("Unknown tool: phone_type")
+    }).pipe(Effect.provide(Logger.layer([logger])))
+  })
+
+  it.effect("skipped registrations leave existing tools and scoped cleanup intact", () =>
     Effect.gen(function* () {
       const service = yield* Tool.Service
-      const error = yield* service
-        .transform((draft) => {
-          draft.add({ ...make(), name: "healthy", options: { codemode: false } })
-          draft.add({
-            name: "phone_type",
-            input: Schema.Struct({}),
-            execute: () => Effect.succeed({ content: "ok" }),
-            options: { codemode: false },
-          } as unknown as Info)
-        })
-        .pipe(Effect.flip)
-
-      expect(error).toBeInstanceOf(Tool.RegistrationError)
-      expect(error.name).toBe("phone_type")
-      expect(error.message).toContain('Expected string\n  at ["description"]')
-      expect((yield* service.snapshot()).definitions.map((tool) => tool.name)).toEqual(["execute"])
+      yield* transform(service, { echo: constant("original") }, { codemode: false })
+      yield* Effect.scoped(
+        Effect.gen(function* () {
+          yield* service.transform((draft) => {
+            draft.add({ ...constant("invalid"), name: "echo", description: undefined } as unknown as Info)
+            draft.add({ ...make(), name: "temporary", options: { codemode: false } })
+          })
+          const snapshot = yield* service.snapshot()
+          expect(snapshot.definitions.map((tool) => tool.name)).toEqual(["echo", "temporary", "execute"])
+          expect((yield* snapshot.execute(call("echo"))).output).toEqual({ text: "original" })
+        }),
+      )
+      expect((yield* service.snapshot()).definitions.map((tool) => tool.name)).toEqual(["echo", "execute"])
     }),
   )
 

packages/core/test/session-runner.test.ts

@@ -79,7 +79,7 @@ import { Provider } from "@opencode-ai/core/provider"
 import { Cause, Deferred, Effect, Exit, Fiber, Layer, Queue, Schema, Scope, Stream } from "effect"
 import { TestClock } from "effect/testing"
 import { HttpClientRequest, HttpClientResponse } from "effect/unstable/http"
-import { asc, desc, eq } from "drizzle-orm"
+import { asc, desc, eq, sql } from "drizzle-orm"
 import { testEffect } from "./lib/effect"
 import { permissionLayer } from "./lib/permission"
 import { agentHost, catalogHost, host } from "./plugin/host"
@@ -1444,6 +1444,49 @@ describe("SessionRunnerLLM", () => {
     }),
   )
 
+  it.effect("delivers controls without preflighting unavailable initial instructions", () =>
+    Effect.gen(function* () {
+      const session = yield* setup
+      const database = yield* Database.Service
+      const bus = yield* Bus.Service
+      const runner = yield* SessionRunner.Service
+      systemUnavailable = true
+      let reads = 0
+      systemLoadHook = Effect.sync(() => {
+        reads++
+      })
+      const compaction = yield* SessionInbox.admitCompaction(database.db, bus, {
+        id: SessionMessage.ID.create(),
+        sessionID,
+        delivery: "queue",
+      })
+      yield* SessionInbox.admit(database.db, bus, {
+        id: SessionMessage.ID.create(),
+        sessionID,
+        item: {
+          type: "move",
+          payload: {
+            location: Location.Ref.make({ directory: AbsolutePath.make("/moved") }),
+            projectID: Project.ID.global,
+          },
+          delivery: "queue",
+        },
+      })
+
+      expect(yield* runner.drain({ sessionID, force: false })).toEqual(SessionRunner.DrainResult.Moved({}))
+
+      expect(reads).toBe(0)
+      expect(requests).toHaveLength(0)
+      expect(yield* session.inbox(sessionID)).toEqual([])
+      expect((yield* session.get(sessionID)).location.directory).toBe(AbsolutePath.make("/moved"))
+      expect((yield* session.messages({ sessionID })).find((message) => message.id === compaction.id)).toMatchObject({
+        type: "compaction",
+        status: "failed",
+        error: { type: "compaction.unavailable", message: "Nothing to compact yet" },
+      })
+    }),
+  )
+
   it.effect("delivers a queued move atomically at the idle boundary", () =>
     Effect.gen(function* () {
       const session = yield* setup
@@ -1554,6 +1597,56 @@ describe("SessionRunnerLLM", () => {
     }),
   )
 
