refactor(tui): simplify plugin update controls

fix(tui): refine plugin update states
docs(cli): document plugin update controls
2026-08-27 12:06:22 +00:00 · 2026-08-25 22:02:54 -04:00 · 2026-08-25 22:02:54 -04:00 · 2026-08-25 22:02:54 -04:00 · 2026-08-25 22:02:54 -04:00 · 2026-08-25 22:02:54 -04:00
523 changed files with 7672 additions and 27414 deletions
@@ -1,5 +0,0 @@
---
-"@opencode-ai/core": patch
---
-
-Include both paths of renamed files in new snapshot change lists so undo restores the original file instead of only deleting the renamed file.
@@ -1,8 +0,0 @@
---
-"@opencode-ai/core": minor
-"@opencode-ai/schema": patch
---
-
-Open durable sessions with in-process model, tool, instruction, and permission capabilities. Live Sources update at safe boundaries through existing instruction epochs, while capability replacement waits for the next busy period. Capability-owned sessions remain pending after restart until their host reopens and drives them.
-
-Close an open's in-process capabilities after settlement without deleting durable history. Tool executors may yield domain errors, which normalize to tool failures while canonical permission declines retain their interruption behavior.
@@ -1,10 +0,0 @@
---
-"@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.
@@ -42,7 +42,7 @@ jobs:
      - name: Find affected packages
        id: packages
        env:
-          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_BASE: ${{ github.event.pull_request.base.sha || github.event.before }}
          TURBO_SCM_HEAD: ${{ github.sha }}
        run: |
          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
@@ -119,7 +119,7 @@ jobs:
          GITHUB_ACTIONS=false bun turbo test --affected
        env:
          OPENCODE_EXPERIMENTAL_DISABLE_FILEWATCHER: ${{ runner.os == 'Windows' && 'true' || 'false' }}
-          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_BASE: ${{ github.event.pull_request.base.sha || github.event.before }}
          TURBO_SCM_HEAD: ${{ github.sha }}

      - name: Verify published codemode package
@@ -137,7 +137,7 @@ jobs:
          fi
          bun turbo verify:package --affected --filter=@opencode-ai/sdk
        env:
-          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_BASE: ${{ github.event.pull_request.base.sha || github.event.before }}
          TURBO_SCM_HEAD: ${{ github.sha }}

      - name: Verify compiled service lifecycle
@@ -237,9 +237,9 @@ jobs:
        working-directory: packages/app
        run: bunx playwright install chromium

-      - name: Run app e2e tests against production build
+      - name: Run app e2e tests
        if: env.E2E_ENABLED == 'true'
-        run: bun --cwd packages/app test:e2e:built
+        run: bun --cwd packages/app test:e2e:local
        env:
          CI: true
        timeout-minutes: 30
@@ -125,7 +125,6 @@
        "@effect/platform-node": "catalog:",
        "@opencode-ai/client": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
-        "@opencode-ai/pty": "0.1.12",
        "@opencode-ai/schema": "workspace:*",
        "@opencode-ai/server": "workspace:*",
        "@opencode-ai/tui": "workspace:*",
@@ -348,8 +347,10 @@
        "@ai-sdk/anthropic": "3.0.82",
        "@ai-sdk/azure": "3.0.88",
        "@ai-sdk/cohere": "3.0.27",
+        "@ai-sdk/deepinfra": "2.0.41",
        "@ai-sdk/gateway": "3.0.104",
        "@ai-sdk/google-vertex": "4.0.128",
+        "@ai-sdk/groq": "3.0.31",
        "@ai-sdk/mistral": "3.0.51",
        "@ai-sdk/openai-compatible": "2.0.41",
        "@ai-sdk/perplexity": "3.0.26",
@@ -364,7 +365,6 @@
        "@opencode-ai/ai": "workspace:*",
        "@opencode-ai/codemode": "workspace:*",
        "@opencode-ai/plugin": "workspace:*",
-        "@opencode-ai/pty": "0.1.12",
        "@opencode-ai/schema": "workspace:*",
        "@opencode-ai/util": "workspace:*",
        "@parcel/watcher": "2.5.1",
@@ -553,20 +553,6 @@
        "@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",
@@ -745,7 +731,6 @@
      },
      "devDependencies": {
        "@happy-dom/global-registrator": "20.0.11",
-        "@playwright/test": "catalog:",
        "@tsconfig/node22": "catalog:",
        "@types/bun": "catalog:",
        "@types/luxon": "catalog:",
@@ -851,7 +836,6 @@
        "@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",
@@ -893,7 +877,6 @@
      "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:*",
@@ -1003,6 +986,7 @@
        "mime-types": "3.0.2",
        "minimatch": "10.2.5",
        "npm-package-arg": "13.0.2",
+        "pacote": "21.5.1",
        "resolve.exports": "catalog:",
      },
      "devDependencies": {
@@ -1012,6 +996,7 @@
        "@types/node": "catalog:",
        "@types/npm-package-arg": "6.1.4",
        "@types/npmcli__arborist": "6.3.3",
+        "@types/pacote": "11.1.8",
        "@typescript/native-preview": "catalog:",
      },
    },
@@ -1195,6 +1180,8 @@

    "@ai-sdk/deepgram": ["@ai-sdk/deepgram@2.0.52", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-8pcrQvEQCbvrrQKnD6hclBbI0hUgSrgyADykRbabxv/g9vPurfMC6n23J7dD+KZ3EcCoW+qz3IUIfySJ58gBOg=="],

+    "@ai-sdk/deepinfra": ["@ai-sdk/deepinfra@2.0.41", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.37", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-y6RoOP7DGWmDSiSxrUSt5p18sbz+Ixe5lMVPmdE7x+Tr5rlrzvftyHhjWHfqlAtoYERZTGFbP6tPW1OfQcrb4A=="],
+
    "@ai-sdk/deepseek": ["@ai-sdk/deepseek@2.0.47", "", { "dependencies": { "@ai-sdk/provider": "3.0.14", "@ai-sdk/provider-utils": "4.0.38" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MzcQ321JO8OY+TVLFI81A7cIIuoeLLxrLCDD+8C1E3Ro6UFyfMtRXo9bw9OhTMRSDMo6hgSDOo4Fekz8aJtQYQ=="],

    "@ai-sdk/elevenlabs": ["@ai-sdk/elevenlabs@2.0.52", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-ZgkausouWvO9U4ZtowNJ093bSNYOvH8zqls3uLC3+oxzWvbbTZO8SOdmFk0+gGafsXFJvq2yUX9+rEeJPwOJLw=="],
@@ -2165,8 +2152,6 @@

    "@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"],
@@ -2175,20 +2160,6 @@

    "@opencode-ai/protocol": ["@opencode-ai/protocol@workspace:packages/protocol"],

-    "@opencode-ai/pty": ["@opencode-ai/pty@0.1.12", "", { "optionalDependencies": { "@opencode-ai/pty-darwin-arm64": "0.1.12", "@opencode-ai/pty-darwin-x64": "0.1.12", "@opencode-ai/pty-linux-arm64-gnu": "0.1.12", "@opencode-ai/pty-linux-arm64-musl": "0.1.12", "@opencode-ai/pty-linux-x64-gnu": "0.1.12", "@opencode-ai/pty-linux-x64-musl": "0.1.12" }, "bin": { "opencode-pty": "bin/opencode-pty.js" } }, "sha512-dl4FyJUhTXThsWYY8txG/8/nwN7dE0M5Sic9r4L9f2pvtJnbR5zrCrPoiPIBIxZle1wVks1dhz4z/CfqLf5sCg=="],
-
-    "@opencode-ai/pty-darwin-arm64": ["@opencode-ai/pty-darwin-arm64@0.1.12", "", { "os": "darwin", "cpu": "arm64" }, "sha512-tMvoriq3VegVlj1uEglc6qE0M7VXy61nyf9Si7tTO7xa8JiyxuFJSXOZ1pGeErDu+pe24hvTyVOR+gkdew8w9g=="],
-
-    "@opencode-ai/pty-darwin-x64": ["@opencode-ai/pty-darwin-x64@0.1.12", "", { "os": "darwin", "cpu": "x64" }, "sha512-Sn5vMLL5giHOhx7J5H6zwDp4YjjXorY+QV0IEYY+SCT4wQfRBliokIyj23pRl6P2RK3u9bDLXJHDNMfDVZ2Rxg=="],
-
-    "@opencode-ai/pty-linux-arm64-gnu": ["@opencode-ai/pty-linux-arm64-gnu@0.1.12", "", { "os": "linux", "cpu": "arm64" }, "sha512-HbnlKZy052l7G527wK0+05EXaUpZ4ykVAmNBEzqWCoi4TeQj2+Nr9kJ9trx9o1KrVcT4Ki58CCvN5QOls6Z0yQ=="],
-
-    "@opencode-ai/pty-linux-arm64-musl": ["@opencode-ai/pty-linux-arm64-musl@0.1.12", "", { "os": "linux", "cpu": "arm64" }, "sha512-2nTN7ggu1h9XgjNcoQMYjP5sirfYnAskpdFCOqjokLqhytX/IMMmkRTQs+foaEaPz0dAIQD3DQplR2jZIgxp1w=="],
-
-    "@opencode-ai/pty-linux-x64-gnu": ["@opencode-ai/pty-linux-x64-gnu@0.1.12", "", { "os": "linux", "cpu": "x64" }, "sha512-FnD5ndnObTQKAoaVvxLKi5W+r3/+dsaMsobz6uK0B9hlmffXxY5CQ6HQyUU/h3aIKLWXxho5XYkA2b9yrp8/gA=="],
-
-    "@opencode-ai/pty-linux-x64-musl": ["@opencode-ai/pty-linux-x64-musl@0.1.12", "", { "os": "linux", "cpu": "x64" }, "sha512-prkrNu6uvjqoffxdGiDHSU5C0Y+kCSfv+lslu7dfRPgPKenVELNpRTAbOduyrWPac2vGt8j5NM61icJyodbJmA=="],
-
    "@opencode-ai/schema": ["@opencode-ai/schema@workspace:packages/schema"],

    "@opencode-ai/script": ["@opencode-ai/script@workspace:packages/script"],
@@ -5941,6 +5912,10 @@

    "@ai-sdk/deepgram/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.46", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8", "undici": "^6.28.0" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-tEtld97plCFiYevsJuOkGkeuhQndeMWFBVrJS4AjnbD5AqrNSXRCe0p+BZ3Cju/sxDeeZ9ym3q9YUV8fASA7aQ=="],

+    "@ai-sdk/deepinfra/@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.37", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-+POSFVcgiu47BK64dhsI6OpcDC0/VAE2ZSaXdXGNNhpC/ava++uSRJYks0k2bpfY0wwCTgpAWZsXn/dG2Yppiw=="],
+
+    "@ai-sdk/deepinfra/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.21", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MtFUYI1/8mgDvRmaBDjbLJPFFrMG777AvSgyIFQtZHIMzm88R/12vYBBpnk7pfiWLFE1DSZzY4WDYzGbKAcmiw=="],
+
    "@ai-sdk/deepseek/@ai-sdk/provider": ["@ai-sdk/provider@3.0.14", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-5X1k57JBJ4H7H1QjX7CnJYAB1I19r/trVZTMcSms7/kLNZ8RaU4Nt2agcwZzv82Hfx6Q7/TOLU7agAKeFfc8cA=="],

    "@ai-sdk/deepseek/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.38", "", { "dependencies": { "@ai-sdk/provider": "3.0.14", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-/HHGmtKllqjg1OLc023v9w9kK3laW7Z6TzfZukYQWCsGBbzB9p60zTvvpXFVcs44NZBVXL3viOa1HRKUbeee8g=="],
@@ -2,7 +2,7 @@
 exact = true
 # Only install newly resolved package versions published at least 3 days ago.
 minimumReleaseAge = 259200
-minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@brendonovich/vite-plugin-opencode", "@opencode-ai/sdk", "@opencode-ai/pty", "@opencode-ai/pty-darwin-arm64", "@opencode-ai/pty-darwin-x64", "@opencode-ai/pty-linux-arm64-gnu", "@opencode-ai/pty-linux-arm64-musl", "@opencode-ai/pty-linux-x64-gnu", "@opencode-ai/pty-linux-x64-musl", "@opentui/core", "@opentui/core-darwin-arm64", "@opentui/core-darwin-x64", "@opentui/core-linux-arm64", "@opentui/core-linux-arm64-musl", "@opentui/core-linux-x64", "@opentui/core-linux-x64-musl", "@opentui/core-win32-arm64", "@opentui/core-win32-x64", "@opentui/keymap", "@opentui/solid", "opentui-spinner", "gitlab-ai-provider", "opencode-gitlab-auth", "@ff-labs/fff-node", "@ff-labs/fff-bun", "@ff-labs/fff-bin-darwin-arm64", "@ff-labs/fff-bin-darwin-x64", "@ff-labs/fff-bin-linux-arm64-gnu", "@ff-labs/fff-bin-linux-arm64-musl", "@ff-labs/fff-bin-linux-x64-gnu", "@ff-labs/fff-bin-linux-x64-musl", "@ff-labs/fff-bin-win32-arm64", "@ff-labs/fff-bin-win32-x64", "@pierre/diffs", "@pierre/theming", "app-builder-lib", "dmg-builder", "electron-builder", "electron-publish", "blume"]
+minimumReleaseAgeExcludes = ["@ai-sdk/amazon-bedrock", "@ai-sdk/anthropic", "@brendonovich/vite-plugin-opencode", "@opencode-ai/sdk", "@opentui/core", "@opentui/core-darwin-arm64", "@opentui/core-darwin-x64", "@opentui/core-linux-arm64", "@opentui/core-linux-arm64-musl", "@opentui/core-linux-x64", "@opentui/core-linux-x64-musl", "@opentui/core-win32-arm64", "@opentui/core-win32-x64", "@opentui/keymap", "@opentui/solid", "opentui-spinner", "gitlab-ai-provider", "opencode-gitlab-auth", "@ff-labs/fff-node", "@ff-labs/fff-bun", "@ff-labs/fff-bin-darwin-arm64", "@ff-labs/fff-bin-darwin-x64", "@ff-labs/fff-bin-linux-arm64-gnu", "@ff-labs/fff-bin-linux-arm64-musl", "@ff-labs/fff-bin-linux-x64-gnu", "@ff-labs/fff-bin-linux-x64-musl", "@ff-labs/fff-bin-win32-arm64", "@ff-labs/fff-bin-win32-x64", "@pierre/diffs", "@pierre/theming", "app-builder-lib", "dmg-builder", "electron-builder", "electron-publish", "blume"]

 [test]
 root = "./do-not-run-tests-from-root"
@@ -1,8 +1,8 @@
 {
  "nodeModules": {
-    "x86_64-linux": "sha256-iYdVrLtyKmjlyypisF9SqzgyriWT90kSCh3crxw9AKU=",
-    "aarch64-linux": "sha256-BV2t4w5ujArbtSC/Qfm3gLzevQW9A6hMgOyPVp94g/o=",
-    "aarch64-darwin": "sha256-EwMq7zaxzzcsmH0Pjqu4ftGdcM8Lna8mvHgKzRcVI8g=",
-    "x86_64-darwin": "sha256-PokzxlkQy6JvHADF2ZMIIDI1u9ZjSNNedpmR9gvHS5c="
+    "x86_64-linux": "sha256-fOM/kGJJ1cipCHQIxioDZEB7NZykpSiqgwm7gIS6THI=",
+    "aarch64-linux": "sha256-XTY2C33HjsBMWO7VIiWc2MynjJrxbTLrOJ+6pM+afI0=",
+    "aarch64-darwin": "sha256-wX6+bC18djtPZ7A9ch+wryM7tDFfrAlT0xx0QTk6EJQ=",
+    "x86_64-darwin": "sha256-dcRRX4bYq5AmG4GcVmYq/M+06dlf4KJHn+clT2JY48g="
  }
 }
@@ -36,12 +36,7 @@ const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
 // Protocols whose wire format ignores inline cache markers (OpenAI's implicit
 // prefix caching, Gemini's implicit + out-of-band CachedContent). Skip the
 // whole policy pass for these — emitting hints would be harmless but pointless.
-const RESPECTS_INLINE_HINTS = new Set([
-  "anthropic-messages",
-  "google-vertex-messages",
-  "bedrock-converse",
-  "openrouter",
-])
+const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "google-vertex-messages", "bedrock-converse", "openrouter"])

 const makeHint = (ttlSeconds: number | undefined): CacheHint =>
  ttlSeconds !== undefined ? new CacheHint({ type: "ephemeral", ttlSeconds }) : new CacheHint({ type: "ephemeral" })
@@ -69,22 +69,14 @@ export interface OptionsInput {
  // SDK Metadata:2649 {user_id?: string | null}
  readonly metadata?: { readonly user_id?: string | null }
  // SDK MessageCreateParamsContainer:2596 ContainerParams|string
-  readonly container?:
-    | string
-    | { readonly id?: string | null; readonly skills?: ReadonlyArray<Record<string, unknown>> | null }
+  readonly container?: string | { readonly id?: string | null; readonly skills?: ReadonlyArray<Record<string, unknown>> | null }
  readonly inference_geo?: string | null
  readonly inferenceGeo?: string | null
  readonly cache_control?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
  readonly cacheControl?: { readonly type: "ephemeral"; readonly ttl?: "5m" | "1h" }
  // SDK OutputConfig:2684 {effort, format: JSONOutputFormat}
-  readonly output_config?: {
-    readonly effort?: string | null
-    readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null
-  }
-  readonly outputConfig?: {
-    readonly effort?: string | null
-    readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null
-  }
+  readonly output_config?: { readonly effort?: string | null; readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null }
+  readonly outputConfig?: { readonly effort?: string | null; readonly format?: { readonly type: "json_schema"; readonly schema: Record<string, unknown> } | null }
 }

 export type ProviderOptionsInput = OptionsInput
@@ -267,11 +259,7 @@ const AnthropicToolChoice = Schema.Union([
    type: Schema.Literals(["auto", "any", "none"]),
    disable_parallel_tool_use: Schema.optional(Schema.Boolean),
  }),
-  Schema.Struct({
-    type: Schema.tag("tool"),
-    name: Schema.String,
-    disable_parallel_tool_use: Schema.optional(Schema.Boolean),
-  }),
+  Schema.Struct({ type: Schema.tag("tool"), name: Schema.String, disable_parallel_tool_use: Schema.optional(Schema.Boolean) }),
 ])

 const AnthropicThinking = Schema.Union([
@@ -406,7 +394,6 @@ 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
@@ -441,18 +428,18 @@ const cacheControl = (breakpoints: Cache.Breakpoints, cache: CacheHint | undefin
  return Cache.ttlBucket(cache.ttlSeconds) === "1h" ? EPHEMERAL_1H : EPHEMERAL_5M
 }

-const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })
+const anthropicMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ anthropic: metadata })

-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 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 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 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 lowerTool = (breakpoints: Cache.Breakpoints, tool: ToolDefinition, inputSchema: JsonSchema): AnthropicTool => ({
@@ -512,21 +499,14 @@ const serverToolResultType = (name: string): AnthropicServerToolResultType | und
  return undefined
 }

-const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult")(function* (
-  part: ToolResultPart,
-  providerMetadataKey: string,
-) {
+const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult")(function* (part: ToolResultPart) {
  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?.[providerMetadataKey]?.["result"] ?? part.result.value
-  return {
-    type: wireType,
-    tool_use_id: scrubToolCallID(part.id),
-    content: payload,
-  } satisfies AnthropicServerToolResultBlock
+  const payload = part.providerMetadata?.anthropic?.["result"] ?? part.result.value
+  return { type: wireType, tool_use_id: scrubToolCallID(part.id), content: payload } satisfies AnthropicServerToolResultBlock
 })

 const fileIdFromMetadata = (metadata: MediaPart["metadata"]): string | undefined => {
@@ -574,7 +554,9 @@ const documentContextFromMetadata = (metadata: MediaPart["metadata"]): string |
  return undefined
 }

-const citationsFromMetadata = (metadata: MediaPart["metadata"]): AnthropicDocumentBlock["citations"] | undefined => {
+const citationsFromMetadata = (
+  metadata: MediaPart["metadata"],
+): AnthropicDocumentBlock["citations"] | undefined => {
  if (!ProviderShared.isRecord(metadata)) return undefined
  const raw = ProviderShared.isRecord(metadata.anthropic)
    ? (metadata.anthropic.citations ?? metadata.citations)
@@ -724,7 +706,8 @@ const lowerToolResultContent = Effect.fnUntraced(function* (part: ToolResultPart
 })

 const requireThinkingSignature = (request: LLMRequest) => {
-  if (request.model.compatibility?.requireSignature !== undefined) return request.model.compatibility.requireSignature
+  if (request.model.compatibility?.requireSignature !== undefined)
+    return request.model.compatibility.requireSignature
  const provider = request.model.provider.toLowerCase()
  const model = request.model.id.toLowerCase()
  const baseURL = (request.model.route.endpoint.baseURL ?? "").toLowerCase()
@@ -808,7 +791,6 @@ 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") {
@@ -854,8 +836,8 @@ 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, providerMetadataKey)
-          const redactedData = redactedDataFromMetadata(part.providerMetadata, providerMetadataKey)
+          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata)
+          const redactedData = redactedDataFromMetadata(part.providerMetadata)
          if (signature === undefined && redactedData !== undefined) {
            content.push({ type: "redacted_thinking", data: redactedData })
            continue
@@ -884,7 +866,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
          continue
        }
        if (part.type === "tool-result" && part.providerExecuted) {
-          content.push(yield* lowerServerToolResult(part, providerMetadataKey))
+          content.push(yield* lowerServerToolResult(part))
          continue
        }
        return yield* invalid(
@@ -918,24 +900,21 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (

 const resolveOptions = Effect.fn("AnthropicMessages.resolveOptions")(function* (request: LLMRequest) {
  const input = request.providerOptions as Record<string, unknown> | undefined
-  const rawServiceTier =
-    (input as Record<string, unknown> | undefined)?.service_tier ??
-    (input as Record<string, unknown> | undefined)?.serviceTier
+  const rawServiceTier = (input as Record<string, unknown> | undefined)?.service_tier ?? (input as Record<string, unknown> | undefined)?.serviceTier
  const service_tier =
    rawServiceTier === "auto" || rawServiceTier === "standard_only"
      ? (rawServiceTier as "auto" | "standard_only")
      : undefined
  const rawMetadata = (input as Record<string, unknown> | undefined)?.metadata
  const metadata =
-    ProviderShared.isRecord(rawMetadata) && (typeof rawMetadata.user_id === "string" || rawMetadata.user_id === null)
+    ProviderShared.isRecord(rawMetadata) &&
+    (typeof rawMetadata.user_id === "string" || rawMetadata.user_id === null)
      ? { user_id: rawMetadata.user_id as string | null }
      : undefined
  const container =
    typeof (input as Record<string, unknown> | undefined)?.container === "string" ||
    ProviderShared.isRecord((input as Record<string, unknown> | undefined)?.container)
-      ? ((input as Record<string, unknown>).container as
-          | string
-          | { id?: string | null; skills?: ReadonlyArray<Record<string, unknown>> | null })
+      ? ((input as Record<string, unknown>).container as string | { id?: string | null; skills?: ReadonlyArray<Record<string, unknown>> | null })
      : undefined
  const rawInferenceGeo =
    (input as Record<string, unknown> | undefined)?.inference_geo ??
@@ -986,7 +965,8 @@ const resolveThinking = Effect.fn("AnthropicMessages.resolveThinking")(function*
    input.display === "summarized" || input.display === "omitted"
      ? (input.display as "summarized" | "omitted")
      : undefined
-  if (input.type === "adaptive") return { type: "adaptive" as const, ...(display === undefined ? {} : { display }) }
+  if (input.type === "adaptive")
+    return { type: "adaptive" as const, ...(display === undefined ? {} : { display }) }
  if (input.type === "disabled") return { type: "disabled" as const }
  if (input.type !== "enabled") return undefined
  const budget =
@@ -1074,7 +1054,7 @@ 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, providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: AnthropicUsage | undefined): Usage | undefined => {
  if (!usage) return undefined
  const nonCached = usage.input_tokens ?? undefined
  const cacheRead = usage.cache_read_input_tokens ?? undefined
@@ -1088,7 +1068,7 @@ const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string
    cacheWriteInputTokens: cacheWrite,
    reasoningTokens: usage.output_tokens_details?.thinking_tokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, usage.output_tokens, undefined),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { anthropic: usage },
  })
 }

@@ -1097,7 +1077,7 @@ const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string
 // 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, providerMetadataKey: string) => {
+const mergeUsage = (left: Usage | undefined, right: Usage | undefined) => {
  if (!left) return right
  if (!right) return left
  const nonCachedInputTokens = right.nonCachedInputTokens ?? left.nonCachedInputTokens
@@ -1115,9 +1095,7 @@ const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerM
    reasoningTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
    providerMetadata: {
-      [providerMetadataKey]:
-        mergeJsonRecords(left.providerMetadata?.[providerMetadataKey], right.providerMetadata?.[providerMetadataKey]) ??
-        {},
+      anthropic: mergeJsonRecords(left.providerMetadata?.["anthropic"], right.providerMetadata?.["anthropic"]) ?? {},
    },
  })
 }
@@ -1135,7 +1113,7 @@ 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, providerMetadataKey: string): LLMEvent | undefined => {
+const serverToolResultEvent = (block: AnthropicStreamBlock): LLMEvent | undefined => {
  if (!block.type || !isServerToolResultType(block.type)) return undefined
  const errorPayload =
    typeof block.content === "object" && block.content !== null && "type" in block.content
@@ -1149,7 +1127,7 @@ const serverToolResultEvent = (block: AnthropicStreamBlock, providerMetadataKey:
    providerExecuted: true,
    // The complete payload is irreducible provider replay state: subsequent
    // stateless requests must round-trip the typed result block verbatim.
-    providerMetadata: providerMetadata(providerMetadataKey, { blockType: block.type, result: block.content }),
+    providerMetadata: anthropicMetadata({ blockType: block.type, result: block.content }),
  })
 }

@@ -1158,8 +1136,8 @@ type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
 const NO_EVENTS: StepResult["1"] = []

 const onMessageStart = (state: ParserState, event: AnthropicEvent): StepResult => {
-  const usage = mapUsage(event.message?.usage, state.providerMetadataKey)
-  return [usage ? { ...state, usage: mergeUsage(state.usage, usage, state.providerMetadataKey) } : state, NO_EVENTS]
+  const usage = mapUsage(event.message?.usage)
+  return [usage ? { ...state, usage: mergeUsage(state.usage, usage) } : state, NO_EVENTS]
 }

 const onContentBlockStart = (
@@ -1211,16 +1189,14 @@ const onContentBlockStart = (
  if (block.type === "thinking" && block.thinking !== undefined) {
    const events: LLMEvent[] = []
    const id = `reasoning-${event.index ?? 0}`
-    const metadata =
-      block.signature === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { signature: block.signature })
-    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, metadata)
+    const providerMetadata =
+      block.signature === undefined ? undefined : anthropicMetadata({ signature: block.signature })
+    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, providerMetadata)
    return [
      {
        ...state,
        lifecycle: block.thinking
-          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, metadata)
+          ? Lifecycle.reasoningDelta(lifecycle, events, id, block.thinking, providerMetadata)
          : lifecycle,
        reasoningSignatures:
          event.index === undefined || block.signature === undefined
@@ -1243,14 +1219,14 @@ const onContentBlockStart = (
          state.lifecycle,
          events,
          `reasoning-${event.index ?? 0}`,
-          providerMetadata(state.providerMetadataKey, { redactedData: block.data }),
+          anthropicMetadata({ redactedData: block.data }),
        ),
      },
      events,
    ]
  }

-  const result = serverToolResultEvent(block, state.providerMetadataKey)
+  const result = serverToolResultEvent(block)
  if (!result) return [state, NO_EVENTS]
  const events: LLMEvent[] = []
  return [{ ...state, lifecycle: Lifecycle.stepStart(state.lifecycle, events) }, [...events, result]]
@@ -1330,7 +1306,7 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
        Lifecycle.textEnd(state.lifecycle, events, `text-${event.index}`),
        events,
        `reasoning-${event.index}`,
-        signature === undefined ? undefined : providerMetadata(state.providerMetadataKey, { signature }),
+        signature === undefined ? undefined : anthropicMetadata({ signature }),
      )
  events.push(...resultEvents)
  const reasoningSignatures = { ...state.reasoningSignatures }
@@ -1342,7 +1318,7 @@ const onMessageDelta = (
  state: ParserState,
  event: AnthropicEvent & { readonly delta?: AnthropicStreamDelta },
 ): StepResult => {
-  const usage = mergeUsage(state.usage, mapUsage(event.usage, state.providerMetadataKey), state.providerMetadataKey)
+  const usage = mergeUsage(state.usage, mapUsage(event.usage))
  return [
    {
      ...state,
@@ -1355,7 +1331,7 @@ const onMessageDelta = (
        providerMetadata:
          event.delta?.stop_sequence === null || event.delta?.stop_sequence === undefined
            ? undefined
-            : providerMetadata(state.providerMetadataKey, { stopSequence: event.delta.stop_sequence }),
+            : anthropicMetadata({ stopSequence: event.delta.stop_sequence }),
      },
    },
    NO_EVENTS,
@@ -1442,7 +1418,9 @@ const step = (state: ParserState, event: AnthropicEvent) => {
      if (event.index === undefined)
        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic ${block.type} missing index`))
      if (!block.id)
-        return Effect.fail(ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`))
+        return Effect.fail(
+          ProviderShared.eventError(ADAPTER, `Anthropic tool_use missing id at index ${event.index}`),
+        )
    }
    return Effect.succeed(onContentBlockStart(state, { ...event, content_block: block }))
  }
@@ -1481,8 +1459,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: Protocol.jsonEvent(AnthropicEvent),
-    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+    initial: () => ({
      tools: ToolStream.empty<number>(),
      reasoningSignatures: {},
      lifecycle: Lifecycle.initial(),
@@ -1496,9 +1473,10 @@ export const route = Route.make({
  provider: "anthropic",
  providerMetadataKey: "anthropic",
  protocol,
-  endpoint: Endpoint.path((input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH), {
-    baseURL: DEFAULT_BASE_URL,
-  }),
+  endpoint: Endpoint.path(
+    (input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH),
+    { baseURL: DEFAULT_BASE_URL },
+  ),
  auth: Auth.none,
  framing,
  headers: () => ({ "anthropic-version": "2023-06-01" }),
@@ -258,21 +258,19 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>
    tool: (name) => ({ tool: { name } }) as const,
  })

-const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })
+const bedrockMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ bedrock: metadata })

-const reasoningSignature = (part: ReasoningPart, providerMetadataKey: string) => {
-  const metadata = part.providerMetadata?.[providerMetadataKey]
+const reasoningSignature = (part: ReasoningPart) => {
+  const bedrock = part.providerMetadata?.bedrock
  return (
    part.encrypted ??
-    (ProviderShared.isRecord(metadata) && typeof metadata.signature === "string" ? metadata.signature : undefined)
+    (ProviderShared.isRecord(bedrock) && typeof bedrock.signature === "string" ? bedrock.signature : undefined)
  )
 }

-const reasoningRedactedData = (part: ReasoningPart, providerMetadataKey: string) => {
-  const metadata = part.providerMetadata?.[providerMetadataKey]
-  return ProviderShared.isRecord(metadata) && typeof metadata.redactedData === "string"
-    ? metadata.redactedData
-    : undefined
+const reasoningRedactedData = (part: ReasoningPart) => {
+  const bedrock = part.providerMetadata?.bedrock
+  return ProviderShared.isRecord(bedrock) && typeof bedrock.redactedData === "string" ? bedrock.redactedData : undefined
 }

 const lowerToolCall = (part: ToolCallPart): BedrockToolUseBlock => ({
@@ -320,7 +318,6 @@ 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") {
@@ -368,8 +365,8 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
          continue
        }
        if (part.type === "reasoning") {
-          const signature = reasoningSignature(part, providerMetadataKey)
-          const redactedData = reasoningRedactedData(part, providerMetadataKey)
+          const signature = reasoningSignature(part)
+          const redactedData = reasoningRedactedData(part)
          if (signature === undefined && redactedData !== undefined) {
            content.push({ reasoningContent: { redactedContent: redactedData } })
            continue
@@ -469,7 +466,7 @@ 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, providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: BedrockUsageSchema | undefined): Usage | undefined => {
  if (!usage) return undefined
  const inputTokens = ProviderShared.sumTokens(
    usage.inputTokens,
@@ -483,12 +480,11 @@ const mapUsage = (usage: BedrockUsageSchema | undefined, providerMetadataKey: st
    cacheReadInputTokens: usage.cacheReadInputTokens,
    cacheWriteInputTokens: usage.cacheWriteInputTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, usage.outputTokens, usage.totalTokens),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { bedrock: 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`
@@ -545,14 +541,20 @@ const step = (state: ParserState, event: BedrockEvent) =>
      const reasoning = event.contentBlockDelta.delta.reasoningContent
      const events: LLMEvent[] = []
      const redactedData = reasoning.redactedContent ?? reasoning.data
-      const metadata = reasoning.signature
-        ? providerMetadata(state.providerMetadataKey, { signature: reasoning.signature })
+      const providerMetadata = reasoning.signature
+        ? bedrockMetadata({ signature: reasoning.signature })
        : redactedData !== undefined
-          ? providerMetadata(state.providerMetadataKey, { redactedData })
+          ? bedrockMetadata({ redactedData })
          : undefined
      const lifecycle =
-        reasoning.text !== undefined || metadata !== undefined
-          ? Lifecycle.reasoningDelta(state.lifecycle, events, `reasoning-${index}`, reasoning.text ?? "", metadata)
+        reasoning.text !== undefined || providerMetadata !== undefined
+          ? Lifecycle.reasoningDelta(
+              state.lifecycle,
+              events,
+              `reasoning-${index}`,
+              reasoning.text ?? "",
+              providerMetadata,
+            )
          : state.lifecycle
      return [
        {
@@ -594,7 +596,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
            events,
            `reasoning-${index}`,
            state.reasoningSignatures[index]
-              ? providerMetadata(state.providerMetadataKey, { signature: state.reasoningSignatures[index] })
+              ? bedrockMetadata({ signature: state.reasoningSignatures[index] })
              : undefined,
          )
      events.push(...resultEvents)
@@ -631,7 +633,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
    }

    if (event.metadata) {
-      const usage = mapUsage(event.metadata.usage, state.providerMetadataKey) ?? state.pendingFinish?.usage
+      const usage = mapUsage(event.metadata.usage) ?? state.pendingFinish?.usage
      return [
        {
          ...state,
@@ -650,9 +652,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
        method: "stream",
        reason: classifyProviderFailure({
          message:
-            event.exception.details.message ??
-            event.exception.details.originalMessage ??
-            "Bedrock Converse stream error",
+            event.exception.details.message ?? event.exception.details.originalMessage ?? "Bedrock Converse stream error",
          code: event.exception.type,
        }),
      })
@@ -696,8 +696,7 @@ export const protocol = Protocol.make({
  },
  stream: {
    event: BedrockEvent,
-    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+    initial: () => ({
      tools: ToolStream.empty<number>(),
      pendingFinish: undefined,
      hasToolCalls: false,
@@ -82,9 +82,7 @@ const consumeFrames = (route: string) => (state: FrameBufferState, chunk: Uint8A
        "Failed to parse Bedrock Converse event-stream payload",
      )) as Record<string, unknown>
      delete parsed.p
-      out.push(
-        messageType === "exception" ? { exception: { type: eventType, details: parsed } } : { [eventType]: parsed },
-      )
+      out.push(messageType === "exception" ? { exception: { type: eventType, details: parsed } } : { [eventType]: parsed })
    }
    return [cursor, out] as const
  })
@@ -229,7 +229,6 @@ type GeminiEvent = Schema.Schema.Type<typeof GeminiEvent>

 interface ParserState {
  readonly route: string
-  readonly providerMetadataKey: string
  readonly finishReason?: string
  readonly hasToolCalls: boolean
  readonly promptFeedback?: GeminiPromptFeedback
@@ -286,23 +285,22 @@ const lowerUserPart = Effect.fn("Gemini.lowerUserPart")(function* (part: TextPar
  return { inlineData: { mimeType: media.mime, data: media.base64 } }
 })

-const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })
+const googleMetadata = (metadata: Record<string, unknown>): ProviderMetadata => ({ google: metadata })

-const thoughtSignature = (metadata: ProviderMetadata | undefined, key: string) => {
-  const value = metadata?.[key]
-  return ProviderShared.isRecord(value) && typeof value.thoughtSignature === "string"
-    ? value.thoughtSignature
+const thoughtSignature = (providerMetadata: ProviderMetadata | undefined) => {
+  const google = providerMetadata?.google
+  return ProviderShared.isRecord(google) && typeof google.thoughtSignature === "string"
+    ? google.thoughtSignature
    : undefined
 }

-const lowerToolCall = (part: ToolCallPart, omitIds: boolean, metadataKey: string) => ({
+const lowerToolCall = (part: ToolCallPart, omitIds: boolean) => ({
  functionCall: { ...(omitIds ? {} : { id: part.id }), name: part.name, args: part.input },
-  thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey),
+  thoughtSignature: thoughtSignature(part.providerMetadata),
 })

 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
@@ -344,19 +342,15 @@ 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, metadataKey) })
+          parts.push({ text: part.text, thoughtSignature: thoughtSignature(part.providerMetadata) })
          continue
        }
        if (part.type === "reasoning") {
-          parts.push({
-            text: part.text,
-            thought: true,
-            thoughtSignature: thoughtSignature(part.providerMetadata, metadataKey),
-          })
+          parts.push({ text: part.text, thought: true, thoughtSignature: thoughtSignature(part.providerMetadata) })
          continue
        }
        if (part.type === "tool-call") {
-          const lowered = lowerToolCall(part, omitCallIds, metadataKey)
+          const lowered = lowerToolCall(part, omitCallIds)
          const signature = lowered.thoughtSignature
          parts.push({
            ...lowered,
@@ -504,7 +498,7 @@ 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, metadataKey: string) => {
+const mapUsage = (usage: GeminiUsage | undefined) => {
  if (!usage) return undefined
  // Explicit provider nulls decode as `null`; normalize to `undefined` so the
  // token arithmetic below treats them like absent counts.
@@ -525,7 +519,7 @@ const mapUsage = (usage: GeminiUsage | undefined, metadataKey: string) => {
    cacheReadInputTokens: cached,
    reasoningTokens: thoughts,
    totalTokens: ProviderShared.totalTokens(promptTokens, outputTokens, usage.totalTokenCount ?? undefined),
-    providerMetadata: providerMetadata(metadataKey, usage),
+    providerMetadata: { google: usage },
  })
 }