+  it.effect("runs a queued control on Location entry before a carried continuation", () =>
+    Effect.gen(function* () {
+      const session = yield* setup
+      const database = yield* Database.Service
+      const bus = yield* Bus.Service
+      const runner = yield* SessionRunner.Service
+      yield* admit(session, "Echo before moving")
+      yield* TestLLM.push(
+        TestLLM.tool("call-entry", "echo", { text: "moving" }),
+        TestLLM.text("Entry summary", "entry-summary"),
+        TestLLM.text("Continued", "entry-continuation"),
+      )
+      const stream = yield* TestLLM.gate
+      const run = yield* runner.drain({ sessionID, force: false }).pipe(Effect.forkChild)
+      yield* stream.started
+      const compaction = yield* SessionInbox.admitCompaction(database.db, bus, {
+        id: SessionMessage.ID.create(),
+        sessionID,
+        delivery: "queue",
+      })
+      yield* SessionInbox.admit(database.db, bus, {
+        id: SessionMessage.ID.create(),
+        sessionID,
+        item: {
+          type: "move",
+          payload: {
+            location: Location.Ref.make({ directory: AbsolutePath.make("/project") }),
+            projectID: Project.ID.global,
+          },
+          delivery: "steer",
+        },
+      })
+      yield* stream.release
+      const moved = yield* Fiber.join(run)
+
+      expect(moved).toEqual(SessionRunner.DrainResult.Moved({ continuation: { step: 2 } }))
+      expect(requests).toHaveLength(1)
+      expect(yield* SessionInbox.find(database.db, compaction.id)).toMatchObject({ id: compaction.id })
+      if (moved._tag !== "Moved") throw new Error("Expected a Location handoff")
+
+      // Location entry considers queued controls even when model work carries across the move.
+      yield* runner.drain({ sessionID, force: false, continuation: moved.continuation })
+
+      expect(requests).toHaveLength(3)
+      expect(userTexts(requests[1])[0]).toContain("Create a new anchored summary")
+      expect(userTexts(requests[2])[0]).toContain("<summary>\nEntry summary\n</summary>")
+      expect(yield* session.inbox(sessionID)).toEqual([])
+    }),
+  )
+
   it.effect("seeds a fork with the parent's newest instruction values", () =>
     Effect.gen(function* () {
       const session = yield* setup
@@ -2562,6 +2655,74 @@ describe("SessionRunnerLLM", () => {
     }),
   )
 
+  it.effect("refreshes preparation after overflow compaction without promoting new input", () =>
+    Effect.gen(function* () {
+      const session = yield* setupOverflowRecovery
+      const bus = yield* Bus.Service
+      let reads = 0
+      let resolutions = 0
+      systemLoadHook = Effect.sync(() => {
+        reads++
+      })
+      modelResolveHook = Effect.sync(() => {
+        resolutions++
+      })
+      yield* admit(session, "Continue")
+      yield* TestLLM.push(
+        [LLMEvent.providerError({ message: "prompt too long", classification: "context-overflow" })],
+        TestLLM.text("Overflow summary", "overflow-summary"),
+        TestLLM.text("Recovered", "overflow-recovered"),
+        TestLLM.stop(),
+        TestLLM.stop(),
+      )
+      const first = yield* TestLLM.gate
+      const run = yield* session.resume(sessionID).pipe(Effect.forkChild)
+      yield* first.started
+      expect(reads).toBe(1)
+      expect(resolutions).toBe(1)
+      expect(requests[0]?.model).toBe(recoveryModel)
+
+      const summary = yield* TestLLM.gate
+      yield* first.release
+      yield* summary.started
+      systemBaseline = "Changed during compaction"
+      yield* bus.publish(SessionEvent.ModelSelected, {
+        sessionID,
+        model: { id: ID.make("replacement"), providerID: Provider.ID.make("fake") },
+      })
+      const queued = yield* session.prompt({
+        sessionID,
+        text: "Queued during compaction",
+        delivery: "queue",
+        resume: false,
+      })
+      const steered = yield* admit(session, "Steered during compaction")
+      const retry = yield* TestLLM.gate
+      yield* summary.release
+      yield* retry.started
+
+      expect(reads).toBe(2)
+      expect(resolutions).toBe(2)
+      expect(requests).toHaveLength(3)
+      expect(requests[2]?.model).toBe(replacementModel)
+      expect(requests[2]?.system.map((part) => part.text)).toEqual([defaultSystem, "Initial context"])
+      expect(systemTexts(requests[2])).toContain("Changed during compaction")
+      expect(userTexts(requests[2])[0]).toContain("<summary>\nOverflow summary\n</summary>")
+      expect(userTexts(requests[2]).join("\n")).not.toContain("Queued during compaction")
+      expect(userTexts(requests[2]).join("\n")).not.toContain("Steered during compaction")
+      expect((yield* session.inbox(sessionID)).map((item) => item.id)).toEqual([queued.id, steered.id])
+
+      yield* retry.release
+      yield* Fiber.join(run)
+
+      expect(requests).toHaveLength(5)
+      expect(userTexts(requests[3])).toContain("Steered during compaction")
+      expect(userTexts(requests[3])).not.toContain("Queued during compaction")
+      expect(userTexts(requests[4])).toContain("Queued during compaction")
+      expect(yield* session.inbox(sessionID)).toEqual([])
+    }),
+  )
+
   it.effect("does not recover provider context overflow when automatic compaction is disabled", () =>
     Effect.gen(function* () {
       const session = yield* setupOverflowRecovery
@@ -3320,6 +3481,71 @@ describe("SessionRunnerLLM", () => {
     }),
   )
 