@@ -573,15 +567,10 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
      lifecycle,
      events,
      "reasoning-0",
-      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.reasoningSignature }),
+      googleMetadata({ thoughtSignature: state.reasoningSignature }),
    )
  if (state.textSignature !== undefined)
-    lifecycle = Lifecycle.textEnd(
-      lifecycle,
-      events,
-      "text-0",
-      providerMetadata(state.providerMetadataKey, { thoughtSignature: state.textSignature }),
-    )
+    lifecycle = Lifecycle.textEnd(lifecycle, events, "text-0", googleMetadata({ thoughtSignature: state.textSignature }))
  Lifecycle.finish(lifecycle, events, {
    reason: {
      normalized:
@@ -590,9 +579,7 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
    },
    usage: state.usage,
    providerMetadata:
-      state.promptFeedback === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { promptFeedback: state.promptFeedback }),
+      state.promptFeedback === undefined ? undefined : googleMetadata({ promptFeedback: state.promptFeedback }),
  })
  return events
 }
@@ -601,9 +588,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
  const nextState = {
    ...state,
    promptFeedback: event.promptFeedback ?? state.promptFeedback,
-    usage: event.usageMetadata
-      ? (mapUsage(event.usageMetadata, state.providerMetadataKey) ?? state.usage)
-      : state.usage,
+    usage: event.usageMetadata ? (mapUsage(event.usageMetadata) ?? state.usage) : state.usage,
  }
  const candidate = event.candidates?.[0]
  if (!candidate?.content)
@@ -647,7 +632,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          events,
          "reasoning-0",
          part.text,
-          signature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: signature }) : undefined,
+          signature ? googleMetadata({ thoughtSignature: signature }) : undefined,
        )
        continue
      }
@@ -655,16 +640,14 @@ const step = (state: ParserState, event: GeminiEvent) => {
        lifecycle,
        events,
        "reasoning-0",
-        reasoningSignature
-          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-          : undefined,
+        reasoningSignature ? googleMetadata({ thoughtSignature: reasoningSignature }) : undefined,
      )
      lifecycle = Lifecycle.textDelta(
        lifecycle,
        events,
        "text-0",
        part.text,
-        textSignature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature }) : undefined,
+        textSignature ? googleMetadata({ thoughtSignature: textSignature }) : undefined,
      )
      textSignature = undefined
      continue
@@ -684,9 +667,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
        lifecycle,
        events,
        "reasoning-0",
-        reasoningSignature
-          ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-          : undefined,
+        reasoningSignature ? googleMetadata({ thoughtSignature: reasoningSignature }) : undefined,
      )
      lifecycle = Lifecycle.stepStart(lifecycle, events)
      events.push(
@@ -694,9 +675,8 @@ const step = (state: ParserState, event: GeminiEvent) => {
          id,
          name: part.functionCall.name,
          input,
-          providerMetadata: part.thoughtSignature
-            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: part.thoughtSignature })
-            : undefined,
+          providerMetadata:
+            part.thoughtSignature ? googleMetadata({ thoughtSignature: part.thoughtSignature }) : undefined,
        }),
      )
      hasToolCalls = true
@@ -734,7 +714,6 @@ 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(),
    }),
@@ -196,7 +196,7 @@ type LoweredInputItem =
 // multiple streamed summary parts into the same item before flushing.
 type OpenResponsesReasoningInput = {
  type: "reasoning"
-  id?: string
+  id: string
  summary: Array<{ type: "summary_text"; text: string }>
  encrypted_content?: string | null
 }
@@ -346,8 +346,7 @@ export const Event = Schema.StructWithRest(
    item_id: Schema.optional(Schema.String),
    output_index: Schema.optional(Schema.Number),
    summary_index: Schema.optional(Schema.Number),
-    // OutputItemAdded/Done permit a null item in the Open Responses OpenAPI schema.
-    item: optionalNull(StreamItem),
+    item: Schema.optional(StreamItem),
    response: Schema.optional(
      Schema.StructWithRest(
        Schema.Struct({
@@ -373,6 +372,9 @@ export const Event = Schema.StructWithRest(
 )
 export type Event = Schema.Schema.Type<typeof Event>

+// Which lowered input item a persisted item id is about to be attached to.
+export type ItemKind = "message" | "reasoning" | "function-call" | "hosted-tool"
+
 export interface Extension {
  readonly id: string
  readonly name: string
@@ -382,6 +384,10 @@ export interface Extension {
    readonly request: LLMRequest
  }) => MediaInput | undefined
  readonly lowerHostedToolItem?: (item: unknown) => ExtendedHostedToolItem | undefined
+  // Optional grammar check applied before a persisted item id is resent as
+  // part of replayed history. Returning false drops the id; every lowered
+  // item treats a dropped id the same as an absent one.
+  readonly acceptsItemID?: (kind: ItemKind, id: string) => boolean
 }

 const BASE: Extension = { id: ADAPTER, name: NAME }
@@ -441,37 +447,53 @@ export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LL
    tool: (toolName) => ({ type: "function" as const, name: toolName }),
  })

-// Server-issued item ids need a nonempty prefix and suffix, but the prefix is
-// provider-defined and does not necessarily identify the item's semantic type.
+// Servers validate item ids on replayed history, and a malformed or oversized
+// id can fail an otherwise valid request. Only server-issued tokens are worth
+// resending; anything else is treated as absent so the item is resent without
+// an id (or skipped, for items that cannot be expressed without one).
+const ITEM_ID_PATTERN = /^[A-Za-z0-9_-]{1,64}$/
 const itemID = (providerMetadata: ProviderMetadata | undefined, providerMetadataKey: string) => {
  const metadata = providerMetadata?.[providerMetadataKey]
-  if (!ProviderShared.isRecord(metadata) || typeof metadata.itemId !== "string") return undefined
-  const separator = metadata.itemId.indexOf("_")
-  return separator > 0 && separator < metadata.itemId.length - 1 ? metadata.itemId : undefined
+  return ProviderShared.isRecord(metadata) &&
+    typeof metadata.itemId === "string" &&
+    ITEM_ID_PATTERN.test(metadata.itemId)
+    ? metadata.itemId
+    : undefined
 }

-const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenResponsesInputItem => {
+const acceptsItemID = (extension: Extension, kind: ItemKind, id: string | undefined): id is string =>
+  id !== undefined && (extension.acceptsItemID?.(kind, id) ?? true)
+
+const lowerToolCall = (
+  part: ToolCallPart,
+  providerMetadataKey: string,
+  extension: Extension,
+): OpenResponsesInputItem => {
  const id = itemID(part.providerMetadata, providerMetadataKey)
  return {
    type: "function_call",
-    ...(id === undefined ? {} : { id }),
+    ...(acceptsItemID(extension, "function-call", id) ? { id } : {}),
    call_id: part.id,
    name: part.name,
    arguments: ProviderShared.encodeJson(part.input),
  }
 }

-const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
+const lowerReasoning = (
+  part: ReasoningPart,
+  providerMetadataKey: string,
+  extension: Extension,
+): OpenResponsesReasoningInput | undefined => {
  const metadata = part.providerMetadata?.[providerMetadataKey]
-  if (!ProviderShared.isRecord(metadata)) return undefined
  const id = itemID(part.providerMetadata, providerMetadataKey)
+  if (!ProviderShared.isRecord(metadata) || !acceptsItemID(extension, "reasoning", id)) return undefined
  const encryptedContent =
    typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
      ? metadata.reasoningEncryptedContent
      : undefined
  return {
    type: "reasoning",
-    ...(id === undefined ? {} : { id }),
+    id,
    summary: part.text.length > 0 ? [{ type: "summary_text", text: part.text }] : [],
    encrypted_content: encryptedContent,
  }
@@ -592,7 +614,8 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
          Array<{ id: string | undefined; phase: MessagePhase | null | undefined; parts: TextPart[] }>
        >((groups, part) => {
          const metadata = part.providerMetadata?.[providerMetadataKey]
-          const id = itemID(part.providerMetadata, providerMetadataKey)
+          const rawID = itemID(part.providerMetadata, providerMetadataKey)
+          const id = acceptsItemID(extension, "message", rawID) ? rawID : undefined
          const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase) : undefined
          const group = groups.at(-1)
          if (group && group.id === id && group.phase === phase) group.parts.push(part)
@@ -617,23 +640,23 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
        }
        if (part.type === "reasoning") {
          flushText()
-          const reasoning = lowerReasoning(part, providerMetadataKey)
+          const reasoning = lowerReasoning(part, providerMetadataKey, extension)
          if (!reasoning) continue
-          const existing = reasoning.id === undefined ? undefined : reasoningItems[reasoning.id]
+          const existing = reasoningItems[reasoning.id]
          if (existing) {
            existing.summary.push(...reasoning.summary)
            if (typeof reasoning.encrypted_content === "string")
              existing.encrypted_content = reasoning.encrypted_content
            continue
          }
-          if (reasoning.id !== undefined) reasoningItems[reasoning.id] = reasoning
+          reasoningItems[reasoning.id] = reasoning
          input.push(reasoning)
          continue
        }
        if (part.type === "tool-call") {
          flushText()
          if (part.providerExecuted === true) continue
-          input.push(lowerToolCall(part, providerMetadataKey))
+          input.push(lowerToolCall(part, providerMetadataKey, extension))
          continue
        }
        if (part.type === "tool-result" && part.providerExecuted === true) {
@@ -645,7 +668,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
              : Schema.is(HostedToolItem)(part.result.value)
                ? part.result.value
                : extension.lowerHostedToolItem?.(part.result.value)
-          if (id !== undefined && hosted?.id === id) {
+          if (acceptsItemID(extension, "hosted-tool", id) && hosted?.id === id) {
            if (!hostedToolItems.has(id)) {
              input.push(hosted)
              hostedToolItems.add(id)
@@ -813,7 +836,7 @@ 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.type === "reasoning" && typeof item.id === "string" && item.id.length > 0

 export type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]

@@ -892,7 +915,7 @@ 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 !== undefined) {
+  if (item?.type === "message" && item.id) {
    const phase = messagePhase(item.phase)
    return [
      {
@@ -923,7 +946,7 @@ 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 !== undefined ? providerMetadata(state, { itemId: item.id }) : undefined
+  const metadata = item.id ? providerMetadata(state, { itemId: item.id }) : undefined
  const events: LLMEvent[] = []
  const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
  return [
@@ -942,7 +965,7 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
 }

 const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResult => {
-  if (event.item_id === undefined || event.summary_index === undefined) return [state, NO_EVENTS]
+  if (!event.item_id || 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]
@@ -989,7 +1012,7 @@ const onReasoningSummaryPartAdded = (state: ParserState, event: Event): StepResu
 }

 const onReasoningSummaryPartDone = (state: ParserState, event: Event): StepResult => {
-  if (event.item_id === undefined || event.summary_index === undefined) return [state, NO_EVENTS]
+  if (!event.item_id || event.summary_index === undefined) return [state, NO_EVENTS]
  const item = state.reasoningItems[event.item_id]
  if (!item) return [state, NO_EVENTS]
  return [
@@ -1014,7 +1037,7 @@ const onFunctionCallArgumentsDelta = Effect.fn("OpenResponses.onFunctionCallArgu
  state: ParserState,
  event: Event,
 ) {
-  if (event.item_id === undefined) return [state, NO_EVENTS] satisfies StepResult
+  if (!event.item_id) 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
@@ -1045,7 +1068,7 @@ 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 !== undefined) {
+  if (item.type === "message" && item.id) {
    const itemPhase = messagePhase(item.phase)
    const phase = itemPhase === undefined ? state.messagePhases[item.id] : itemPhase
    const events: LLMEvent[] = []
@@ -1076,7 +1099,7 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
      : ToolStream.start(state.tools, id, {
          id: item.call_id,
          name: item.name,
-          providerMetadata: item.id !== undefined ? providerMetadata(state, { itemId: item.id }) : undefined,
+          providerMetadata: item.id ? providerMetadata(state, { itemId: item.id }) : undefined,
        })
    const result =
      item.arguments === undefined
@@ -1137,7 +1160,7 @@ const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (
          ([current, events], item) => {
            const id = item.id ?? (item.type === "function_call" ? item.call_id : undefined)
            if (
-              id === undefined ||
+              !id ||
              ((item.type !== "function_call" || !current.tools[id]) &&
                (item.type !== "reasoning" || !current.reasoningItems[id]))
            )
@@ -1220,13 +1243,12 @@ 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 !== undefined && outputItemID(state, input) !== input.item_id
+    input.item_id && 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 === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(
      event.type === "response.output_text.delta"
        ? onOutputTextDelta(state, event, event.item_id)
@@ -1235,7 +1257,7 @@ 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 === undefined || typeof value !== "string")
+    if (!event.item_id || typeof value !== "string")
      return ProviderShared.eventError(state.id, `${event.type} is malformed`)
    return Effect.succeed(
      event.type === "response.refusal.delta"
@@ -1244,7 +1266,7 @@ export const step = (state: ParserState, input: Event) => {
    )
  }
  if (event.type === "response.reasoning.delta" || event.type === "response.reasoning_summary_text.delta") {
-    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(onReasoningDelta(state, event, event.item_id))
  }
  if (
@@ -1252,24 +1274,24 @@ export const step = (state: ParserState, input: Event) => {
    event.type === "response.reasoning_summary_text.done" ||
    event.type === "response.reasoning_text.done"
  ) {
-    if (event.item_id === undefined) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
+    if (!event.item_id) return ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
    return Effect.succeed(onReasoningDone(state, event, event.item_id))
  }
  if (event.type === "response.reasoning_summary_part.added")
-    return event.item_id !== undefined
+    return event.item_id
      ? Effect.succeed(onReasoningSummaryPartAdded(state, event))
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
  if (event.type === "response.reasoning_summary_part.done")
-    return event.item_id !== undefined
+    return event.item_id
      ? Effect.succeed(onReasoningSummaryPartDone(state, event))
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
  if (event.type === "response.output_item.added") {
-    if (event.item?.type === "message" && event.item.id === undefined)
+    if (event.item?.type === "message" && !event.item.id)
      return ProviderShared.eventError(state.id, `${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 !== undefined
+        event.output_index !== undefined && id
          ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: id } }
          : state,
        event,
@@ -1277,11 +1299,11 @@ 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 !== undefined
+    return event.item_id
      ? onFunctionCallArgumentsDelta(state, event)
      : ProviderShared.eventError(state.id, `${event.type} is missing item_id`)
  if (event.type === "response.output_item.done") {
-    if (event.item?.type === "message" && event.item.id === undefined)
+    if (event.item?.type === "message" && !event.item.id)
      return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
    return onOutputItemDone(state, event)
  }
@@ -253,7 +253,6 @@ interface PendingToolDelta {
 }

 export interface ParserState {
-  readonly providerMetadataKey: string
  readonly tools: ToolStream.State<number>
  readonly pendingTools: Partial<Record<number, PendingToolDelta>>
  readonly toolCallEvents: ReadonlyArray<LLMEvent>
@@ -325,18 +324,17 @@ 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, providerMetadataKey: string) => {
-  const field = part.providerMetadata?.[providerMetadataKey]?.reasoningField
+const reasoningField = (part: ReasoningPart) => {
+  const field = part.providerMetadata?.openai?.reasoningField
  return typeof field === "string" ? field : undefined
 }

-const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown, providerMetadataKey: string) => {
+const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown) => {
  const observed = parts.flatMap((part) => {
-    const details = part.providerMetadata?.[providerMetadataKey]?.reasoningDetails
+    const details = part.providerMetadata?.openai?.reasoningDetails
    return Array.isArray(details) ? details : []
  })
-  if (parts.some((part) => Array.isArray(part.providerMetadata?.[providerMetadataKey]?.reasoningDetails)))
-    return observed
+  if (parts.some((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))) return observed
  if (isRecord(native) && Array.isArray(native.reasoning_details)) return native.reasoning_details
 }

@@ -368,7 +366,7 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
  message: OpenAIChatRequestMessage,
  configuredField: string | undefined,
  requireReasoning: boolean,
-  options: LoweringOptions & { readonly providerMetadataKey: string },
+  options: LoweringOptions,
 ) {
  const content: TextPart[] = []
  const reasoning: ReasoningPart[] = []
@@ -390,14 +388,10 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
    }
  }
  const text = reasoning.map((part) => part.text).join("")
-  const details = reasoningDetails(reasoning, message.native?.openaiCompatible, options.providerMetadataKey)
-  const observedField = reasoning
-    .map((part) => reasoningField(part, options.providerMetadataKey))
-    .find((value) => value !== undefined)
+  const details = reasoningDetails(reasoning, message.native?.openaiCompatible)
+  const observedField = reasoning.map(reasoningField).find((value) => value !== undefined)
  const nativeReasoning = openAICompatibleReasoningContent(message.native?.openaiCompatible)
-  const fullyStructured = reasoning.every((part) =>
-    Array.isArray(part.providerMetadata?.[options.providerMetadataKey]?.reasoningDetails),
-  )
+  const fullyStructured = reasoning.every((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))
  const field = (() => {
    if (configuredField !== undefined && (requireReasoning || reasoning.length > 0 || nativeReasoning !== undefined))
      return configuredField
@@ -465,7 +459,7 @@ const lowerMessage = Effect.fn("OpenAIChat.lowerMessage")(function* (
  message: OpenAIChatRequestMessage,
  reasoningField: string | undefined,
  requireReasoning: boolean,
-  options: LoweringOptions & { readonly providerMetadataKey: string },
+  options: LoweringOptions,
 ) {
  if (message.role === "user") return [yield* lowerUserMessage(message, options)]
  if (message.role === "assistant")
@@ -501,13 +495,8 @@ 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
-          .replace(/[^a-zA-Z0-9]/g, "")
-          .slice(0, 9)
-          .padEnd(9, "0")
+      if (mistral) return id.replace(/[^a-zA-Z0-9]/g, "").slice(0, 9).padEnd(9, "0")
      if (modelID.includes("claude")) return id.replace(/[^a-zA-Z0-9_-]/g, "_")
      if (request.model.provider === "openai" || request.model.provider === "azure" || modelID.startsWith("openai/"))
        return id.slice(0, 40)
@@ -516,8 +505,7 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
  }
  const requireAssistantAfterTool = request.model.compatibility?.requireAssistantAfterTool ?? mistral
  const bridgeTools = () => {
-    if (requireAssistantAfterTool && messages.at(-1)?.role === "tool")
-      messages.push({ role: "assistant", content: "Done." })
+    if (requireAssistantAfterTool && messages.at(-1)?.role === "tool") messages.push({ role: "assistant", content: "Done." })
  }
  const pendingImages: Array<Schema.Schema.Type<typeof OpenAIChatUserContent>> = []
  const flushImages = () => {
@@ -569,10 +557,7 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
        )
      continue
    }
-    if (
-      message.role === "assistant" &&
-      message.content.every((part) => part.type === "text" && part.text.trim() === "")
-    )
+    if (message.role === "assistant" && message.content.every((part) => part.type === "text" && part.text.trim() === ""))
      continue
    if (message.role === "tool") {
      const lowered = yield* lowerToolMessages(message, lowering)
@@ -603,10 +588,7 @@ const hasToolHistory = (messages: ReadonlyArray<LLMRequest["messages"][number]>)
 // models.dev provider naming: DeepSeek, Moonshot AI, Together AI, ZAI
 // (Zhipu + Coding Plan variants), Nvidia, Cerebras, Chutes, etc. still
 // require `max_tokens`.
-const detectMaxTokensField = (
-  provider: string,
-  baseURL: string | undefined,
-): "max_tokens" | "max_completion_tokens" => {
+const detectMaxTokensField = (provider: string, baseURL: string | undefined): "max_tokens" | "max_completion_tokens" => {
  const p = provider.toLowerCase()
  const url = (baseURL ?? "").toLowerCase()
  if (
@@ -656,8 +638,7 @@ const detectSupportsStore = (provider: string, baseURL: string | undefined): boo
  const isChutes = p === "chutes" || url.includes("chutes.ai")
  const isCloudflareWorkersAI = p === "cloudflare-workers-ai" || url.includes("api.cloudflare.com")
  const isCloudflareAiGateway = p === "cloudflare-ai-gateway" || url.includes("gateway.ai.cloudflare.com")
-  const isVercelAiGateway =
-    p === "vercel-ai-gateway" || url.includes("ai-gateway.vercel.sh") || url.includes("vercel.sh")
+  const isVercelAiGateway = p === "vercel-ai-gateway" || url.includes("ai-gateway.vercel.sh") || url.includes("vercel.sh")
  const isAntLing = p === "ant-ling" || url.includes("api.ant-ling.com")
  const isOpencode = p === "opencode" || url.includes("opencode.ai")
  const isNonStandard =
@@ -689,7 +670,11 @@ const detectSupportsStrictMode = (provider: string, baseURL: string | undefined)
  return !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia
 }

-const detectZaiToolStream = (provider: string, baseURL: string | undefined, modelID: string): boolean => {
+const detectZaiToolStream = (
+  provider: string,
+  baseURL: string | undefined,
+  modelID: string,
+): boolean => {
  const p = provider.toLowerCase()
  const url = (baseURL ?? "").toLowerCase()
  const isZai =
@@ -739,10 +724,10 @@ export const fromRequest = Effect.fn("OpenAIChat.fromRequest")(function* (
  const supportsStore = request.model.compatibility?.supportsStore ?? detectSupportsStore(provider, baseURL)
  const supportsUsageInStreaming =
    request.model.compatibility?.supportsUsageInStreaming ?? detectSupportsUsageInStreaming()
-  const supportsStrictMode =
-    request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
+  const supportsStrictMode = request.model.compatibility?.supportsStrictMode ?? detectSupportsStrictMode(provider, baseURL)
  const zaiToolStream =
-    request.model.compatibility?.zaiToolStream ?? detectZaiToolStream(provider, baseURL, request.model.id)
+    request.model.compatibility?.zaiToolStream ??
+    detectZaiToolStream(provider, baseURL, request.model.id)
  const hasHistory = hasToolHistory(request.messages)
  const hasActiveTools = request.tools.length > 0
  return {
@@ -827,14 +812,15 @@ 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"], providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
  if (!usage) return undefined
  const input = usage.prompt_tokens ?? undefined
  const output = usage.completion_tokens ?? undefined
-  const cached = (usage.prompt_tokens_details?.cached_tokens ??
-    (usage as { prompt_cache_hit_tokens?: number | null }).prompt_cache_hit_tokens ??
-    (usage as { cached_tokens?: number | null }).cached_tokens ??
-    undefined) as number | undefined
+  const cached =
+    (usage.prompt_tokens_details?.cached_tokens ??
+      (usage as { prompt_cache_hit_tokens?: number | null }).prompt_cache_hit_tokens ??
+      (usage as { cached_tokens?: number | null }).cached_tokens ??
+      undefined) as number | undefined
  const cacheWrite = usage.prompt_tokens_details?.cache_write_tokens ?? undefined
  const reasoning = usage.completion_tokens_details?.reasoning_tokens ?? undefined
  const nonCached = ProviderShared.subtractTokens(input, ProviderShared.sumTokens(cached, cacheWrite))
@@ -846,7 +832,7 @@ const mapUsage = (usage: OpenAIChatEvent["usage"], providerMetadataKey: string):
    cacheWriteInputTokens: cacheWrite,
    reasoningTokens: reasoning,
    totalTokens: ProviderShared.totalTokens(input, output, usage.total_tokens ?? undefined),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { openai: usage },
  })
 }

@@ -920,12 +906,8 @@ 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 = (
-  providerMetadataKey: string,
-  field: ParserState["reasoningField"],
-  details?: ReadonlyArray<unknown>,
-) => ({
-  [providerMetadataKey]: {
+const reasoningMetadata = (field: ParserState["reasoningField"], details?: ReadonlyArray<unknown>) => ({
+  openai: {
    ...(field ? { reasoningField: field } : {}),
    ...(details ? { reasoningDetails: details } : {}),
  },
@@ -952,17 +934,15 @@ 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, state.providerMetadataKey) ??
-      (choiceUsage ? mapUsage(choiceUsage, state.providerMetadataKey) : undefined) ??
-      state.usage
+    const usage = mapUsage(event.usage) ?? (choiceUsage ? mapUsage(choiceUsage) : undefined) ?? state.usage
    const rawFinishReason = choice?.finish_reason
-    const finishReason = rawFinishReason
-      ? {
-          normalized: yield* mapFinishReason(event, rawFinishReason),
-          raw: choice?.native_finish_reason ?? rawFinishReason,
-        }
-      : state.finishReason
+    const finishReason =
+      rawFinishReason
+        ? {
+            normalized: yield* mapFinishReason(event, rawFinishReason),
+            raw: choice?.native_finish_reason ?? rawFinishReason,
+          }
+        : state.finishReason
    const delta = choice?.delta
    const toolDeltas = delta?.tool_calls ?? []
    let tools = state.tools
@@ -993,7 +973,7 @@ 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(state.providerMetadataKey, reasoningField)
+    const deltaMetadata = reasoningMetadata(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 (
@@ -1009,11 +989,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        lifecycle,
        events,
        "reasoning-0",
-        reasoningMetadata(
-          state.providerMetadataKey,
-          reasoningField,
-          reasoningDetailsObserved ? state.reasoningDetails : undefined,
-        ),
+        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
      )
      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.content)
    }
@@ -1023,11 +999,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
        lifecycle,
        events,
        "reasoning-0",
-        reasoningMetadata(
-          state.providerMetadataKey,
-          reasoningField,
-          reasoningDetailsObserved ? state.reasoningDetails : undefined,
-        ),
+        reasoningMetadata(reasoningField, reasoningDetailsObserved ? state.reasoningDetails : undefined),
      )
      lifecycle = Lifecycle.textDelta(lifecycle, events, "text-0", delta.refusal)
    }
@@ -1088,7 +1060,6 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>

    return [
      {
-        providerMetadataKey: state.providerMetadataKey,
        tools: finished?.tools ?? tools,
        pendingTools,
        toolCallEvents: finished?.events ?? state.toolCallEvents,
@@ -1132,18 +1103,12 @@ 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.providerMetadataKey, state.reasoningField),
-        )
+      ? Lifecycle.reasoningStart(state.lifecycle, events, "reasoning-0", reasoningMetadata(state.reasoningField))
      : state.lifecycle
  const ended = Lifecycle.reasoningEnd(started, events, "reasoning-0", metadata)
  const lifecycle = toolCallEvents.length ? Lifecycle.stepStart(ended, events) : ended
@@ -1170,7 +1135,6 @@ 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: [],
@@ -86,10 +86,28 @@ const OpenAIResponsesBody = Schema.Struct({
 })
 export type OpenAIResponsesBody = Schema.Schema.Type<typeof OpenAIResponsesBody>

+// Replayed items are paired with stored server state by id, so a foreign or
+// synthetic token can fail request validation even when `call_id` pairing is
+// intact. Only resend ids in each item kind's own grammar; hosted tool
+// items keep generic validation because every hosted tool mints its own
+// prefix. The same allowlist approach codex uses before resending history
+// (codex-rs core/src/client.rs, `prepare_response_items_for_request`).
+const ITEM_ID_PREFIXES: Record<OpenResponses.ItemKind, ReadonlyArray<string>> = {
+  message: ["msg_"],
+  reasoning: ["rs_"],
+  "function-call": ["fc_"],
+  // Every hosted tool mints its own id prefix, so items keep generic validation.
+  "hosted-tool": [],
+}
+
 const extension = {
  id: ADAPTER,
  name: NAME,
  lowerHostedToolItem: (item: unknown) => (Schema.is(OpenAIResponsesHostedToolItem)(item) ? item : undefined),
+  acceptsItemID: (kind: OpenResponses.ItemKind, id: string) => {
+    const prefixes = ITEM_ID_PREFIXES[kind]
+    return prefixes.length === 0 || prefixes.some((prefix) => id.startsWith(prefix))
+  },
 } satisfies OpenResponses.Extension

 const nativeImageToolInput = (tool: ToolDefinition) => {
@@ -184,7 +202,7 @@ const HOSTED_TOOLS = {

 const step = (state: OpenResponses.ParserState, event: OpenResponses.Event) => {
  if (event.type === "response.reasoning_text.delta")
-    return event.item_id !== undefined
+    return event.item_id
      ? Effect.succeed(
          OpenResponses.onReasoningDelta(state, event, OpenResponses.outputItemID(state, event) ?? event.item_id),
        )
@@ -29,9 +29,10 @@ export type ResponseIncludable = (typeof ResponseIncludables)[number] | (string

 export const ServiceTiers = ["auto", "default", "flex", "priority"] as const
 export type ServiceTier = (typeof ServiceTiers)[number] | (string & {})
-export const ServiceTier = Schema.declare<ServiceTier>((value): value is ServiceTier => typeof value === "string", {
-  title: "ServiceTier",
-})
+export const ServiceTier = Schema.declare<ServiceTier>(
+  (value): value is ServiceTier => typeof value === "string",
+  { title: "ServiceTier" },
+)

 export const Truncations = ["auto", "disabled"] as const
 export type Truncation = (typeof Truncations)[number]
@@ -34,35 +34,37 @@ export const onDone: (
  state: OpenResponses.ParserState,
  item: Item,
  tools: Definitions,
-) => Effect.Effect<OpenResponses.StepResult, AIError> = Effect.fn("ResponsesHostedTools.onDone")(
-  function* (state, item, tools) {
-    const tool = tools[item.type]
-    if (!tool) return [state, []] satisfies OpenResponses.StepResult
-    const providerMetadata = OpenResponses.providerMetadata(state, { itemId: item.id })
-    const events: LLMEvent[] = []
-    const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
-    events.push(
-      LLMEvent.toolCall({
-        id: item.id,
-        name: tool.name,
-        input: tool.input(item),
-        providerExecuted: true,
-        providerMetadata,
-      }),
-      LLMEvent.toolResult({
-        id: item.id,
-        name: tool.name,
-        result: tool.result
-          ? yield* tool.result(item)
-          : item.error !== undefined && item.error !== null
-            ? { type: "error", value: item.error }
-            : { type: "json", value: item },
-        providerExecuted: true,
-        providerMetadata,
-      }),
-    )
-    return [{ ...state, lifecycle }, events] satisfies OpenResponses.StepResult
-  },
-)
+) => Effect.Effect<OpenResponses.StepResult, AIError> = Effect.fn("ResponsesHostedTools.onDone")(function* (
+  state,
+  item,
+  tools,
+) {
+  const tool = tools[item.type]
+  if (!tool) return [state, []] satisfies OpenResponses.StepResult
+  const providerMetadata = OpenResponses.providerMetadata(state, { itemId: item.id })
+  const events: LLMEvent[] = []
+  const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
+  events.push(
+    LLMEvent.toolCall({
+      id: item.id,
+      name: tool.name,
+      input: tool.input(item),
+      providerExecuted: true,
+      providerMetadata,
+    }),
+    LLMEvent.toolResult({
+      id: item.id,
+      name: tool.name,
+      result: tool.result
+        ? yield* tool.result(item)
+        : item.error !== undefined && item.error !== null
+          ? { type: "error", value: item.error }
+          : { type: "json", value: item },
+      providerExecuted: true,
+      providerMetadata,
+    }),
+  )
+  return [{ ...state, lifecycle }, events] satisfies OpenResponses.StepResult
+})

 export * as ResponsesHostedTools from "./responses-hosted-tools.js"
@@ -40,16 +40,15 @@ const patterns = [
  /model_context_window_exceeded/i,
  /too many tokens/i,
  /token limit exceeded/i,
-  /request_too_large/i,
 ]

-const payloadPatterns = [/request entity too large/i, /payload too large/i, /request too large/i]
+const payloadPatterns = [/request_too_large/i, /request entity too large/i, /payload too large/i, /request too large/i]

 const exclusions = [/^(throttling error|service unavailable):/i, /rate limit/i, /too many requests/i]

 export const isContextOverflow = (message: string) =>
  !exclusions.some((pattern) => pattern.test(message)) &&
-  (patterns.some((pattern) => pattern.test(message)) || /^4(?:00|13)\s*(status code)?\s*\(no body\)/i.test(message))
+  (patterns.some((pattern) => pattern.test(message)) || /^400\s*(status code)?\s*\(no body\)/i.test(message))

 export const isPayloadTooLarge = (message: string) => payloadPatterns.some((pattern) => pattern.test(message))

@@ -107,7 +106,6 @@ export function classifyProviderFailure(input: ProviderFailure): AIError["reason
    clientScoped &&
    (codes.includes("context_length_exceeded") ||
      codes.includes("model_context_window_exceeded") ||
-      codes.includes("request_too_large") ||
      isContextOverflow(text))
  )
    return new InvalidRequestReason({ ...common, classification: "context-overflow" })
@@ -1,5 +1,5 @@
 import { Auth } from "../route/auth.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
+import type { Route as RouteDef, RouteDefaultsInput } from "../route/client.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { OpenAIChat } from "../protocols/openai-chat.js"
 import { OpenAIResponses } from "../protocols/openai-responses.js"
@@ -23,30 +23,22 @@ 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({
+const responsesRoute = OpenAIResponses.route.with({
  id: "bedrock-mantle-responses",
  provider: id,
-  providerMetadataKey: "mantle",
-  protocol: OpenAIResponses.protocol,
-  endpoint: OpenAIResponses.route.endpoint,
-  auth: OpenAIResponses.route.auth,
-  transport: OpenAIResponses.httpTransport,
-  defaults: OpenAIResponses.route.defaults,
 })

 const chatRoute = OpenAIChat.route.with({
  id: "bedrock-mantle-chat",
  provider: id,
-  providerMetadataKey: "mantle",
 })

 export const routes = [responsesRoute, chatRoute]

-const configuredRoute = <Body, Prepared>(route: Route<Body, Prepared>, input: Config) => {
+const configuredRoute = <Body, Prepared>(route: RouteDef<Body, Prepared>, input: Config) => {
  const region = input.region ?? input.credentials?.region ?? "us-east-1"
  const credentials = input.credentials === undefined ? undefined : { ...input.credentials, region }
  return route.with({
@@ -78,7 +70,7 @@ export const configure = (input: Config = {}) => {

  return {
    id,
-    model: responses,
+    model: chat,
    chat,
    responses,
    configure,
@@ -96,7 +88,6 @@ 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,
@@ -112,4 +103,4 @@ export const responsesModel: ProviderPackage.Definition<Settings, OpenAIProvider
  modelID,
  settings,
 ) => configure(config(settings)).responses(modelID)
-export const model = responsesModel
+export const model = chatModel
@@ -35,7 +35,6 @@ 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),
  })
@@ -1,2 +1,2 @@
-export { responsesModel as model } from "../amazon-bedrock-mantle.js"
+export { chatModel as model } from "../amazon-bedrock-mantle.js"
 export type { Settings } from "../amazon-bedrock-mantle.js"
@@ -1,61 +0,0 @@
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import type { RouteDefaultsInput } from "../route/client.js"
-import { ProviderID, type ModelID } from "../schema/index.js"
-import { profiles } from "./openai-compatible-profile.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("deepinfra")
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: OpenAIProviderOptionsInput
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: OpenAIProviderOptionsInput
-}
-
-export const route = OpenAICompatibleChat.route.with({
-  id: "deepinfra-chat",
-  provider: id,
-  endpoint: { baseURL: profiles.deepinfra.baseURL },
-})
-
-export const routes = [route]
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
-  const root = baseURL?.replace(/\/+$/, "")
-  const configured = route.with({
-    ...defaults,
-    endpoint: {
-      baseURL: root === undefined ? profiles.deepinfra.baseURL : root.endsWith("/openai") ? root : `${root}/openai`,
-    },
-    auth: AuthOptions.bearer(input, "DEEPINFRA_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) =>
-      configured.model<OpenAIProviderOptionsInput>({
-        id: modelID,
-        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning_content", supportsStore: false },
-      }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
@@ -27,7 +27,6 @@ export interface Settings extends ProviderPackage.Settings {
 const route = OpenAICompatibleChat.route.with({
  id: "google-vertex-chat",
  provider: id,
-  providerMetadataKey: "vertex",
 })

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

@@ -68,7 +68,7 @@ const protocol = {
 const route = Route.make({
  id: "google-vertex-gemini",
  provider: id,
-  providerMetadataKey: "vertex",
+  providerMetadataKey: "google",
  protocol,
  endpoint: Endpoint.path(({ request }) => {
    const model = String(request.model.id)
@@ -1,116 +0,0 @@
-import { Effect, Schema } from "effect"
-import type { ProviderPackage } from "../provider-package.js"
-import { OpenAIChat } from "../protocols/openai-chat.js"
-import { ProviderShared } from "../protocols/shared.js"
-import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
-import { Route, type RouteDefaultsInput } from "../route/client.js"
-import { Endpoint } from "../route/endpoint.js"
-import { Framing } from "../route/framing.js"
-import { Protocol } from "../route/protocol.js"
-import { ProviderID, type ModelID, type LLMRequest } from "../schema/index.js"
-import { profiles } from "./openai-compatible-profile.js"
-import type { OpenAIProviderOptionsInput } from "./openai-options.js"
-
-export const id = ProviderID.make("groq")
-
-export type ProviderOptions = Pick<OpenAIProviderOptionsInput, "reasoningEffort"> & {
-  /** Controls visible reasoning on GPT-OSS; other models always use parsed reasoning. */
-  readonly includeReasoning?: boolean
-  readonly parallelToolCalls?: boolean
-  readonly serviceTier?: "on_demand" | "flex" | "auto" | "performance" | (string & {})
-  readonly user?: string
-}
-
-export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
-  ProviderAuthOption<"optional"> & {
-    readonly baseURL?: string
-    readonly providerOptions?: ProviderOptions
-  }
-
-export interface Settings extends ProviderPackage.Settings {
-  readonly apiKey?: string
-  readonly baseURL?: string
-  readonly providerOptions?: ProviderOptions
-}
-
-const Options = Schema.Struct({
-  includeReasoning: Schema.optional(Schema.Boolean),
-  parallelToolCalls: Schema.optional(Schema.Boolean),
-  serviceTier: Schema.optional(Schema.String),
-  user: Schema.optional(Schema.String),
-})
-
-export const protocol = Protocol.make({
-  id: "groq-chat",
-  body: {
-    schema: Schema.Struct({
-      ...OpenAIChat.bodyFields,
-      reasoning_format: Schema.optional(Schema.Literal("parsed")),
-      include_reasoning: Schema.optional(Schema.Boolean),
-      parallel_tool_calls: Schema.optional(Schema.Boolean),
-      service_tier: Schema.optional(Schema.String),
-      user: Schema.optional(Schema.String),
-    }),
-    from: Effect.fn("Groq.fromRequest")(function* (request: LLMRequest) {
-      const options = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(Options))(
-        request.providerOptions ?? {},
-      )
-      const gptOSS = request.model.id.startsWith("openai/gpt-oss-")
-      return {
-        ...(yield* OpenAIChat.fromRequest(request)),
-        reasoning_format: gptOSS ? undefined : ("parsed" as const),
-        include_reasoning: gptOSS ? options.includeReasoning : undefined,
-        parallel_tool_calls: options.parallelToolCalls,
-        service_tier: options.serviceTier,
-        user: options.user,
-      }
-    }),
-  },
-  stream: OpenAIChat.protocol.stream,
-})
-
-export const route = Route.make({
-  id: "groq-chat",
-  provider: id,
-  providerMetadataKey: "openai",
-  protocol,
-  endpoint: Endpoint.path("/chat/completions", { baseURL: profiles.groq.baseURL }),
-  framing: Framing.sse,
-})
-
-export const configure = (input: LanguageModelOptions = {}) => {
-  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
-  const configured = route.with({
-    ...defaults,
-    endpoint: { baseURL: baseURL ?? profiles.groq.baseURL },
-    auth: AuthOptions.bearer(input, "GROQ_API_KEY"),
-  })
-  return {
-    id,
-    model: (modelID: string | ModelID) =>
-      configured.model<ProviderOptions>({
-        id: modelID,
-        compatibility: {
-          maxTokensField: "max_completion_tokens",
-          reasoningField: "reasoning",
-          requireReasoning: false,
-          supportsStore: false,
-          supportsStrictMode: false,
-        },
-      }),
-    configure,
-  }
-}
-
-export const provider = configure()
-
-export const model: ProviderPackage.Definition<Settings, ProviderOptions>["model"] = (modelID, settings) =>
-  configure({
-    apiKey: settings.apiKey,
-    baseURL: settings.baseURL,
-    headers: settings.headers === undefined ? undefined : { ...settings.headers },
-    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
-    providerOptions: settings.providerOptions,
-  }).model(modelID)
-
-export * as Groq from "./groq.js"
@@ -6,13 +6,11 @@ export * as Azure from "./azure.js"
 export * as Cerebras from "./cerebras.js"
 export * as Cloudflare from "./cloudflare.js"
 export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare.js"
-export * as DeepInfra from "./deepinfra.js"
 export * as Google from "./google.js"
 export * as GoogleVertex from "./google-vertex.js"
 export * as GoogleVertexChat from "./google-vertex-chat.js"
 export * as GoogleVertexMessages from "./google-vertex-messages.js"
 export * as GoogleVertexResponses from "./google-vertex-responses.js"
-export * as Groq from "./groq.js"
 export * as OpenAI from "./openai.js"
 export * as OpenAICompatible from "./openai-compatible.js"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
@@ -164,7 +164,6 @@ 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,
@@ -89,7 +89,6 @@ 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>
@@ -290,16 +289,11 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
      defaults: routeInput.defaults ?? {},
      body: protocol.body,
      with: (patch: RoutePatch<Body, Prepared>) => {
-        const { id, provider, providerMetadataKey, auth, transport, endpoint, ...defaults } = patch
+        const { id, provider, 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,
@@ -345,7 +339,9 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
                return onHalt
                  ? parsed.pipe(
                      Stream.concat(
-                        Stream.suspend(() => Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable)))),
+                        Stream.suspend(() =>
+                          Stream.unwrap(onHalt(state).pipe(Effect.map(Stream.fromIterable))),
+                        ),
                      ),
                    )
                  : parsed
@@ -40,6 +40,17 @@ 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)
@@ -136,12 +147,14 @@ 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,
  })