+  it.effect("keeps queued input parked when a steer is cancelled during preparation", () =>
+    Effect.gen(function* () {
+      const session = yield* setup
+      const runner = yield* SessionRunner.Service
+      yield* admit(session, "A")
+      yield* TestLLM.push(TestLLM.stop(), TestLLM.stop(), TestLLM.stop())
+      const stream = yield* TestLLM.gate
+      const run = yield* runner.drain({ sessionID, force: false }).pipe(Effect.forkChild)
+      yield* stream.started
+
+      yield* session.prompt({ sessionID, text: "B", delivery: "queue", resume: false })
+      const steer = yield* admit(session, "S")
+      systemLoadHook = Effect.gen(function* () {
+        systemLoadHook = Effect.void
+        yield* session.cancelInbox({ sessionID, inboxID: steer.id }).pipe(Effect.orDie)
+      })
+      yield* stream.release
+      yield* Fiber.join(run)
+
+      expect(requests.map(userTexts)).toEqual([["A"], ["A"], ["A", "B"]])
+      expect((yield* session.messages({ sessionID })).some((message) => message.id === steer.id)).toBe(false)
+      expect(yield* session.inbox(sessionID)).toEqual([])
+    }),
+  )
+
+  it.effect("dispatches a queued move when a steer is cancelled during preparation", () =>
+    Effect.gen(function* () {
+      const session = yield* setup
+      const runner = yield* SessionRunner.Service
+      const database = yield* Database.Service
+      const bus = yield* Bus.Service
+      const location = Location.Ref.make({ directory: AbsolutePath.make("/moved") })
+      yield* admit(session, "A")
+      yield* TestLLM.push(TestLLM.stop(), TestLLM.stop())
+      const stream = yield* TestLLM.gate
+      const run = yield* runner.drain({ sessionID, force: false }).pipe(Effect.forkChild)
+      yield* stream.started
+
+      yield* SessionInbox.admit(database.db, bus, {
+        id: SessionMessage.ID.create(),
+        sessionID,
+        item: {
+          type: "move",
+          payload: { location, projectID: Project.ID.global },
+          delivery: "queue",
+        },
+      })
+      const steer = yield* admit(session, "S")
+      systemLoadHook = Effect.gen(function* () {
+        systemLoadHook = Effect.void
+        yield* session.cancelInbox({ sessionID, inboxID: steer.id }).pipe(Effect.orDie)
+      })
+      yield* stream.release
+
+      expect({ result: yield* Fiber.join(run), location: (yield* session.get(sessionID)).location }).toEqual({
+        result: SessionRunner.DrainResult.Moved({}),
+        location,
+      })
+      expect(requests.map(userTexts)).toEqual([["A"], ["A"]])
+      expect(closedTransports).toEqual([sessionID])
+      expect(yield* recordedEventTypes(sessionID)).toContain(Bus.versionedType(SessionEvent.Moved.type, 1))
+      expect(yield* session.inbox(sessionID)).toEqual([])
+    }),
+  )
+
   it.effect("preserves durable queued input for a later wake after interruption", () =>
     Effect.gen(function* () {
       const session = yield* setup
@@ -3640,6 +3866,81 @@ describe("SessionRunnerLLM", () => {
     }),
   )
 
+  it.effect("preserves a stale subagent child session in its model-visible failure", () =>
+    Effect.gen(function* () {
+      const session = yield* setup
+      const bus = yield* Bus.Service
+      const database = yield* Database.Service
+      yield* admit(session, "Recover interrupted subagent")
+      yield* SessionInbox.promote(database.db, bus, sessionID, "steer")
+      const assistantMessageID = SessionMessage.ID.create()
+      yield* bus.publish(SessionEvent.Step.Started, {
+        sessionID,
+        assistantMessageID,
+        agent: Agent.ID.make("build"),
+        model: { id: ID.make("fake-model"), providerID: Provider.ID.make("fake") },
+      })
+      yield* bus.publish(SessionEvent.Tool.Input.Started, {
+        sessionID,
+        assistantMessageID,
+        id: "call-interrupted-subagent",
+        name: "subagent",
+      })
+      yield* bus.publish(SessionEvent.Tool.Input.Ended, {
+        sessionID,
+        assistantMessageID,
+        id: "call-interrupted-subagent",
+        text: '{"agent":"general"}',
+      })
+      yield* bus.publish(SessionEvent.Tool.Called, {
+        sessionID,
+        assistantMessageID,
+        id: "call-interrupted-subagent",
+        input: { agent: "general" },
+        executed: false,
+      })
+      yield* database.db
+        .update(SessionMessageTable)
+        .set({
+          data: sql`json_set(
+            ${SessionMessageTable.data},
+            '$.content[0].state.metadata',
+            json('{"sessionID":"ses_existing_child","status":"running","internal":"private"}')
+          )`,
+        })
+        .where(eq(SessionMessageTable.id, assistantMessageID))
+        .run()
+        .pipe(Effect.orDie)
+      requests.length = 0
+      yield* TestLLM.push([])
+      yield* session.resume(sessionID)
+
+      expect(yield* session.context(sessionID)).toMatchObject([
+        { type: "user", text: "Recover interrupted subagent" },
+        {
+          type: "assistant",
+          content: [
+            {
+              type: "tool",
+              id: "call-interrupted-subagent",
+              state: {
+                status: "error",
+                error: {
+                  type: "aborted",
+                  message: "Tool execution interrupted: subagent (sessionID: ses_existing_child)",
+                },
+                metadata: { sessionID: "ses_existing_child", status: "running", internal: "private" },
+              },
+            },
+          ],
+        },
+      ])
+      const modelResult = JSON.stringify(requests[0]?.messages.at(-1))
+      expect(modelResult).toContain("ses_existing_child")
+      expect(modelResult).not.toContain("private")
+    }),
+  )
+
   it.effect("durably fails hosted tools left running by a prior process before continuing inline", () =>
     Effect.gen(function* () {
       const session = yield* setup