@@ -166,6 +179,7 @@ const statusError =
            request,
            response,
            body: details,
+            requestId: requestId(headers),
            rateLimit,
          }),
        }),
@@ -202,6 +216,7 @@ export const classifyHttpFailure = (input: {
          ? undefined
          : new HttpResponseDetails({ status: input.status, headers: headerDetails(Headers.fromInput(headers)) }),
      ...details,
+      requestId: requestId(headers),
      rateLimit,
    }),
  })
@@ -29,6 +29,7 @@ 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),
 }) {}

@@ -211,28 +211,6 @@ describe("RequestExecutor", () => {
    }).pipe(Effect.provide(responsesLayer([new Response("request too large", { status: 413 })]))),
  )

-  it.effect("classifies Anthropic request_too_large as context overflow", () =>
-    Effect.gen(function* () {
-      const executor = yield* RequestExecutor.Service
-      const error = yield* executor.execute(request).pipe(Effect.flip)
-
-      expectAIError(error)
-      expect(error.reason).toMatchObject({
-        _tag: "InvalidRequest",
-        classification: "context-overflow",
-        http: { response: { status: 413 } },
-      })
-    }).pipe(
-      Effect.provide(
-        responsesLayer([
-          new Response('{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}', {
-            status: 413,
-          }),
-        ]),
-      ),
-    ),
-  )
-
  it.effect("does not classify ordinary invalid requests as context overflow", () =>
    Effect.gen(function* () {
      const executor = yield* RequestExecutor.Service
@@ -327,6 +305,7 @@ 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",
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/chat-streams-text",
    "recordedAt": "2026-08-23T17:21:53.198Z"
  },
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/responses-calls-a-tool",
    "recordedAt": "2026-08-23T17:21:55.170Z"
  },
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/responses-continues-after-a-tool-result",
    "recordedAt": "2026-08-23T17:21:56.397Z"
  },
@@ -1,7 +1,10 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:azure", "provider:azure"],
+    "tags": [
+      "prefix:azure",
+      "provider:azure"
+    ],
    "name": "azure/responses-streams-text",
    "recordedAt": "2026-08-23T17:21:54.158Z"
  },
@@ -2,7 +2,11 @@
  "version": 1,
  "metadata": {
    "model": "openai.gpt-oss-120b",
-    "tags": ["prefix:bedrock-mantle", "provider:amazon-bedrock", "protocol:openai-responses"],
+    "tags": [
+      "prefix:bedrock-mantle",
+      "provider:amazon-bedrock",
+      "protocol:openai-responses"
+    ],
    "name": "bedrock-mantle/streams-text",
    "recordedAt": "2026-08-25T03:29:02.968Z"
  },
@@ -7,7 +7,13 @@
    "route": "cloudflare-workers-ai",
    "transport": "http",
    "model": "@cf/openai/gpt-oss-20b",
-    "tags": ["prefix:cloudflare-workers-ai", "provider:cloudflare-workers-ai", "tool", "tool-call", "golden"]
+    "tags": [
+      "prefix:cloudflare-workers-ai",
+      "provider:cloudflare-workers-ai",
+      "tool",
+      "tool-call",
+      "golden"
+    ]
  },
  "interactions": [
    {
@@ -29,4 +35,4 @@
      }
    }
  ]
-}
+}
@@ -1,32 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "deepinfra",
-    "route": "deepinfra-chat",
-    "transport": "http",
-    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "text", "golden"],
-    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-text",
-    "recordedAt": "2026-08-26T00:34:03.019Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"You are concise.\"},{\"role\":\"user\",\"content\":\"Reply exactly with: Hello!\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":40,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Hello\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"!\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":25,\"total_tokens\":28,\"completion_tokens\":3,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":25,\"total_tokens\":28,\"completion_tokens\":3,\"estimated_cost\":3.46e-6,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,32 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "deepinfra",
-    "route": "deepinfra-chat",
-    "transport": "http",
-    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "tool", "tool-call", "golden"],
-    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-tool-call",
-    "recordedAt": "2026-08-26T00:34:04.173Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_SMfBjXa8eCmHLyjfeyARxe3a\",\"function\":{\"arguments\":\"{\\\"city\\\": \\\"Paris\\\"}\",\"name\":\"get_weather\"},\"type\":\"function\"}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":230,\"total_tokens\":244,\"completion_tokens\":14,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":230,\"total_tokens\":244,\"completion_tokens\":14,\"estimated_cost\":0.000027480000000000005,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,50 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "provider": "deepinfra",
-    "route": "deepinfra-chat",
-    "transport": "http",
-    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "tool", "tool-loop", "golden"],
-    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-tool-loop",
-    "recordedAt": "2026-08-26T00:34:05.656Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"function\":{\"arguments\":\"{\\\"city\\\": \\\"Paris\\\"}\",\"name\":\"get_weather\"},\"type\":\"function\"}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":243,\"total_tokens\":257,\"completion_tokens\":14,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":243,\"total_tokens\":257,\"completion_tokens\":14,\"estimated_cost\":0.000028780000000000002,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    },
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"},{\"role\":\"assistant\",\"content\":null,\"tool_calls\":[{\"id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"}}],\"reasoning_content\":\"\"},{\"role\":\"tool\",\"tool_call_id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"content\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream; charset=utf-8"
-        },
-        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Paris\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\" is\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\" sunny\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\".\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":276,\"total_tokens\":281,\"completion_tokens\":5,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":276,\"total_tokens\":281,\"completion_tokens\":5,\"estimated_cost\":0.0000292,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,7 +1,11 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
+    "tags": [
+      "prefix:google-vertex",
+      "provider:google-vertex",
+      "protocol:gemini"
+    ],
    "name": "google-vertex/calls-a-tool",
    "recordedAt": "2026-08-23T17:21:51.036Z"
  },
@@ -1,7 +1,11 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
+    "tags": [
+      "prefix:google-vertex",
+      "provider:google-vertex",
+      "protocol:gemini"
+    ],
    "name": "google-vertex/continues-after-a-tool-result",
    "recordedAt": "2026-08-23T17:21:51.853Z"
  },
@@ -1,7 +1,11 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:google-vertex", "provider:google-vertex", "protocol:gemini"],
+    "tags": [
+      "prefix:google-vertex",
+      "provider:google-vertex",
+      "protocol:gemini"
+    ],
    "name": "google-vertex/streams-text",
    "recordedAt": "2026-08-23T17:21:50.112Z"
  },
@@ -1,29 +0,0 @@
-{
-  "version": 1,
-  "metadata": {
-    "model": "openai/gpt-oss-20b",
-    "tags": ["prefix:groq-chat", "provider:groq", "protocol:groq-chat", "text", "usage"],
-    "name": "groq-chat/streams-text-with-usage",
-    "recordedAt": "2026-08-26T14:40:09.833Z"
-  },
-  "interactions": [
-    {
-      "transport": "http",
-      "request": {
-        "method": "POST",
-        "url": "https://api.groq.com/openai/v1/chat/completions",
-        "headers": {
-          "content-type": "application/json"
-        },
-        "body": "{\"model\":\"openai/gpt-oss-20b\",\"messages\":[{\"role\":\"user\",\"content\":\"Reply with exactly one word: hello\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"reasoning_effort\":\"low\",\"max_completion_tokens\":512,\"include_reasoning\":false,\"service_tier\":\"on_demand\",\"user\":\"recorded-test\"}"
-      },
-      "response": {
-        "status": 200,
-        "headers": {
-          "content-type": "text/event-stream"
-        },
-        "body": "data: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\"},\"logprobs\":null,\"finish_reason\":null}],\"x_groq\":{\"id\":\"req_01m0z87915eep9bpf10gg7331e\",\"seed\":94036161}}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"hello\"},\"logprobs\":null,\"finish_reason\":null}]}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[{\"index\":0,\"delta\":{},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"x_groq\":{\"id\":\"req_01m0z87915eep9bpf10gg7331e\",\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}}},\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}}}\n\ndata: {\"id\":\"chatcmpl-660048b0-3c99-4215-a582-e116b77eb881\",\"object\":\"chat.completion.chunk\",\"created\":1787755209,\"model\":\"openai/gpt-oss-20b\",\"system_fingerprint\":\"fp_66891002f6\",\"choices\":[],\"usage\":{\"queue_time\":0.10886435,\"prompt_tokens\":78,\"prompt_time\":0.003693734,\"completion_tokens\":20,\"completion_time\":0.020459983,\"total_tokens\":98,\"total_time\":0.024153717,\"completion_tokens_details\":{\"reasoning_tokens\":10}},\"service_tier\":\"on_demand\"}\n\ndata: [DONE]\n\n"
-      }
-    }
-  ]
-}
@@ -1,7 +1,12 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:openai-responses-cache", "provider:openai", "protocol:openai-responses", "cache"],
+    "tags": [
+      "prefix:openai-responses-cache",
+      "provider:openai",
+      "protocol:openai-responses",
+      "cache"
+    ],
    "name": "openai-responses-cache/reports-cached-tokens-on-identical-second-call",
    "recordedAt": "2026-08-25T03:29:25.124Z"
  },
@@ -5,7 +5,14 @@
    "route": "openai-responses",
    "transport": "http",
    "model": "gpt-5.5",
-    "tags": ["prefix:openai-responses", "provider:openai", "flagship", "tool", "tool-loop", "golden"],
+    "tags": [
+      "prefix:openai-responses",
+      "provider:openai",
+      "flagship",
+      "tool",
+      "tool-loop",
+      "golden"
+    ],
    "name": "openai-responses/openai-responses-gpt-5-5-tool-loop",
    "recordedAt": "2026-08-20T06:30:22.262Z"
  },
@@ -1,7 +1,14 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:pdf", "pdf", "provider:openai", "protocol:openai-responses", "tool", "tool-result"],
+    "tags": [
+      "prefix:pdf",
+      "pdf",
+      "provider:openai",
+      "protocol:openai-responses",
+      "tool",
+      "tool-result"
+    ],
    "name": "pdf/openai-tool-result",
    "recordedAt": "2026-08-25T03:29:08.297Z"
  },
@@ -1,7 +1,13 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:pdf", "pdf", "provider:openai", "protocol:openai-responses", "user-input"],
+    "tags": [
+      "prefix:pdf",
+      "pdf",
+      "provider:openai",
+      "protocol:openai-responses",
+      "user-input"
+    ],
    "name": "pdf/openai-user-input",
    "recordedAt": "2026-08-25T03:29:05.645Z"
  },
@@ -1,7 +1,14 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:pdf", "pdf", "provider:xai", "protocol:xai-responses", "tool", "tool-result"],
+    "tags": [
+      "prefix:pdf",
+      "pdf",
+      "provider:xai",
+      "protocol:xai-responses",
+      "tool",
+      "tool-result"
+    ],
    "name": "pdf/xai-tool-result",
    "recordedAt": "2026-08-25T03:29:11.774Z"
  },
@@ -1,7 +1,13 @@
 {
  "version": 1,
  "metadata": {
-    "tags": ["prefix:pdf", "pdf", "provider:xai", "protocol:xai-responses", "user-input"],
+    "tags": [
+      "prefix:pdf",
+      "pdf",
+      "provider:xai",
+      "protocol:xai-responses",
+      "user-input"
+    ],
    "name": "pdf/xai-user-input",
    "recordedAt": "2026-08-25T03:29:10.612Z"
  },
@@ -52,4 +52,4 @@
      }
    }
  ]
-}
+}
@@ -2,7 +2,12 @@
  "version": 1,
  "metadata": {
    "model": "anthropic/claude-sonnet-4.6",
-    "tags": ["prefix:openai-compatible-chat", "provider:vercel-ai-gateway", "protocol:openai-chat", "reasoning"],
+    "tags": [
+      "prefix:openai-compatible-chat",
+      "provider:vercel-ai-gateway",
+      "protocol:openai-chat",
+      "reasoning"
+    ],
    "name": "vercel-ai-gateway-reasoning",
    "recordedAt": "2026-07-18T11:28:42.077Z"
  },
@@ -26,4 +31,4 @@
      }
    }
  ]
-}
+}
@@ -18,19 +18,13 @@ describe("provider error classification", () => {
    expect(messages.every(isContextOverflow)).toBe(true)
  })

-  test("classifies Anthropic request_too_large as recoverable overflow", () => {
-    expect(
+  test("classifies request size failures separately from context overflow", () => {
+    const failures = [
+      classifyProviderFailure({ message: "request too large", status: 413 }),
      classifyProviderFailure({
        message: '{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}',
        status: 400,
      }),
-    ).toMatchObject({ _tag: "InvalidRequest", classification: "context-overflow" })
-    expect(isContextOverflow("413 status code (no body)")).toBe(true)
-  })
-
-  test("classifies generic request size failures separately from context overflow", () => {
-    const failures = [
-      classifyProviderFailure({ message: "request too large", status: 413 }),
      classifyProviderFailure({ message: "upstream request entity too large", status: 502 }),
    ]

@@ -39,6 +33,7 @@ describe("provider error classification", () => {
        expect.objectContaining({ _tag: "InvalidRequest", classification: "payload-too-large" }),
      ),
    )
+    expect(isContextOverflow("413 status code (no body)")).toBe(false)
  })

  test("does not classify rate limits as context overflow", () => {
@@ -89,7 +84,9 @@ describe("provider error classification", () => {

  test("classifies network error text as provider internal", () => {
    expect(
-      ["network error", "network-error", "network_error"].map((message) => classifyProviderFailure({ message })._tag),
+      ["network error", "network-error", "network_error"].map(
+        (message) => classifyProviderFailure({ message })._tag,
+      ),
    ).toEqual(["ProviderInternal", "ProviderInternal", "ProviderInternal"])
  })

@@ -28,8 +28,6 @@ describe("provider package entrypoints", () => {
      import("@opencode-ai/ai/providers/amazon-bedrock/mantle/responses"),
      import("@opencode-ai/ai/providers/togetherai"),
      import("@opencode-ai/ai/providers/cerebras"),
-      import("@opencode-ai/ai/providers/deepinfra"),
-      import("@opencode-ai/ai/providers/groq"),
    ])

    for (const module of modules) expect(module.model).toBeFunction()
@@ -39,24 +37,6 @@ describe("provider package entrypoints", () => {
    expect(modules[19].model).toBe(modules[20].model)
  })

-  test("maps DeepInfra package settings onto its native executable model", async () => {
-    const DeepInfra = await import("@opencode-ai/ai/providers/deepinfra")
-    const settings = {
-      apiKey: "fixture",
-      baseURL: "https://provider.example.test/v1/",
-      headers: { "x-application": "opencode" },
-      body: { service_tier: "priority" },
-      providerOptions: { reasoningEffort: "high" as const },
-    }
-    const deepinfra = DeepInfra.model("google/gemma-3-27b-it", settings)
-
-    expect(deepinfra.route.id).toBe("deepinfra-chat")
-    expect(deepinfra.route.endpoint.baseURL).toBe("https://provider.example.test/v1/openai")
-    expect(deepinfra.route.defaults.providerOptions).toEqual(settings.providerOptions)
-    expect(deepinfra.route.defaults.headers).toEqual(settings.headers)
-    expect(deepinfra.route.defaults.http?.body).toEqual(settings.body)
-  })
-
  test("maps OpenRouter and xAI package settings onto executable models", async () => {
    const OpenRouter = await import("@opencode-ai/ai/providers/openrouter")
    const XAI = await import("@opencode-ai/ai/providers/xai")
@@ -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, Endpoint, LLMClient, Route } from "../../src/route.js"
+import { Auth, LLMClient } 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"
@@ -810,99 +810,6 @@ 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(
@@ -569,57 +569,6 @@ 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(
@@ -1,13 +1,11 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
-import { LLM, Message } from "../../src/index.js"
+import { LLM } 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"
-import { dynamicResponse, fixedResponse } from "../lib/http.js"
+import { dynamicResponse } from "../lib/http.js"
 import { sseEvents } from "../lib/sse.js"
 import { recordedTests } from "../recorded-test.js"

@@ -18,16 +16,12 @@ const credentials = {
 }

 describe("Amazon Bedrock Mantle provider", () => {
-  it.effect("uses Responses by default and exposes Chat explicitly", () =>
+  it.effect("uses Chat by default and exposes Responses", () =>
    Effect.gen(function* () {
      const provider = AmazonBedrockMantle.configure({ credentials })
-      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 chat = yield* compileRequest(LLM.request({ model: provider.model("openai.gpt-oss-120b"), prompt: "Hi" }))
      const responses = yield* compileRequest(
-        LLM.request({ model: provider.model("openai.gpt-oss-120b"), prompt: "Hi" }),
+        LLM.request({ model: provider.responses("openai.gpt-oss-120b"), prompt: "Hi" }),
      )

      expect(chat).toMatchObject({
@@ -40,23 +34,6 @@ 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)
    }),
  )

@@ -106,42 +83,6 @@ describe("Amazon Bedrock Mantle provider", () => {
      expect(seen).toEqual([{ url: "https://mantle.test/v1/chat/completions", authorization: "Bearer test-key" }])
    }),
  )
-
-  it.effect("replays reasoning with Mantle's message-prefixed item ids", () =>
-    Effect.gen(function* () {
-      const model = AmazonBedrockMantle.configure({ apiKey: "test-key" }).responses("openai.gpt-oss-120b")
-      const item = { type: "reasoning", id: "msg_95d4d0af4350432a", encrypted_content: "mantle-state" }
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think." })).pipe(
-        Effect.provide(
-          fixedResponse(
-            sseEvents(
-              { type: "response.output_item.added", item },
-              { type: "response.reasoning_summary_text.delta", item_id: item.id, delta: "Considering." },
-              { type: "response.output_item.done", item },
-              { type: "response.completed", response: { id: "resp_1" } },
-            ),
-          ),
-        ),
-      )
-
-      const prepared = yield* compileRequest(
-        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",
-          id: "msg_95d4d0af4350432a",
-          summary: [{ type: "summary_text", text: "Considering." }],
-          encrypted_content: "mantle-state",
-        },
-        { role: "user", content: [{ type: "input_text", text: "Continue." }] },
-      ])
-    }),
-  )
 })

 const recorded = recordedTests({
@@ -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({
-        "cloudflare-ai-gateway": { reasoningField: "reasoning", reasoningDetails: merged },
+        openai: { reasoningField: "reasoning", reasoningDetails: merged },
      })

      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
@@ -515,10 +515,7 @@ describe("Gemini route", () => {
        {
          role: "model",
          parts: [
-            {
-              functionCall: { id: "call_image", name: "read", args: { path: "pixel.png" } },
-              thoughtSignature: "sig_1",
-            },
+            { functionCall: { id: "call_image", name: "read", args: { path: "pixel.png" } }, thoughtSignature: "sig_1" },
          ],
        },
        {
@@ -609,7 +606,10 @@ describe("Gemini route", () => {
      expect(prepared.body.contents).toEqual([
        {
          role: "model",
-          parts: [{ functionCall: { name: "shot", args: {} } }, { functionCall: { name: "shot", args: {} } }],
+          parts: [
+            { functionCall: { name: "shot", args: {} } },
+            { functionCall: { name: "shot", args: {} } },
+          ],
        },
        {
          role: "user",
@@ -1071,9 +1071,7 @@ describe("Gemini route", () => {
      const prepared = yield* compileRequest(
        LLM.request({
          model,
-          messages: [
-            Message.assistant([{ type: "text", text: "All done.", providerMetadata: delta?.providerMetadata }]),
-          ],
+          messages: [Message.assistant([{ type: "text", text: "All done.", providerMetadata: delta?.providerMetadata }])],
        }),
      )
      expect(prepared.body.contents).toEqual([
@@ -1574,7 +1572,9 @@ describe("Gemini route", () => {
              { candidates: [{ content: { role: "model", parts: null } }] },
              { candidates: [{ content: null, finishReason: null }] },
              {
-                candidates: [{ content: { role: "model", parts: [{ text: "Hello" }] }, finishReason: "STOP" as const }],
+                candidates: [
+                  { content: { role: "model", parts: [{ text: "Hello" }] }, finishReason: "STOP" as const },
+                ],
              },
            ),
          ),
@@ -1,6 +1,6 @@
 import * as Anthropic from "../../src/providers/anthropic.js"
 import * as AnthropicCompatible from "../../src/providers/anthropic-compatible.js"
-import { Cerebras, DeepInfra, TogetherAI } from "../../src/providers/index.js"
+import { Cerebras, TogetherAI } from "../../src/providers/index.js"
 import { CloudflareAIGateway, CloudflareWorkersAI } from "../../src/providers/cloudflare.js"
 import * as Google from "../../src/providers/google.js"
 import * as OpenAI from "../../src/providers/openai.js"
@@ -55,9 +55,6 @@ const cerebras = Cerebras.configure({ apiKey: process.env.CEREBRAS_API_KEY ?? "f
 const groq = OpenAICompatible.groq
  .configure({ apiKey: process.env.GROQ_API_KEY ?? "fixture" })
  .model("llama-3.3-70b-versatile")
-const deepInfra = DeepInfra.configure({ apiKey: process.env.DEEPINFRA_API_KEY ?? "fixture" }).model(
-  "meta-llama/Llama-3.3-70B-Instruct-Turbo",
-)
 const openRouter = OpenRouter.configure({ apiKey: process.env.OPENROUTER_API_KEY ?? "fixture" })
 const openrouter = openRouter.model("openai/gpt-4o-mini")
 const openrouterGpt55 = openRouter.model("openai/gpt-5.5")
@@ -225,13 +222,6 @@ describeRecordedGoldenScenarios([
    requires: ["GROQ_API_KEY"],
    scenarios: ["text", "tool-call", { id: "tool-loop", timeout: 30_000 }],
  },
-  {
-    name: "DeepInfra Llama 3.3 70B",
-    prefix: "deepinfra-chat",
-    model: deepInfra,
-    requires: ["DEEPINFRA_API_KEY"],
-    scenarios: ["text", "tool-call", { id: "tool-loop", timeout: 30_000 }],
-  },
  {
    name: "OpenRouter gpt-4o-mini",
    prefix: "openai-compatible-chat",
@@ -26,7 +26,9 @@ const recorded = recordedTests({
 describe("Google Vertex Gemini recorded", () => {
  recorded.effect("streams text", () =>
    Effect.gen(function* () {
-      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Reply with exactly one word: hello" }))
+      const response = yield* LLMClient.generate(
+        LLM.request({ model, prompt: "Reply with exactly one word: hello" }),
+      )

      expect(response.text.toLowerCase()).toContain("hello")
    }),
@@ -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, fixedResponse } from "../lib/http.js"
+import { dynamicResponse } 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: { vertex: { functionCallId: "provider_call_1" } },
+                providerMetadata: { google: { functionCallId: "provider_call_1" } },
              }),
            ]),
            Message.tool({
@@ -97,7 +97,7 @@ describe("Google Vertex providers", () => {
              name: "lookup",
              result: "sunny",
              resultType: "text",
-              providerMetadata: { vertex: { functionCallId: "provider_call_1" } },
+              providerMetadata: { google: { functionCallId: "provider_call_1" } },
            }),
          ],
        }),
@@ -122,91 +122,6 @@ 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({
@@ -1,185 +0,0 @@
-import { configure } from "@opencode-ai/ai/providers/groq"
-import { describe, expect } from "bun:test"
-import { Effect } from "effect"
-import { LLM, LLMEvent, LLMRequest, LLMResponse, Message, ToolChoice, ToolDefinition } from "../../src/index.js"
-import { LLMClient } from "../../src/route.js"
-import { compileRequest } from "../../src/route/client.js"
-import { recordedTests } from "../recorded-test.js"
-
-const apiKey = process.env.GROQ_API_KEY ?? "fixture"
-const recorded = recordedTests({
-  prefix: "groq-chat",
-  provider: "groq",
-  protocol: "groq-chat",
-  requires: ["GROQ_API_KEY"],
-})
-
-const weather = ToolDefinition.make({
-  name: "lookup_weather",
-  description: "Look up the current weather for a city",
-  inputSchema: {
-    type: "object",
-    properties: { city: { type: "string", enum: ["Paris", "London"] } },
-    required: ["city"],
-    additionalProperties: false,
-  },
-})
-
-describe("Groq recorded", () => {
-  recorded.effect.with(
-    "streams text with usage",
-    { tags: ["text", "usage"], metadata: { model: "openai/gpt-oss-20b" } },
-    () =>
-      Effect.gen(function* () {
-        const request = LLM.request({
-          model: configure({
-            apiKey,
-            providerOptions: {
-              includeReasoning: false,
-              reasoningEffort: "low",
-              serviceTier: "on_demand",
-              user: "recorded-test",
-            },
-          }).model("openai/gpt-oss-20b"),
-          prompt: "Reply with exactly one word: hello",
-          generation: { maxTokens: 512 },
-        })
-        const compiled = yield* compileRequest(request)
-        expect(compiled.body).toMatchObject({
-          max_completion_tokens: 512,
-          stream_options: { include_usage: true },
-          include_reasoning: false,
-          service_tier: "on_demand",
-          user: "recorded-test",
-        })
-        expect(compiled.body.max_tokens).toBeUndefined()
-        expect(compiled.body.store).toBeUndefined()
-        expect(compiled.body.reasoning_format).toBeUndefined()
-
-        const response = yield* LLMClient.generate(request)
-        expect(response.text.toLowerCase().trim()).toBe("hello")
-        expect(response.reasoning).toBe("")
-        expect(response.events.some(LLMEvent.is.textDelta)).toBe(true)
-        expectUsage(response)
-      }),
-    60_000,
-  )
-
-  for (const item of [
-    {
-      name: "continues Qwen parallel tool calls",
-      model: configure({ apiKey, providerOptions: { parallelToolCalls: true, reasoningEffort: "none" } }).model(
-        "qwen/qwen3.6-27b",
-      ),
-      cities: ["Paris", "London"],
-      reasoning: false,
-    },
-    {
-      name: "replays GPT OSS reasoning through a tool loop",
-      model: configure({ apiKey, providerOptions: { includeReasoning: true, reasoningEffort: "low" } }).model(
-        "openai/gpt-oss-20b",
-      ),
-      cities: ["Paris"],
-      reasoning: true,
-    },
-  ]) {
-    recorded.effect.with(
-      item.name,
-      {
-        tags: ["tool", "tool-loop", "usage", item.reasoning ? "reasoning" : "parallel"],
-        metadata: { model: item.model.id },
-      },
-      () =>
-        Effect.gen(function* () {
-          const request = LLM.request({
-            model: item.model,
-            prompt: `Look up the current weather in ${item.cities.join(" and ")}. Call lookup_weather once for each city in the same response before answering. After receiving all results, report each city's weather in one short sentence.`,
-            tools: [weather],
-            toolChoice: "required",
-            generation: { maxTokens: 1536 },
-          })
-          const compiled = yield* compileRequest(request)
-          expect(compiled.body.stream_options).toEqual({ include_usage: true })
-          expect(compiled.body.store).toBeUndefined()
-          expect(compiled.body.reasoning_format).toBe(item.reasoning ? undefined : "parsed")
-          expect(compiled.body.tools[0].function.strict).toBeUndefined()
-          if (!item.reasoning) expect(compiled.body.parallel_tool_calls).toBe(true)
-
-          const first = yield* LLMClient.generate(request)
-          expect(first.finishReason.normalized).toBe("tool-calls")
-          expect(first.toolCalls).toHaveLength(item.cities.length)
-          expect(new Set(first.toolCalls.map((call) => call.id)).size).toBe(item.cities.length)
-          expect(first.toolCalls.map((call) => call.input)).toEqual(
-            expect.arrayContaining(item.cities.map((city) => ({ city }))),
-          )
-          expect(first.toolCalls.every((call) => call.name === "lookup_weather")).toBe(true)
-          expectUsage(first)
-          if (item.reasoning) {
-            expect(first.reasoning.length).toBeGreaterThan(0)
-            expect(first.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
-          }
-
-          const followUp = LLMRequest.update(request, {
-            toolChoice: ToolChoice.make("none"),
-            messages: [
-              ...request.messages,
-              first.message,
-              ...first.toolCalls.map((call) =>
-                Message.tool({ id: call.id, name: call.name, result: { condition: "sunny", temperature: "18C" } }),
-              ),
-            ],
-          })
-          const replay = yield* compileRequest(followUp)
-          if (item.reasoning) {
-            expect(replay.body.messages).toEqual(
-              expect.arrayContaining([expect.objectContaining({ role: "assistant", reasoning: first.reasoning })]),
-            )
-          }
-          expect(replay.body.reasoning_format).toBe(item.reasoning ? undefined : "parsed")
-
-          const second = yield* LLMClient.generate(followUp)
-          expect(second.finishReason.normalized).toBe("stop")
-          expect(second.toolCalls).toHaveLength(0)
-          expect(second.text.toLowerCase()).toContain("sunny")
-          item.cities.forEach((city) => expect(second.text).toContain(city))
-          expectUsage(second)
-        }),
-      60_000,
-    )
-  }
-
-  recorded.effect.with(
-    "streams Qwen parsed reasoning",
-    { tags: ["reasoning", "usage"], metadata: { model: "qwen/qwen3.6-27b" } },
-    () =>
-      Effect.gen(function* () {
-        const request = LLM.request({
-          model: configure({
-            apiKey,
-            providerOptions: { reasoningEffort: "default" },
-          }).model("qwen/qwen3.6-27b"),
-          prompt:
-            "What is 173 multiplied by 219? Think through the arithmetic, then reply with only the final integer.",
-          generation: { maxTokens: 2048 },
-        })
-        const compiled = yield* compileRequest(request)
-        expect(compiled.body).toMatchObject({ reasoning_format: "parsed", reasoning_effort: "default" })
-        expect(compiled.body.include_reasoning).toBeUndefined()
-
-        const response = yield* LLMClient.generate(request)
-        expect(response.text.replaceAll(",", "").trim()).toBe("37887")
-        expect(response.text).not.toContain("<think>")
-        expect(response.reasoning.length).toBeGreaterThan(0)
-        expect(response.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
-        expectUsage(response)
-      }),
-    60_000,
-  )
-})
-
-function expectUsage(response: LLMResponse) {
-  expect(response.usage).toBeDefined()
-  expect(response.usage?.inputTokens).toBeGreaterThan(0)
-  expect(response.usage?.outputTokens).toBeGreaterThan(0)
-  expect(response.events.filter(LLMEvent.is.finish)).toHaveLength(1)
-}
@@ -1,112 +0,0 @@
-import { expect } from "bun:test"
-import { Effect } from "effect"
-import { LanguageModel, LLM, Message } from "../../src/index.js"
-import { OpenAIChat } from "../../src/protocols/openai-chat.js"
-import { Groq } from "../../src/providers/groq.js"
-import { compileRequest } from "../../src/route/client.js"
-import { it } from "../lib/effect.js"
-import { weatherTool } from "../recorded-scenarios.js"
-
-it.effect("Groq reuses Chat streaming and defaults to parsed reasoning", () =>
-  Effect.gen(function* () {
-    expect(Groq.protocol.stream).toBe(OpenAIChat.protocol.stream)
-    const model = Groq.configure({ apiKey: "fixture" }).model("llama-3.3-70b-versatile")
-    expect(model.route.endpoint.baseURL).toBe("https://api.groq.com/openai/v1")
-    const compiled = yield* compileRequest(
-      LLM.request({ model, prompt: "Hello", tools: [weatherTool], generation: { maxTokens: 64 } }),
-    )
-    expect(compiled.body).toMatchObject({
-      max_completion_tokens: 64,
-      stream_options: { include_usage: true },
-      reasoning_format: "parsed",
-    })
-    for (const key of ["store", "max_tokens", "include_reasoning", "parallel_tool_calls", "service_tier", "user"])
-      expect(compiled.body[key]).toBeUndefined()
-    expect(compiled.body.tools?.[0]?.function).not.toHaveProperty("strict")
-  }),
-)
-
-it.effect("Groq lowers its own options for custom catalog identities and endpoints", () =>
-  Effect.gen(function* () {
-    const model = LanguageModel.update(
-      Groq.model("qwen/qwen3.6-27b", {
-        apiKey: "fixture",
-        baseURL: "https://gateway.example/v1",
-        headers: { "x-client": "test" },
-        body: { custom: "value" },
-        providerOptions: {
-          reasoningEffort: "default",
-          parallelToolCalls: true,
-          serviceTier: "flex",
-          user: "test-user",
-        },
-      }),
-      { provider: "custom-groq" },
-    )
-    const compiled = yield* compileRequest(
-      LLM.request({ model, prompt: "Hello", providerOptions: { parallelToolCalls: false, includeReasoning: false } }),
-    )
-    expect(model.route.endpoint.baseURL).toBe("https://gateway.example/v1")
-    expect(model.route.defaults.headers).toEqual({ "x-client": "test" })
-    expect(model.route.defaults.http?.body).toEqual({ custom: "value" })
-    expect(compiled.body).toMatchObject({
-      reasoning_effort: "default",
-      reasoning_format: "parsed",
-      parallel_tool_calls: false,
-      service_tier: "flex",
-      user: "test-user",
-    })
-    expect(compiled.body.include_reasoning).toBeUndefined()
-    for (const key of ["reasoningFormat", "reasoningEffort", "parallelToolCalls", "serviceTier"])
-      expect(compiled.body).not.toHaveProperty(key)
-  }),
-)
-
-it.effect("Groq replays reasoning only when present and preserves explicit reasoning exclusion", () =>
-  Effect.gen(function* () {
-    const compiled = yield* compileRequest(
-      LLM.request({
-        model: Groq.configure({ apiKey: "fixture" }).model("openai/gpt-oss-20b"),
-        messages: [
-          Message.user("Think"),
-          Message.assistant([
-            { type: "reasoning", text: "Thinking" },
-            { type: "text", text: "Answer" },
-          ]),
-          Message.user("Again"),
-          Message.assistant("Answer only"),
-          Message.user("Continue"),
-        ],
-        providerOptions: { reasoningEffort: "low", includeReasoning: false },
-      }),
-    )
-    expect(compiled.body).toMatchObject({ reasoning_effort: "low", include_reasoning: false })
-    expect(compiled.body.reasoning_format).toBeUndefined()
-    expect(compiled.body.messages[1]).toMatchObject({ reasoning: "Thinking", content: "Answer" })
-    expect(compiled.body.messages[1]).not.toHaveProperty("reasoning_content")
-    expect(compiled.body.messages[3]).not.toHaveProperty("reasoning")
-  }),
-)
-
-it.effect("Groq omits reasoning_format for the GPT-OSS family by default", () =>
-  Effect.gen(function* () {
-    for (const id of ["openai/gpt-oss-20b", "openai/gpt-oss-120b", "openai/gpt-oss-safeguard-20b"]) {
-      const compiled = yield* compileRequest(
-        LLM.request({ model: Groq.configure({ apiKey: "fixture" }).model(id), prompt: "Hello" }),
-      )
-      expect(compiled.body.reasoning_format).toBeUndefined()
-      expect(compiled.body.include_reasoning).toBeUndefined()
-    }
-  }),
-)
-
-it.effect("Groq validates option types", () =>
-  Effect.gen(function* () {
-    for (const providerOptions of [{ includeReasoning: "false" }, { parallelToolCalls: "false" }]) {
-      const error = yield* compileRequest(
-        LLM.request({ model: Groq.configure({ apiKey: "fixture" }).model("qwen"), prompt: "Hello", providerOptions }),
-      ).pipe(Effect.flip)
-      expect(error.reason._tag).toBe("InvalidRequest")
-    }
-  }),
-)
@@ -2,80 +2,13 @@ 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 {
-  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 { Cerebras, TogetherAI } 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")
@@ -96,18 +29,6 @@ describe("native OpenAI-compatible providers", () => {
    }),
  )

-  it.effect("preserves native DeepInfra provider and route identity", () =>
-    Effect.gen(function* () {
-      const deepinfra = DeepInfra.configure({ apiKey: "fixture" }).model("google/gemma-3-27b-it")
-      expect(deepinfra).toMatchObject({
-        provider: "deepinfra",
-        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning_content", supportsStore: false },
-        route: { id: "deepinfra-chat", protocol: "openai-chat" },
-      })
-      expect(deepinfra.route.endpoint.baseURL).toBe("https://api.deepinfra.com/v1/openai")
-    }),
-  )
-
  it.effect("applies native provider request defaults even with a custom gateway URL", () =>
    Effect.gen(function* () {
      const together = yield* compileRequest(
@@ -161,21 +82,6 @@ describe("native OpenAI-compatible providers", () => {
    }),
  )