packages/core/test/shell-parse-compatibility.test.ts

@@ -0,0 +1,234 @@
+import { describe, expect, test } from "bun:test"
+import { Effect } from "effect"
+import { ShellParse } from "../src/shell/parse.js"
+
+describe("portable shell parser compatibility", () => {
+  test.each([
+    ["bash", "echo $((1+1))", {}],
+    ["bash", "echo $((1 + $(printf hidden)))", {}],
+    ["bash", "cd ~/project", {}],
+    ["bash", "cd src&&cd..", {}],
+    ["bash", "cd src && cd .. && git status", {}],
+    ["zsh", "git status", {}],
+    ["fish", "git status", {}],
+    ["bash", "git status", { BASH_ENV: "/startup" }],
+    ["bash", "git status", { "BASH_FUNC_cd%%": "() { :; }" }],
+    ["bash", "cd $HOME; pwd", { HOME: "/session-home" }],
+    ["bash", "cd; pwd", { HOME: "/session-home" }],
+    ["bash", 'target=/outside; cd "$target"; pwd', {}],
+    ["bash", "cd 'src dir' && git status", {}],
+    ["bash", 'cd "src dir"; cd escaped\\ space', {}],
+    ["bash", "cd '$HOME'; cd '~/outside'", {}],
+    ["bash", 'g""it status', {}],
+    ["bash", 'npm "run" test', {}],
+    ["bash", "git '*'", {}],
+    ["bash", "FOO=bar git status", {}],
+    ["bash", "HOME=/outside; cd; pwd", {}],
+    ["bash", "CDPATH=/outside cd child", {}],
+    ["bash", "cd child; pwd", { CDPATH: "/outside" }],
+    ["bash", "cd /workspace > output", {}],
+    ["bash", "export X=value; unset X; git status", {}],
+    ["bash", "printf ok && git status > output", {}],
+    ["bash", "printf ok | cat < input > output", {}],
+    ["bash", "cd -- -/../../../etc; pwd", {}],
+    ["bash", "cd -; pushd; popd; pwd", {}],
+    ["bash", "command cd /outside; builtin cd /elsewhere", {}],
+    ["pwsh", "Get-ChildItem | ForEach-Object { Write-Output $_ }", {}],
+    ["pwsh", "ForEach-Object { Remove-Item victim }", {}],
+    ["pwsh", "Set-Location -LiteralPath '../outside'; Get-ChildItem", {}],
+    ["pwsh", "Set-Location -LiteralPath:/outside", {}],
+    ["pwsh", "Set-Location -LiteralPath:'/outside path'", {}],
+    ["pwsh", 'Set-Location -PATH:"../outside path"', {}],
+    ["pwsh", "Set-Location $HOME; Set-Location $PWD; Set-Location $target", { HOME: "/session-home" }],
+    ["pwsh", "Set-Item Env:T /outside; Set-Location $env:T", { T: "/workspace" }],
+    ["pwsh", "sl /outside; Microsoft.PowerShell.Management\\Set-Location /outside", {}],
+  ] as const)(
+    "matches supported legacy resources, saved prefixes, and directories natively: %s %s %j",
+    async (shell, command, env) => {
+      if (Object.keys(env).length > 0) {
+        const child = Bun.spawn({
+          cmd: [
+            process.execPath,
+            "--eval",
+            `
+          import { Effect } from "effect"
+          import { ShellParse } from "./src/shell/parse.ts"
+          const command = ${JSON.stringify(command)}
+          const shell = ${JSON.stringify(shell)}
+          const legacy = await Effect.runPromise(ShellParse.scan(command, shell, "/workspace"))
+          const portable = await Effect.runPromise(ShellParse.scan(command, shell, "/workspace", { portable: true }))
+          const native = await Effect.runPromise(ShellParse.scanPortable(command, shell, "/workspace"))
+          console.log(JSON.stringify([legacy, portable, native]))
+        `,
+          ],
+          cwd: `${import.meta.dir}/..`,
+          env: { ...process.env, ...env },
+          stdout: "pipe",
+          stderr: "pipe",
+        })
+        const [output, error, code] = await Promise.all([
+          new Response(child.stdout).text(),
+          new Response(child.stderr).text(),
+          child.exited,
+        ])
+        expect(code, error).toBe(0)
+        const [legacy, portable, native] = JSON.parse(output)
+        expect(portable).toEqual(legacy)
+        expect(native).toEqual(legacy)
+        return
+      }
+      const legacy = await Effect.runPromise(ShellParse.scan(command, shell, "/workspace"))
+      const portable = await Effect.runPromise(ShellParse.scan(command, shell, "/workspace", { portable: true }))
+      expect(portable).toEqual(legacy)
+      expect(await Effect.runPromise(ShellParse.scanPortable(command, shell, "/workspace"))).toEqual(legacy)
+    },
+  )
+
+  test("derives the legacy prefix for long argument lists", async () => {
+    const command = `echo ${"x ".repeat(16_000)}`.trimEnd()
+    const result = await Effect.runPromise(ShellParse.scan(command, "bash", "/workspace", { portable: true }))
+    expect(result).toEqual({ commands: [{ resource: command, save: "echo *" }], directories: [] })
+  })
+
+  test("extracts inline PowerShell directory flags with case-insensitive names and quoted values", async () => {
+    const result = await Effect.runPromise(
+      ShellParse.scanPortable(
+        "Set-Location -LITERALPATH:C:\\outside; Set-Location -pAtH:'../other dir'",
+        "pwsh",
+        "/workspace",
+      ),
+    )
+    expect(result).toEqual({ commands: [], directories: ["C:\\outside", "../other dir"] })
+  })
+})
+
+describe("current native and legacy parity gaps", () => {
+  // These are observed parser gaps, not permission-policy changes that must be preserved.