-  it.effect("normalizes DeepInfra API roots without duplicating the OpenAI path", () =>
-    Effect.gen(function* () {
-      for (const baseURL of [
-        "https://gateway.example/v1",
-        "https://gateway.example/v1/",
-        "https://gateway.example/v1/openai",
-        "https://gateway.example/v1/openai/",
-      ]) {
-        expect(DeepInfra.configure({ apiKey: "fixture", baseURL }).model("gemma").route.endpoint.baseURL).toBe(
-          "https://gateway.example/v1/openai",
-        )
-      }
-    }),
-  )
-
  it.effect("maps package settings onto native executable models", () =>
    Effect.gen(function* () {
      for (const native of [TogetherAI, Cerebras]) {
@@ -216,18 +122,6 @@ describe("native OpenAI-compatible providers", () => {
          token: "cerebras-secret",
          url: "https://api.cerebras.ai/v1/chat/completions",
        },
-        {
-          model: DeepInfra.configure().model("gemma"),
-          env: { DEEPINFRA_API_KEY: "deepinfra-secret" },
-          token: "deepinfra-secret",
-          url: "https://api.deepinfra.com/v1/openai/chat/completions",
-        },
-        {
-          model: Groq.configure().model("llama"),
-          env: { GROQ_API_KEY: "groq-secret" },
-          token: "groq-secret",
-          url: "https://api.groq.com/openai/v1/chat/completions",
-        },
      ]

      yield* Effect.forEach(scenarios, (scenario) =>
@@ -68,13 +68,11 @@ 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?.[
-            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)
+          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)
          if (!item.structured) return
-          const details = metadata?.reasoningDetails
+          const details = metadata?.openai?.reasoningDetails
          if (!Array.isArray(details)) return
          expect(
            details.some(
@@ -128,11 +126,7 @@ for (const item of cases) {
          ).toMatch(/^Paris is sunny\.?$/)
          const details = events
            .filter(LLMEvent.is.reasoningEnd)
-            .map(
-              (event) =>
-                event.providerMetadata?.[item.model.route.providerMetadataKey ?? String(item.model.provider)]
-                  ?.reasoningDetails,
-            )
+            .map((event) => event.providerMetadata?.openai?.reasoningDetails)
            .find(Array.isArray)
          expect(Array.isArray(details)).toBe(item.structured)
          if (!item.structured || !Array.isArray(details)) return
@@ -903,70 +903,6 @@ 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" } })
@@ -437,7 +437,7 @@ describe("OpenAI-compatible Chat route", () => {
        outputTokens: undefined,
        totalTokens: undefined,
        providerMetadata: {
-          deepseek: {
+          openai: {
            prompt_tokens: null,
            completion_tokens: null,
            total_tokens: null,
@@ -195,19 +195,15 @@ describe("Open Responses-compatible route", () => {
          model,
          messages: [
            Message.assistant([
-              { type: "text", text: "Kept.", providerMetadata: { "openai-compatible": { itemId: "history_1" } } },
+              // The baseline does not enforce a provider id grammar, so a
+              // non-OpenAI but well-formed token is resent as-is.
+              { type: "text", text: "Kept.", providerMetadata: { openresponses: { itemId: "history_1" } } },
+              // Shape violations are dropped even without a grammar policy.
              {
                type: "text",
-                text: "Long.",
-                providerMetadata: { "openai-compatible": { itemId: `history_${"a".repeat(64)}` } },
+                text: "Dropped.",
+                providerMetadata: { openresponses: { itemId: `m${"a".repeat(64)}` } },
              },
-              {
-                type: "text",
-                text: "Opaque.",
-                providerMetadata: { "openai-compatible": { itemId: "provider_value/with+symbols" } },
-              },
-              { type: "text", text: "No suffix.", providerMetadata: { "openai-compatible": { itemId: "msg_" } } },
-              { type: "text", text: "No prefix.", providerMetadata: { "openai-compatible": { itemId: "_item" } } },
            ]),
          ],
        }),
@@ -222,23 +218,8 @@ describe("Open Responses-compatible route", () => {
        },
        {
          type: "message",
-          id: `history_${"a".repeat(64)}`,
          role: "assistant",
-          content: [{ type: "output_text", text: "Long." }],
-        },
-        {
-          type: "message",
-          id: "provider_value/with+symbols",
-          role: "assistant",
-          content: [{ type: "output_text", text: "Opaque." }],
-        },
-        {
-          type: "message",
-          role: "assistant",
-          content: [
-            { type: "output_text", text: "No suffix." },
-            { type: "output_text", text: "No prefix." },
-          ],
+          content: [{ type: "output_text", text: "Dropped." }],
        },
      ])
    }),
@@ -267,7 +248,7 @@ describe("Open Responses-compatible route", () => {
              name: item.type,
              result: { type: "json", value: item },
              providerExecuted: true,
-              providerMetadata: { example: { itemId: item.id } },
+              providerMetadata: { openresponses: { itemId: item.id } },
            }),
          ),
        }),
@@ -302,296 +283,11 @@ describe("Open Responses-compatible route", () => {
      )

      expect(response.message.content).toEqual([
-        { type: "text", text: "Indexed", providerMetadata: { "openai-compatible": { itemId: "msg_1" } } },
+        { type: "text", text: "Indexed", providerMetadata: { openresponses: { 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({
@@ -666,7 +362,7 @@ describe("Open Responses-compatible route", () => {

      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
        input: { query: "complete" },
-        providerMetadata: { example: { itemId: "item_1" } },
+        providerMetadata: { openresponses: { itemId: "item_1" } },
      })
    }),
  )
@@ -698,7 +394,7 @@ describe("Open Responses-compatible route", () => {
      )

      expect(response.events.find((event) => event.type === "reasoning-end")).toMatchObject({
-        providerMetadata: { "openai-compatible": { itemId: "rs_raw", reasoningEncryptedContent: "raw-state" } },
+        providerMetadata: { openresponses: { itemId: "rs_raw", reasoningEncryptedContent: "raw-state" } },
      })
    }),
  )
@@ -747,7 +443,7 @@ describe("Open Responses-compatible route", () => {
            Message.assistant({
              type: "text",
              text: "Unclassified.",
-              providerMetadata: { "openai-compatible": { phase: null } },
+              providerMetadata: { openresponses: { phase: null } },
            }),
          ],
        }),
@@ -806,7 +502,7 @@ describe("Open Responses-compatible route", () => {
        {
          type: "text",
          text: "I can't help with that.",
-          providerMetadata: { example: { itemId: "msg_refusal" } },
+          providerMetadata: { openresponses: { itemId: "msg_refusal" } },
        },
      ])

@@ -895,7 +591,7 @@ describe("Open Responses-compatible route", () => {

      expect(response.toolCalls).toEqual([])
      expect(response.events.find(LLMEvent.is.finish)).toMatchObject({
-        providerMetadata: { example: { responseId: "resp_1" } },
+        providerMetadata: { openresponses: { responseId: "resp_1" } },
      })
    }),
  )
@@ -2229,35 +2229,6 @@ 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(
@@ -3179,50 +3150,38 @@ describe("OpenAI Responses route", () => {
    }),
  )

-  it.effect("preserves provider-issued item ids and removes malformed ids without dropping items", () =>
+  it.effect("drops replayed item ids outside the server's grammar", () =>
    Effect.gen(function* () {
      const prepared = yield* compileRequest(
        LLM.request({
          model,
          messages: [
            Message.assistant([
+              // Fails the message id prefix.
              {
                type: "text",
                text: "Hello",
                providerMetadata: { openai: { itemId: "history_1" } },
              },
+              // Oversized for the Responses item id limit.
              {
                type: "text",
                text: "World",
-                providerMetadata: { openai: { itemId: `message_${"a".repeat(64)}` } },
+                providerMetadata: { openai: { itemId: `m${"a".repeat(64)}` } },
              },
+              // Fails the reasoning id prefix, so the whole item is unreplayable
+              // statelessly and is skipped rather than sent malformed.
              {
                type: "reasoning",
                text: "Checked the diff.",
                providerMetadata: { openai: { itemId: "thinking_1", reasoningEncryptedContent: "encrypted-state" } },
              },
-              {
-                type: "reasoning",
-                text: "Missing suffix.",
-                providerMetadata: { openai: { itemId: "rs_", reasoningEncryptedContent: "another-state" } },
-              },
-              {
-                type: "reasoning",
-                text: "No prefix separator.",
-                providerMetadata: { openai: { itemId: "550e8400-e29b-41d4-a716-446655440000" } },
-              },
              ToolCallPart.make({
                id: "call_1",
                name: "lookup",
                input: { query: "weather" },
                providerMetadata: { openai: { itemId: "toolu_01A" } },
              }),
-              ToolCallPart.make({
-                id: "call_2",
-                name: "lookup",
-                input: { query: "news" },
-                providerMetadata: { openai: { itemId: "fc_" } },
-              }),
            ]),
          ],
        }),
@@ -3231,44 +3190,18 @@ describe("OpenAI Responses route", () => {
      expect(prepared.body.input).toEqual([
        {
          type: "message",
-          id: "history_1",
          role: "assistant",
-          content: [{ type: "output_text", text: "Hello" }],
-        },
-        {
-          type: "message",
-          id: `message_${"a".repeat(64)}`,
-          role: "assistant",
-          content: [{ type: "output_text", text: "World" }],
-        },
-        {
-          type: "reasoning",
-          id: "thinking_1",
-          summary: [{ type: "summary_text", text: "Checked the diff." }],
-          encrypted_content: "encrypted-state",
-        },
-        {
-          type: "reasoning",
-          summary: [{ type: "summary_text", text: "Missing suffix." }],
-          encrypted_content: "another-state",
-        },
-        {
-          type: "reasoning",
-          summary: [{ type: "summary_text", text: "No prefix separator." }],
+          content: [
+            { type: "output_text", text: "Hello" },
+            { type: "output_text", text: "World" },
+          ],
        },
        {
          type: "function_call",
-          id: "toolu_01A",
          call_id: "call_1",
          name: "lookup",
          arguments: '{"query":"weather"}',
        },
-        {
-          type: "function_call",
-          call_id: "call_2",
-          name: "lookup",
-          arguments: '{"query":"news"}',
-        },
      ])
    }),
  )
@@ -295,7 +295,7 @@ describe("OpenRouter", () => {
              {
                type: "reasoning",
                text: "Thinking",
-                providerMetadata: { openrouter: { reasoningField: "reasoning", reasoningDetails: details } },
+                providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: details } },
              },
            ]),
          ],
@@ -328,7 +328,7 @@ describe("OpenRouter", () => {
            Message.assistant({
              type: "reasoning",
              text: "Thinking",
-              providerMetadata: { openrouter: { reasoningField: "reasoning", reasoningDetails: details } },
+              providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: details } },
            }),
          ],
        }),
@@ -354,7 +354,7 @@ describe("OpenRouter", () => {
            Message.assistant({
              type: "reasoning",
              text: "AB",
-              providerMetadata: { openrouter: { reasoningField: "reasoning", reasoningDetails: details } },
+              providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: details } },
            }),
          ],
        }),
@@ -80,9 +80,7 @@ describe("Z.ai Images", () => {

  it.effect("sanitizes unpaired surrogates in outbound image requests", () =>
    Image.generate({
-      model: ZAI.configure({ apiKey: "test", http: { body: { metadata: { source: "default\uDC00" } } } }).image(
-        "model",
-      ),
+      model: ZAI.configure({ apiKey: "test", http: { body: { metadata: { source: "default\uDC00" } } } }).image("model"),
      prompt: "A red circle \uD800 on a white background \u{1F600}",
    }).pipe(
      Effect.provide(
@@ -40,13 +40,4 @@ 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")
-  })
 })
@@ -1,5 +1,3 @@
 src/assets/theme.css
 e2e/test-results
 e2e/playwright-report
-component-tests/test-results
-component-tests/playwright-report
@@ -31,7 +31,7 @@ Your app is ready to be deployed!

 ## E2E Testing

-Locally, Playwright starts the Vite dev server automatically via `webServer`, or reuses one already running at the configured address. The browser suite uses isolated API fixtures rather than a live opencode backend.
+Playwright starts the Vite dev server automatically via `webServer`, and UI tests expect an opencode backend at `localhost:4096` by default.

 ```bash
 bunx playwright install chromium
@@ -39,35 +39,11 @@ bun run test:e2e:local
 bun run test:e2e:local -- --grep "settings"
 ```

-CI builds the app once and runs the same suite against Vite preview, serving production assets from `dist`. Managed built runs never reuse an existing server, so a running dev server cannot silently replace the production build. To run this mode locally:
-
-```bash
-bun run test:e2e:built
-bun run test:e2e:built -- --grep "settings"
-```
-
-To test an already-running dev server without starting or building a server:
-
-```bash
-PLAYWRIGHT_BASE_URL=http://127.0.0.1:4444 bun run test:e2e
-```
-
-For an already-running production build, also set `PLAYWRIGHT_BUILD=1` so the fixture API uses the app's origin:
-
-```bash
-PLAYWRIGHT_BUILD=1 PLAYWRIGHT_BASE_URL=http://127.0.0.1:4444 bun run test:e2e
-```
-
-External targets must use HTTP because fixture URLs use HTTP. `PLAYWRIGHT_BASE_URL` skips server startup and building in either mode.
-
-Compiled CLI startup and service lifecycle coverage runs separately in CI via `packages/cli/script/service-smoke.ts`.
-
 Environment options:

- `PLAYWRIGHT_BUILD=1` (build and preview locally; always enabled when `CI` is set)
- `PLAYWRIGHT_SERVER_HOST` / `PLAYWRIGHT_SERVER_PORT` (dev fixture API address, default: `127.0.0.1:4096`; built runs use the app's origin, matching production)
- `PLAYWRIGHT_PORT` (managed dev or preview server port, default: `3000`)
- `PLAYWRIGHT_BASE_URL` (use an externally managed app instead of starting a server; otherwise defaults to `http://127.0.0.1:<PLAYWRIGHT_PORT>`)
+- `PLAYWRIGHT_SERVER_HOST` / `PLAYWRIGHT_SERVER_PORT` (backend address, default: `localhost:4096`)
+- `PLAYWRIGHT_PORT` (Vite dev server port, default: `3000`)
+- `PLAYWRIGHT_BASE_URL` (override base URL, default: `http://localhost:<PLAYWRIGHT_PORT>`)

 ## Deployment

@@ -1,26 +0,0 @@
-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()
-})
@@ -44,7 +44,6 @@ The suite contains:
 - single-session tab close timing through stable home restoration
 - cached session repaint and mutation tracing
 - streaming timeline throughput, RAF-gap, long-task, geometry, and remount diagnostics
- retained renderer heap with a large model catalog across repeated session navigation

 All benchmarks import the shared `benchmark` fixture. Pages created through Playwright's `page` fixture automatically capture main-frame navigation history and emit a Chrome trace when `OPENCODE_PERFORMANCE_TRACE_DIR` is set. Benchmarks that need isolated browser contexts use `withBenchmarkPage`, which owns the context and the same diagnostics lifecycle.

@@ -79,26 +78,6 @@ Benchmarks do not assert machine-dependent performance budgets. Streaming proces

 Committed smoke and regression tests continue to own correctness coverage for pagination, tab paint, context resize, collapse state, and composer spacing.

-Tab-switch timing starts at `mousedown`, when mouse-selected tabs actually navigate, with a `click` fallback for keyboard activation. The probe excludes hidden/transparent content and intersects answers with their virtual-row clip and viewport. The tab workload requires the destination's final answer to be visible with Markdown ready. These results are not directly comparable to older click-start, geometry-only measurements. `stableObservedMs` includes confirmation across three correct samples; `firstCorrectObservedMs` is the first sample meeting all content and geometry checks. Neither is a compositor presentation timestamp.
-
-Each tab scenario reports one sample, including its raw observations. Use Playwright's `--repeat-each=5` for repeated measurements. Cached scenarios warm the destination at the same panel width before leaving it; a separate resized scenario validates reuse after opening the review pane changes that width.
-
-```sh
-bunx playwright test --config e2e/performance/playwright.config.ts \
-  timeline/session-tab-switch-benchmark.spec.ts --repeat-each=5
-```
-
-## Retained renderer memory
-
-Run the catalog workload against the production app bundle:
-
-```sh
-bunx playwright test --config e2e/performance/playwright.config.ts \
-  timeline/provider-memory-benchmark.spec.ts --repeat-each=3
-```
-
-`PROVIDER_MEMORY_MODELS` defaults to 1,200 and `PROVIDER_MEMORY_SWITCHES` defaults to 10. Each sample records Chromium's `Runtime.getHeapUsage` and `Memory.getDOMCounters` after an explicit garbage collection. This measures retained state, not allocation peaks or normal GC timing. It does not include worker heaps, the Electron main/GPU processes, or the OpenCode server, and must not be reported as total desktop RAM. Use identical model counts and navigation sequences for before/after comparisons.
-
 ## Chrome traces

 Set `OPENCODE_PERFORMANCE_TRACE_DIR` to emit a standard Chrome DevTools trace for every benchmark page automatically:
@@ -1,7 +1,7 @@
 import config from "../../playwright.config"

 const port = Number(process.env.PLAYWRIGHT_PORT ?? 3000)
-process.env.PLAYWRIGHT_SERVER_PORT = new URL(process.env.PLAYWRIGHT_BASE_URL ?? `http://127.0.0.1:${port}`).port || "80"
+process.env.PLAYWRIGHT_SERVER_PORT = String(port)
 process.env.OPENCODE_PERFORMANCE_RUN_ID ??= `${new Date().toISOString().replace(/[:.]/g, "-")}-${process.pid}`

 export default {
@@ -12,11 +12,9 @@ export default {
  fullyParallel: false,
  workers: 1,
  reporter: [["html", { outputFolder: "../playwright-report/performance", open: "never" }], ["line"]],
-  webServer: config.webServer
-    ? {
-        ...config.webServer,
-        command: `bun run build && bun run serve -- --host 0.0.0.0 --port ${port} --strictPort`,
-        reuseExistingServer: false,
-      }
-    : undefined,
+  webServer: {
+    ...config.webServer,
+    command: `bun run build && bun run serve -- --host 0.0.0.0 --port ${port} --strictPort`,
+    reuseExistingServer: false,
+  },
 }
@@ -126,19 +126,13 @@ test("keeps moving upward while drag-selecting above the timeline", async ({ pag
      )
    })
  })
-  const textBox = await text.evaluate((element) => {
-    const range = document.createRange()
-    range.selectNodeContents(element)
-    const rect = range.getClientRects()[0]
-    return rect ? { x: rect.x, y: rect.y, width: rect.width, height: rect.height } : null
-  })
+  const textBox = await text.boundingBox()
  const scrollBox = await scroller.boundingBox()
  expect(textBox).not.toBeNull()
  expect(scrollBox).not.toBeNull()
  if (!textBox || !scrollBox) return

-  // Start on a text line, not the empty right edge or gap between wrapped lines.
-  await page.mouse.move(textBox.x + Math.min(20, textBox.width / 2), textBox.y + textBox.height / 2)
+  await page.mouse.move(textBox.x + textBox.width - 10, textBox.y + textBox.height / 2)
  await page.mouse.down()
  await page.mouse.move(textBox.x + 20, scrollBox.y - 120, { steps: 30 })

@@ -201,45 +195,6 @@ test("does not pull a keyboard-scrolled user during shell remeasurement", async
  await reportVisualStability(testInfo, "keyboard-during-resize", trace, anchorPlan(regions))
 })

-test("keeps an older answer selected while scrolling within the interaction buffer", async ({ page }) => {
-  await setupTimeline(page, {
-    messages: history(80),
-    viewport: { width: 1400, height: 700 },
-    reducedMotion: true,
-  })
-  const scroller = page.locator(".scroll-view__viewport", { has: page.locator("[data-timeline-row]") })
-  const answer = page.getByText("History 78.", { exact: false })
-  await expect(answer).toBeVisible()
-  await expect
-    .poll(() =>
-      answer.evaluate((element) => element.closest('[data-component="markdown"]')?.hasAttribute("data-markdown-ready")),
-    )
-    .toBe(true)
-  const textBox = await answer.evaluate((element) => {
-    const range = document.createRange()
-    range.selectNodeContents(element)
-    const rect = range.getClientRects()[0]
-    return { x: rect.x, y: rect.y, width: rect.width, height: rect.height }
-  })
-  const scrollBox = await scroller.boundingBox()
-  expect(scrollBox).not.toBeNull()
-  if (!scrollBox) return
-  await page.mouse.move(textBox.x + Math.min(180, textBox.width - 2), textBox.y + textBox.height / 2)
-  await page.mouse.down()
-  await page.mouse.move(textBox.x + 2, textBox.y + textBox.height / 2, { steps: 30 })
-  await page.mouse.up()
-  await expect.poll(() => page.evaluate(() => window.getSelection()?.toString())).toContain("History 78.")
-  await page.mouse.move(scrollBox.x + scrollBox.width / 2, scrollBox.y + scrollBox.height / 2)
-  await page.mouse.wheel(0, -450)
-  await expect
-    .poll(() => scroller.evaluate((element) => element.scrollHeight - element.clientHeight - element.scrollTop))
-    .toBeGreaterThan(400)
-  await expect(answer).toHaveCount(1)
-  await expect.poll(() => page.evaluate(() => window.getSelection()?.toString())).toContain("History 78.")
-  await page.getByRole("heading", { name: "Timeline visual stability" }).click()
-  await expect.poll(() => page.evaluate(() => window.getSelection()?.isCollapsed)).toBe(true)
-})
-
 test("tracks keyboard scrolling from a focused timeline descendant", async ({ page }, testInfo) => {
  const shellID = "prt_descendant_keyboard_01_shell"
  const timeline = await setupTimeline(page, {
@@ -304,16 +259,12 @@ test("does not claim keyboard scrolling owned by a nested scrollable", async ({
  const before = await scroller.evaluate((element) => element.scrollTop)
  const nestedBefore = await nested.evaluate((element) => element.scrollTop)
  await nested.press("PageUp")
-  await expect.poll(() => nested.evaluate((element) => element.scrollTop)).toBeLessThan(nestedBefore)
+  await page.waitForTimeout(300)
  expect(await scroller.evaluate((element) => element.scrollTop)).toBe(before)
+  expect(await nested.evaluate((element) => element.scrollTop)).toBeLessThan(nestedBefore)