+  for (const fixture of [
+    {
+      name: "native omits the legacy empty command-name node for an assignment with redirection",
+      shell: "bash",
+      command: "FOO=bar > output",
+      legacy: { commands: [{ resource: "FOO=bar > output", save: " *" }], directories: [] },
+      native: { commands: [], directories: [] },
+    },
+    {
+      name: "native retains nested executable commands without the legacy empty assignment command-name node",
+      shell: "bash",
+      command: "FOO=$(printf value) > output",
+      legacy: {
+        commands: [
+          { resource: "FOO=$(printf value) > output", save: " *" },
+          { resource: "printf value", save: "printf *" },
+        ],
+        directories: [],
+      },
+      native: { commands: [{ resource: "printf value", save: "printf *" }], directories: [] },
+    },
+    {
+      name: "native keeps numeric arguments in saved prefixes",
+      shell: "bash",
+      command: "git 2 status",
+      legacy: { commands: [{ resource: "git 2 status", save: "git status *" }], directories: [] },
+      native: { commands: [{ resource: "git 2 status", save: "git 2 *" }], directories: [] },
+    },
+    {
+      name: "native keeps numeric directory names and operator-shaped arguments",
+      shell: "bash",
+      command: "cd 123; git == value",
+      legacy: { commands: [{ resource: "git == value", save: "git value *" }], directories: [] },
+      native: { commands: [{ resource: "git == value", save: "git == *" }], directories: ["123"] },
+    },
+    {
+      name: "native preserves substitution source in saved prefixes instead of skipping the argument",
+      shell: "bash",
+      command: "git $(printf status) diff",
+      legacy: {
+        commands: [
+          { resource: "git $(printf status) diff", save: "git diff *" },
+          { resource: "printf status", save: "printf *" },
+        ],
+        directories: [],
+      },
+      native: {
+        commands: [
+          { resource: "git $(printf status) diff", save: "git $(printf status) *" },
+          { resource: "printf status", save: "printf *" },
+        ],
+        directories: [],
+      },
+    },
+    {
+      name: "directory line continuations remain unresolved source rather than legacy split operands",
+      shell: "bash",
+      command: "cd before\\\nafter",
+      legacy: { commands: [], directories: ["before", "after"] },
+      native: { commands: [], directories: ["before\\\nafter"] },
+    },
+    {
+      name: "native recognizes PowerShell carriage-return separators omitted by the legacy AST",
+      shell: "pwsh",
+      command: "Get-ChildItem\rRemove-Item victim",
+      legacy: { commands: [], directories: [] },
+      native: {
+        commands: [
+          { resource: "Get-ChildItem", save: "Get-ChildItem *" },
+          { resource: "Remove-Item victim", save: "Remove-Item *" },
+        ],
+        directories: [],
+      },
+    },
+    {
+      name: "native recognizes tab-separated PowerShell commands omitted by the legacy AST",
+      shell: "pwsh",
+      command: "git\tstatus",
+      legacy: { commands: [], directories: [] },
+      native: { commands: [{ resource: "git\tstatus", save: "git\tstatus *" }], directories: [] },
+    },
+    {
+      name: "native preserves complete PowerShell flag=value resources",
+      shell: "pwsh",
+      command: "git --flag=value",
+      legacy: { commands: [{ resource: "git --flag", save: "git --flag *" }], directories: [] },
+      native: { commands: [{ resource: "git --flag=value", save: "git --flag=value *" }], directories: [] },
+    },
+    {
+      name: "native does not split comma-separated PowerShell directory operands",
+      shell: "pwsh",
+      command: "Set-Location a,b",
+      legacy: { commands: [], directories: ["a", ",b"] },
+      native: { commands: [], directories: ["a,b"] },
+    },
+  ]) {
+    test(fixture.name, async () => {
+      const native = await Effect.runPromise(ShellParse.scanPortable(fixture.command, fixture.shell, "/workspace"))
+      expect(native).toEqual(fixture.native)
+      expect(await Effect.runPromise(ShellParse.scan(fixture.command, fixture.shell, "/workspace"))).toEqual(
+        fixture.legacy,
+      )
+      expect(
+        await Effect.runPromise(ShellParse.scan(fixture.command, fixture.shell, "/workspace", { portable: true })),
+      ).toEqual(native)
+      expect(native).not.toEqual(fixture.legacy)
+    })
+  }
+})
+
+describe("legacy directory command behavior", () => {
+  test.each(["bash", "zsh", "pwsh"])("retains the original shared directory command set: %s", async (shell) => {
+    const result = await Effect.runPromise(
+      ShellParse.scan(
+        "chdir /outside; set-location /elsewhere; push-location /stack; sl .; pop-location",
+        shell,
+        "/workspace",
+      ),
+    )
+    expect(result).toEqual({
+      commands: [
+        { resource: "sl .", save: "sl *" },
+        { resource: "pop-location", save: "pop-location *" },
+      ],
+      directories: ["/outside", "/elsewhere", "/stack"],
+    })
+  })
+})

packages/core/test/shell-parse-parity.test.ts

@@ -3,30 +3,104 @@ import { Effect } from "effect"
 import { ShellParse } from "../src/shell/parse.js"
 import { ShellScan } from "../src/shell/scan.js"
 
-describe("ShellParse portable parity", () => {
-  test("matches tree-sitter for generated supported syntax", async () => {
-    for (const [shell, command] of generated()) {
+describe("ShellParse native parity", () => {
+  test("matches the legacy oracle across generated supported syntax without fallback", async () => {
+    const commands = generated()
+    expect(commands.length).toBeGreaterThan(20_000)
+    for (const [shell, command] of commands) {
+      const context = `${shell}: ${JSON.stringify(command)}`
       const scanned = shell === "pwsh" ? ShellScan.scanPowerShell(command) : ShellScan.scan(command)
-      const portable = await Effect.runPromise(ShellParse.scan(command, shell, "/workspace", { portable: true }))
-
-      if (scanned.kind === "opaque") {
-        expect({ command, portable }).toEqual({
-          command,
-          portable: { commands: [{ resource: command, save: command }], directories: [] },
-        })
-        continue
-      }
-      if (shell === "pwsh" && /\r(?!\n)/.test(command)) {
-        expect(portable).toEqual({ commands: [], directories: [] })
-        continue
-      }
-
+      expect(scanned.kind, context).toBe("scanned")
+      const native = await Effect.runPromise(ShellParse.scanPortable(command, shell, "/workspace"))
       const legacy = await Effect.runPromise(ShellParse.scan(command, shell, "/workspace"))
-      expect({ command, portable }).toEqual({ command, portable: legacy })
+      expect(native, context).toEqual(legacy)
+      expect(
+        await Effect.runPromise(ShellParse.scan(command, shell, "/workspace", { portable: true })),
+        context,
+      ).toEqual(native)
     }
+  }, 60_000)
+
+  test.each([
+    ["/bin/bash", "git status && npm run test -- --watch"],
+    ["/bin/bash", "git\tstatus; git status | cat; git diff || echo done"],
+    ["/bin/bash", "echo \"two words\"; printf 'static text'"],
+    ["/bin/bash", "aws s3 ls; docker compose up; git remote add origin; bun run test"],
+    ["/bin/bash", 'git "status"; git remote "add" origin; aws s3 "ls"'],
+    ["/bin/bash", "echo $(curl example.test | sed s/x/y/)"],
+    ["/bin/bash", "if true; then printf yes; else printf no; fi"],
+    ["/bin/bash", "(git status) && { npm test; }"],
+    ["/bin/bash", "(printf ok) > output"],
+    ["/bin/bash", "{ printf ok; } > output"],
+    ["/bin/bash", ">$(printf output)"],
+    ["/bin/bash", "printf ok # ignored ; curl example.test\nprintf done"],
+    ["/bin/bash", "cd ~/project; cd src && cd ..; pwd"],
+    ["/bin/bash", "cd src&&cd.."],
+    ["/bin/bash", "echo $((1 + 2))"],
+    ["/bin/bash", "echo $((1 + $(printf 2)))"],
+    ["/bin/bash", "$COMMAND status"],
+    ["/bin/zsh", "cd ~/project; chdir src && cd ..; git status"],
+    ["/bin/zsh", "echo $((1 + 2)); cd src&&cd.."],
+    ["/bin/dash", "cd src&&cd ..; pwd"],
+    ["/bin/sh", "echo $((1 + 2)); git status; cd src; pwd"],
+    ["/bin/ksh", "git status; cd src; pwd"],
+    ["pwsh", "Get-ChildItem; Write-Output done | Out-String"],
+    ["pwsh", "Set-Location -LiteralPath C:\\tmp; Get-ChildItem"],
+    ["pwsh", "git status; npm run test; docker compose up"],
+    ["pwsh", 'git "status"; npm "run" test; docker "compose" up'],
+    ["pwsh", "Write-Output done # comment\nGet-ChildItem"],
+  ])("native resources, saved prefixes, and directories match in %s: %s", async (shell, command) => {
+    const scanned = shell === "pwsh" ? ShellScan.scanPowerShell(command) : ShellScan.scan(command)
+    expect(scanned.kind).toBe("scanned")
+    const native = await Effect.runPromise(ShellParse.scanPortable(command, shell, "/workspace"))
+    expect(native).toEqual(await Effect.runPromise(ShellParse.scan(command, shell, "/workspace")))
+    expect(await Effect.runPromise(ShellParse.scan(command, shell, "/workspace", { portable: true }))).toEqual(native)
+  })
+
+  test.each(["> output", "FOO=bar", "2>> output"])(
+    "returns an explicit empty result for statements without executable command nodes: %s",
+    async (command) => {
+      expect(ShellScan.scan(command)).toEqual({ kind: "scanned", commands: [] })
+      const native = await Effect.runPromise(ShellParse.scanPortable(command, "bash", "/workspace"))
+      expect(native).toEqual({ commands: [], directories: [] })
+      expect(await Effect.runPromise(ShellParse.scan(command, "bash", "/workspace"))).toEqual(native)
+      expect(await Effect.runPromise(ShellParse.scan(command, "bash", "/workspace", { portable: true }))).toEqual(
+        native,
+      )
+    },
+  )
+})
+
+describe("ShellParse malformed native syntax", () => {
+  test.each([
+    ["bash", 'echo "unterminated', "unterminated-quote"],
+    ["bash", "printf done &&", "invalid-structure"],
+    ["bash", "cat >", "invalid-redirect"],
+    ["bash", ">", "invalid-redirect"],
+    ["bash", "FOO=bar >", "invalid-redirect"],
+    ["bash", "echo \\", "unterminated-escape"],
+    ["pwsh", 'Write-Output "unterminated', "unterminated-quote"],
+    ["pwsh", "git 12>bar", "invalid-redirect"],
+    ["pwsh", "Write-Output `", "unterminated-escape"],
+  ] as const)("fails explicitly for malformed %s syntax: %s", async (shell, command, reason) => {
+    const scanned = shell === "pwsh" ? ShellScan.scanPowerShell(command) : ShellScan.scan(command)
+    expect(scanned).toEqual({ kind: "opaque", reason })
+    expect(await Effect.runPromise(Effect.result(ShellParse.scanPortable(command, shell, "/workspace")))).toMatchObject(
+      {
+        _tag: "Failure",
+        failure: { message: `Portable shell scanner cannot analyze command: ${reason}` },
+      },
+    )
+    expect(
+      await Effect.runPromise(Effect.result(ShellParse.scan(command, shell, "/workspace", { portable: true }))),
+    ).toMatchObject({
+      _tag: "Failure",
+      failure: { message: `Portable shell scanner cannot analyze command: ${reason}` },
+    })
   })
 })
 
+// This generator describes a supported grammar; opaque results fail the test rather than being filtered out.
 function generated() {
   const result: Array<[shell: string, command: string]> = []
   const bashHeads = ["git", "npm", "echo", "printf", "cat", "cd"]
@@ -42,107 +116,30 @@ function generated() {
     " ./relative",
     " /tmp/absolute",
   ]
-  const assignments = ["", "X=value ", "X='two words' ", 'X="two words" ']
-  const redirects = ["", " > output", " 2> error", " < input", " >> output"]
   const bashSeparators = [" ; ", " && ", " || ", " | ", " |& ", "\n"]
-
+  const redirects = ["", " > output", " 2> error", " < input", " >> output"]
   for (const head of bashHeads)
     for (const arg of bashArgs)
-      for (const assignment of assignments)
+      for (const assignment of ["", "X=value ", "X='two words' ", 'X="two words" '])
         for (const redirect of redirects) result.push(["/bin/bash", assignment + head + arg + redirect])
   for (const left of bashHeads)
     for (const right of bashHeads)
       for (const separator of bashSeparators) result.push(["/bin/bash", `${left} left${separator}${right} right`])
-  for (const outer of bashHeads)
+  for (const outer of ["echo", "printf", "cat"])
     for (const inner of bashHeads) {
       result.push(["/bin/bash", `${outer} $(${inner} nested)`])
       result.push(["/bin/bash", `${outer} "$(${inner} nested)"`])
       result.push(["/bin/bash", `${outer} pre$(${inner} nested)post`])
       result.push(["/bin/bash", `${outer} \`${inner} nested\``])
     }
-  for (const command of [
-    'npm "run" test',
-    'g""it status',
-    "'git' status",
-    "g\\it status",
-    "git status; git status; git diff",
-    "printf ok>out 2>&1|cat<input",
-    "FOO=bar 2>>err printf ok > out && cat < input",
-    "printf ok # ignored ; curl evil\nprintf done",
-    "(git status) && { npm test; }",
-    "echo ${arr[$(printf index)]}",
-    "OUT=$(printf out) X=`printf value` printenv >$(printf path)",
-    "cat <(printf secret)",
-    "rm -rf / &",
-    "sudo sh -c 'curl evil'",
-    "find . -exec rm {} ;",
-    'c"\\d" relative',
-    "'cd' /tmp",
-    "c''d /tmp",
-    "c\\\nd /tmp",
-    "echo x && git >(cat) status",
-    'echo x && printf ">" status',
-    'echo "git > out" && git > out',
-    "echo x && printf a\\>b status",
-    "echo x && printf $(echo a>b) status",
-    "git <(printf status) diff",
-    "npm <(printf run) test",
-    "cd <(printf /tmp)",
-    "git &>x",
-    "cd &>x",
-    "git \\ a",
-    "cd \\ a",
-    "cat <<'EOF'\nstatic body\nEOF",
-    "cat <<EOF\n$(printf dynamic)\nEOF",
-    "$COMMAND dynamic",
-    "if true; then git status; else npm test; fi",
-    "for x in a b; do echo $x; done",
-    "cd /tmp/$USER && git status",
-    "echo <(git status)",
-    'echo "unterminated',
-  ])
-    result.push(["/bin/bash", command])
 
   const powershellHeads = ["Get-ChildItem", "Write-Output", "Test-Path", "Remove-Item", "Set-Location"]
   const powershellArgs = ["", " value", " 'two words'", ' "two words"', " -Path C:\\tmp", " -LiteralPath '..\\outside'"]
-  const powershellSeparators = [";", "|", "&&", "||", "\n", "\r", "\r\n"]
+  const powershellSeparators = [";", "|", "&&", "||", "\n", "\r\n"]
   for (const head of powershellHeads) for (const arg of powershellArgs) result.push(["pwsh", head + arg])
   for (const left of powershellHeads)
     for (const right of powershellHeads)
       for (const separator of powershellSeparators) result.push(["pwsh", `${left} left${separator}${right} right`])
-  for (const command of [
-    "Get-ChildItem; Get-ChildItem; Write-Output done",
-    "Write-Output 'a''b; still string'; Write-Output \"a`\"; still string\"",
-    "Get-Content in.txt > out.txt 2>&1 | Out-File all.log",
-    "Write-Output ok > output.txt # ignored\nGet-ChildItem",
-    "Write-Output ok > output.txt # ignored\rGet-ChildItem",
-    "Write-Output ok > output.txt # ignored\r\nGet-ChildItem",
-    "& git status",
-    ". ./deploy.ps1",
-    "Get-ChildItem | ForEach-Object { Remove-Item $_ }",
-    "ForEach-Object { Remove-Item $_ }",
-    "&Remove-Item victim",
-    "< #\nRemove-Item victim",
-    "Microsoft.PowerShell.Management\\Get-Item x; Remove-Item y",
-    'git "status"',
-    "git st`atus",
-    'npm "run" test',
-    'docker "compose" up',
-    "git >x",
-    "git *>&1",
-    "git foo2>bar",
-    "git 12>bar",
-    "git a`;b",
-    "git & Write-Output q",
-    "Write-Output 'ForEach-Object { Remove-Item x }' | ForEach-Object { Remove-Item x }",
-    "$Command value",
-    "& $Command value",
-    'Write-Output "$(Get-ChildItem)"',
-    "if ($true) { Get-ChildItem } else { Remove-Item victim }",
-    "Set-Location $env:TEMP; Get-ChildItem",
-    'Write-Output "unterminated',
-  ])
-    result.push(["pwsh", command])
 
   let state = 0x5eed1234
   const random = (length: number) => {
@@ -156,10 +153,10 @@ function generated() {
     const separator = bashSeparators[random(bashSeparators.length)]
     const bashForms = [
       `${left}${arg}${separator}${right} fuzz${index}`,
-      `${left}${arg} $(${right} fuzz${index})`,
+      `echo $(${right} fuzz${index})`,
       `${left}${arg} # ignored\n${right} fuzz${index}`,
       `X=value ${left}${arg}${redirects[random(redirects.length)]}`,
-      `${left} before\\\nafter${separator}${right} fuzz${index}`,
+      `${left} 'two words'${separator}${right} fuzz${index}`,
     ]
     result.push(["/bin/bash", bashForms[index % bashForms.length]])
 
@@ -171,10 +168,9 @@ function generated() {
       `${powershellLeft}${powershellArg}${powershellSeparator}${powershellRight} fuzz${index}`,
       `${powershellLeft}${powershellArg} # ignored\n${powershellRight} fuzz${index}`,
       `${powershellLeft} fuzz${index} > output; ${powershellRight}${powershellArg}`,
-      `${powershellLeft}\`\n fuzz${index}; ${powershellRight}${powershellArg}`,
+      `${powershellLeft} "fuzz${index}" | ${powershellRight}${powershellArg}`,
     ]
     result.push(["pwsh", powershellForms[index % powershellForms.length]])
   }
-
   return result
 }