-  await nested.evaluate((element) => element.scrollTo({ top: 0, behavior: "instant" }))
-  await expect.poll(() => nested.evaluate((element) => element.scrollTop)).toBe(0)
-  await scroller.evaluate((element) => {
-    element.dispatchEvent(new WheelEvent("wheel", { bubbles: true, cancelable: true, deltaY: -1 }))
-    element.scrollTo({ top: Math.min(300, element.scrollHeight - element.clientHeight), behavior: "instant" })
-  })
-  await expect.poll(() => scroller.evaluate((element) => element.scrollTop)).toBeLessThan(500)
+  await nested.evaluate((element) => (element.scrollTop = 0))
+  await scroller.evaluate((element) => (element.scrollTop = Math.min(300, element.scrollHeight - element.clientHeight)))
  const boundaryBefore = await scroller.evaluate((element) => element.scrollTop)
  expect(boundaryBefore).toBeGreaterThan(0)
  await nested.press("PageUp")
@@ -1,98 +0,0 @@
-import { benchmark, expect } from "../benchmark"
-import { mockOpenCodeServer } from "../../utils/mock-server"
-import { expectSessionTitle } from "../../utils/waits"
-import { fixture, pageMessages } from "./session-timeline-stress.fixture"
-import { installStressSessionTabs, installTimelineSettings, stressSessionHref } from "./timeline-test-helpers"
-import { waitForStableTimeline } from "./session-tab-switch-probe"
-import type { CatalogUpdated } from "@opencode-ai/client/promise"
-
-benchmark("measures retained renderer memory with a large model catalog", async ({ page, report }) => {
-  benchmark.setTimeout(120_000)
-  const count = Number(process.env.PROVIDER_MEMORY_MODELS ?? 1200)
-  const switches = Number(process.env.PROVIDER_MEMORY_SWITCHES ?? 10)
-  const provider = fixture.provider.all[0]
-  const selected = { ...provider.models["claude-opus-4-6"] }
-  await mockOpenCodeServer(page, {
-    directory: fixture.directory,
-    project: fixture.project,
-    sessions: fixture.sessions,
-    pageMessages,
-    provider: {
-      ...fixture.provider,
-      all: [
-        {
-          ...provider,
-          models: {
-            [selected.id]: selected,
-            ...Object.fromEntries(
-              Array.from({ length: count - 1 }, (_, index) => {
-                const id = `catalog-model-${index}`
-                return [
-                  id,
-                  {
-                    id,
-                    name: `Catalog model ${index}`,
-                    cost: { input: 1, output: 2 },
-                    limit: { context: 200_000, output: 8192 },
-                    variants: { high: { reasoningEffort: "high" } },
-                  },
-                ]
-              }),
-            ),
-          },
-        },
-      ],
-    },
-  })
-  await installTimelineSettings(page)
-  await installStressSessionTabs(page)
-  await page.goto(stressSessionHref(fixture.sourceID))
-  await expectSessionTitle(page, fixture.expected.sourceTitle)
-  await waitForStableTimeline(page, fixture.expected.sourceMessageIDs.at(-1)!)
-  await expect(page.locator('[data-action="composer-model"]')).toContainText("Claude Opus 4.6")
-  const cdp = await page.context().newCDPSession(page)
-  const samples = []
-  for (let index = 0; index <= switches; index++) {
-    if (index > 0) {
-      const target = index % 2 === 1
-      const id = target ? fixture.targetID : fixture.sourceID
-      await page.locator(`[data-slot="titlebar-tabs"] a[href="${stressSessionHref(id)}"]`).click()
-      await expectSessionTitle(page, target ? fixture.expected.targetTitle : fixture.expected.sourceTitle)
-      await waitForStableTimeline(
-        page,
-        (target ? fixture.expected.targetMessageIDs : fixture.expected.sourceMessageIDs).at(-1)!,
-      )
-      await expect(page.locator('[data-action="composer-model"]')).toContainText("Claude Opus 4.6")
-    }
-    // GC is an explicit retained-heap measurement, not an application optimization or readiness wait.
-    await cdp.send("HeapProfiler.collectGarbage")
-    samples.push({
-      switches: index,
-      heap: await cdp.send("Runtime.getHeapUsage"),
-      dom: await cdp.send("Memory.getDOMCounters"),
-    })
-  }
-  expect(samples).toHaveLength(switches + 1)
-  expect(samples.every((sample) => sample.heap.usedSize > 0)).toBe(true)
-  selected.name = "Updated catalog model"
-  await page.evaluate(
-    (event) => {
-      const host = window as Window & { __mockServerStream?: { push: (events: CatalogUpdated[]) => void } }
-      if (!host.__mockServerStream) throw new Error("Missing fixture event stream")
-      host.__mockServerStream.push([event])
-    },
-    {
-      id: "evt_catalog_refresh",
-      created: Date.now(),
-      type: "catalog.updated",
-      location: { directory: fixture.directory },
-      data: {},
-    } satisfies CatalogUpdated,
-  )
-  await expect(page.locator('[data-action="composer-model"]')).toContainText(selected.name)
-  report(
-    { samples },
-    { models: count, switches, gc: "explicit", scope: "renderer main isolate; not total desktop RAM" },
-  )
-  await cdp.detach()
-})
@@ -11,73 +11,115 @@ import {
 } from "./timeline-test-helpers"
 import { measureSessionSwitch, waitForStableTimeline } from "./session-tab-switch-probe"

-const scenarios = [
-  { cached: false, review: false, resized: false },
-  { cached: false, review: true, resized: false },
-  { cached: true, review: false, resized: false },
-  { cached: true, review: true, resized: false },
-  { cached: true, review: true, resized: true },
-]
+type Result = Awaited<ReturnType<typeof measureSessionSwitch>>

-scenarios.forEach((scenario) => {
-  const name = `tab switch: ${scenario.cached ? "cached" : "unmounted"}, review ${scenario.review ? "open" : "closed"}${scenario.resized ? ", resized" : ""}`
-  benchmark(name, async ({ browser, report }, testInfo) => {
-    const result = await withBenchmarkPage(
-      browser,
-      name,
-      async (page) => {
-        await mockStressTimeline(page, { vcsDiff: createReviewDiffs() })
-        await installTimelineSettings(page)
-        await installStressSessionTabs(page)
-        await page.goto(stressSessionHref(fixture.sourceID))
-        await expectSessionTitle(page, fixture.expected.sourceTitle)
-        await waitForStableTimeline(page, fixture.expected.sourceMessageIDs.at(-1)!)
-        if (scenario.review && !scenario.resized) await openReviewPane(page)
-        if (scenario.cached) {
-          await switchSession(page, fixture.targetID, fixture.expected.targetTitle)
-          const answer = page.locator(`[data-timeline-part-id="${fixture.expected.targetPartIDs.at(-1)}"]`)
-          await expect(answer.locator('[data-component="markdown"]')).toHaveAttribute("data-markdown-ready", "")
-          await expect
-            .poll(() =>
-              answer.evaluate((element) => element.checkVisibility({ checkOpacity: true, checkVisibilityCSS: true })),
-            )
-            .toBe(true)
-          await waitForStableTimeline(page, fixture.expected.targetMessageIDs.at(-1)!)
-          await switchSession(page, fixture.sourceID, fixture.expected.sourceTitle)
+benchmark(
+  "benchmarks session tab switching with and without the review pane",
+  async ({ browser, report }, testInfo) => {
+    benchmark.setTimeout(360_000)
+    const runs = Number(process.env.SESSION_TAB_SWITCH_RUNS ?? 5)
+    const results = {
+      closed: { cold: [] as Result[], hot: [] as Result[] },
+      open: { cold: [] as Result[], hot: [] as Result[] },
+    }
+    for (const reviewPane of ["closed", "open"] as const) {
+      for (const mode of ["cold", "hot"] as const) {
+        for (let run = 0; run < runs; run++) {
+          results[reviewPane][mode].push(
+            await withBenchmarkPage(
+              browser,
+              `session-tab-switch-${reviewPane}-${mode}-${run}`,
+              (page) => trial(page, mode, reviewPane),
+              testInfo,
+            ),
+          )
        }
-        if (scenario.resized) await openReviewPane(page)
-        await waitForStableTimeline(page, fixture.expected.sourceMessageIDs.at(-1)!)
+      }
+    }
+    report({ results, summary: summarizeReviewPane(results) }, { runs, reviewDiffs: createReviewDiffs().length })
+  },
+)

-        return measureSessionSwitch(page, {
-          destinationIDs: fixture.messages[fixture.targetID].map((message) => message.id),
-          sourceIDs: fixture.messages[fixture.sourceID].map((message) => message.id),
-          lastID: fixture.expected.targetMessageIDs.at(-1)!,
-          requiredPartID: fixture.expected.targetPartIDs.at(-1),
-          href: stressSessionHref(fixture.targetID),
-          switch: () => switchSession(page, fixture.targetID, fixture.expected.targetTitle),
-        })
-      },
-      testInfo,
-    )
-    expect(result.unknownSamples).toBe(0)
-    expect(result.wrongDestinationSamples).toBe(0)
-    if (scenario.cached) expect(result.blankSamples).toBe(0)
-    report(result, { ...scenario, inputEvent: "mousedown", requireReadyAnswer: true })
+async function trial(page: Page, mode: "cold" | "hot", reviewPane: "closed" | "open") {
+  const reviewDiffs = createReviewDiffs()
+  await mockStressTimeline(page, { vcsDiff: reviewDiffs })
+  await installTimelineSettings(page)
+  await installStressSessionTabs(page)
+  if (mode === "hot") {
+    await page.goto(stressSessionHref(fixture.targetID))
+    await expectSessionTitle(page, fixture.expected.targetTitle)
+    await waitForStableTimeline(page, fixture.expected.targetMessageIDs.at(-1)!)
+    await switchSession(page, fixture.sourceID, fixture.expected.sourceTitle)
+  } else {
+    await page.goto(stressSessionHref(fixture.sourceID))
+    await expectSessionTitle(page, fixture.expected.sourceTitle)
+  }
+  await waitForStableTimeline(page, fixture.expected.sourceMessageIDs.at(-1)!)
+  if (reviewPane === "open") {
+    await openReviewPane(page)
+    await waitForStableTimeline(page, fixture.expected.sourceMessageIDs.at(-1)!)
+  }
+
+  const destinationIDs = fixture.messages[fixture.targetID].map((message) => message.id)
+  const sourceIDs = fixture.messages[fixture.sourceID].map((message) => message.id)
+  const lastID = fixture.expected.targetMessageIDs.at(-1)!
+  const href = stressSessionHref(fixture.targetID)
+  const result = await measureSessionSwitch(page, {
+    destinationIDs,
+    sourceIDs,
+    lastID,
+    href,
+    switch: () => switchSession(page, fixture.targetID, fixture.expected.targetTitle),
  })
-})
+  return result
+}
+
+function summarize(results: Record<"cold" | "hot", Result[]>) {
+  const stats = (values: (number | null)[]) => {
+    const sorted = values.filter((value): value is number => value !== null).sort((a, b) => a - b)
+    return {
+      min: sorted[0] ?? null,
+      median: sorted[Math.floor(sorted.length / 2)] ?? null,
+      max: sorted.at(-1) ?? null,
+      missing: values.length - sorted.length,
+    }
+  }
+  return Object.fromEntries(
+    Object.entries(results).map(([mode, values]) => [
+      mode,
+      {
+        firstDestinationObservedMs: stats(values.map((value) => value.firstDestinationObservedMs)),
+        firstCorrectObservedMs: stats(values.map((value) => value.firstCorrectObservedMs)),
+        stableObservedMs: stats(values.map((value) => value.stableObservedMs)),
+      },
+    ]),
+  )
+}
+
+function summarizeReviewPane(results: Record<"closed" | "open", Record<"cold" | "hot", Result[]>>) {
+  return Object.fromEntries(
+    Object.entries(results).map(([reviewPane, values]) => [
+      reviewPane,
+      summarize(values as Record<"cold" | "hot", Result[]>),
+    ]),
+  )
+}

 async function switchSession(page: Page, sessionID: string, title: string) {
-  const tab = page.locator(`[data-slot="titlebar-tabs"] a[href="${stressSessionHref(sessionID)}"]`)
-  await expect(tab).toHaveCount(1)
+  const href = stressSessionHref(sessionID)
+  const tab = page.locator(`[data-slot="titlebar-tabs"] a[href="${href}"]`).first()
+  await expect(tab).toBeVisible()
  await tab.click()
  await expectSessionTitle(page, title)
 }

 async function openReviewPane(page: Page) {
  await page.getByRole("button", { name: "Toggle review" }).click()
-  await expect(page.locator("#review-panel")).toBeVisible()
+  const panel = page.locator("#review-panel")
+  await expect(panel).toBeVisible()
  await page.waitForFunction(() => {
-    const text = document.querySelector("#review-panel")?.textContent ?? ""
+    const panel = document.querySelector<HTMLElement>("#review-panel")
+    const text = panel?.textContent ?? ""
    return text.includes("generated-000.ts") && text.includes("+3")
  })
 }
@@ -20,10 +20,9 @@ export function classifySessionSwitch(samples: SessionSwitchSample[]) {
  const firstCorrect = samples.findIndex(isCorrectDestination)
  const stable = samples.findIndex((_, index) => isStableSessionSwitch(samples.slice(index, index + 3)))
  return {
-    samples,
    firstDestinationObservedMs: samples[firstDestination]?.observedAtMs ?? null,
    firstCorrectObservedMs: samples[firstCorrect]?.observedAtMs ?? null,
-    stableObservedMs: stable < 0 ? null : samples[stable + 2].observedAtMs,
+    stableObservedMs: samples[stable + 2]?.observedAtMs ?? null,
    wrongDestinationSamples: samples
      .slice(firstDestination)
      .filter((sample) => sample.destination.length > 0 && !sample.last).length,
@@ -1,67 +0,0 @@
-import { benchmark, expect } from "../benchmark"
-import { measureSessionSwitch } from "./session-tab-switch-probe"
-import type { SessionSwitchSample } from "./session-tab-switch-metrics"
-
-benchmark("starts at mousedown and excludes hidden or unfinished destination content", async ({ page, report }) => {
-  await page.setContent(`
-    <a href="/session/destination">Destination</a>
-    <div class="scroll-view__viewport" style="height:200px;overflow:auto">
-      <div data-timeline-row="message" data-timeline-key="row" data-message-id="source">
-        <div data-timeline-part-id="answer"><div data-component="markdown">Destination answer</div></div>
-      </div>
-    </div>
-  `)
-  await page.evaluate(() => {
-    document.querySelector("a")!.addEventListener("mousedown", () => {
-      const row = document.querySelector<HTMLElement>("[data-message-id]")!
-      row.dataset.messageId = "destination"
-      row.style.visibility = "hidden"
-    })
-  })
-  const result = await measureSessionSwitch(page, {
-    destinationIDs: ["destination"],
-    sourceIDs: ["source"],
-    lastID: "destination",
-    requiredPartID: "answer",
-    requireBottomAnchor: false,
-    href: "/session/destination",
-    switch: async () => {
-      // No click is dispatched: the probe must observe the event that activates tabs.
-      await page.getByRole("link", { name: "Destination" }).dispatchEvent("mousedown", { button: 0 })
-      await page.waitForFunction(() => {
-        const host = window as Window & { __sessionSwitchProbe?: { samples: SessionSwitchSample[] } }
-        return host.__sessionSwitchProbe?.samples.some((sample) => !sample.hasVisibleRows)
-      })
-      await page.locator("[data-message-id]").evaluate((row) => row.style.removeProperty("visibility"))
-      await page.waitForFunction(() => {
-        const host = window as Window & { __sessionSwitchProbe?: { samples: SessionSwitchSample[] } }
-        return host.__sessionSwitchProbe?.samples.some(
-          (sample) => sample.destination.length > 0 && sample.requiredPartVisible === false,
-        )
-      })
-      const beforeClip = await page.evaluate(() => {
-        const row = document.querySelector<HTMLElement>("[data-timeline-key]")!
-        row.style.cssText = "height:10px;position:relative;overflow:clip"
-        const answer = row.querySelector<HTMLElement>("[data-timeline-part-id]")!
-        answer.style.cssText = "position:absolute;top:30px;width:150px"
-        answer.querySelector('[data-component="markdown"]')!.setAttribute("data-markdown-ready", "")
-        return (
-          (window as Window & { __sessionSwitchProbe?: { samples: SessionSwitchSample[] } }).__sessionSwitchProbe
-            ?.samples.length ?? 0
-        )
-      })
-      await page.waitForFunction((count) => {
-        const host = window as Window & { __sessionSwitchProbe?: { samples: SessionSwitchSample[] } }
-        return host.__sessionSwitchProbe?.samples.slice(count).some((sample) => sample.requiredPartVisible === false)
-      }, beforeClip)
-      await page.locator("[data-timeline-key]").evaluate((row) => {
-        row.style.height = "100px"
-      })
-    },
-  })
-  expect(result.blankSamples).toBeGreaterThan(0)
-  expect(result.firstCorrectObservedMs).not.toBeNull()
-  expect(result.stableObservedMs).not.toBeNull()
-  expect(result.firstCorrectObservedMs).toBeGreaterThan(result.firstDestinationObservedMs!)
-  report(result)
-})
@@ -25,7 +25,7 @@ async function installSessionSwitchProbe(
    let running = true
    const reviewLevels: Record<string, string> = {
      panel: "#review-panel",
-      tabs: '#review-panel [data-component="tabs"]',
+    tabs: '#review-panel [data-component="tabs"]',
      body: '#review-panel [data-slot="session-review-v2-body"]',
      review: '#review-panel [data-component="session-review-v2"]',
      preview: '#review-panel [data-slot="session-review-v2-preview"]',
@@ -37,6 +37,7 @@ async function installSessionSwitchProbe(
      if (!running || started === undefined) return
      setTimeout(() => {
        if (!running || started === undefined) return
+        const observedAtMs = performance.now() - started
        const reviewPanel = document.querySelector<HTMLElement>("#review-panel")
        const reviewFile = reviewPanel?.querySelector('[data-component="file"][data-mode="diff"]')
        const initialReviewFile = initialReviewNodes.file
@@ -62,30 +63,26 @@ async function installSessionSwitchProbe(
        )
        if (root) {
          const view = root.getBoundingClientRect()
-          const inViewport = (element: HTMLElement) => {
-            if (!element.checkVisibility({ checkOpacity: true, checkVisibilityCSS: true })) return false
-            const rect = element.getBoundingClientRect()
-            const clip = element.closest<HTMLElement>("[data-timeline-key]")?.getBoundingClientRect() ?? view
-            return (
-              Math.min(rect.bottom, clip.bottom, view.bottom) > Math.max(rect.top, clip.top, view.top) &&
-              Math.min(rect.right, clip.right, view.right) > Math.max(rect.left, clip.left, view.left)
-            )
-          }
          const visible = [...root.querySelectorAll<HTMLElement>("[data-message-id]")]
-            .filter(inViewport)
+            .filter((element) => {
+              const rect = element.getBoundingClientRect()
+              return rect.bottom > view.top && rect.top < view.bottom
+            })
            .map((element) => element.dataset.messageId!)
-          const hasVisibleRows = [...root.querySelectorAll<HTMLElement>("[data-timeline-key]")].some(inViewport)
+          const hasVisibleRows = [...root.querySelectorAll<HTMLElement>("[data-timeline-key]")].some((element) => {
+            const rect = element.getBoundingClientRect()
+            return rect.bottom > view.top && rect.top < view.bottom
+          })
          const requiredPartVisible = requiredPartID
            ? [...root.querySelectorAll<HTMLElement>("[data-timeline-part-id]")].some((element) => {
                if (element.dataset.timelinePartId !== requiredPartID) return false
-                if (!element.textContent?.trim()) return false
-                if (element.querySelector('[data-component="markdown"]:not([data-markdown-ready])')) return false
-                return inViewport(element)
+                const rect = element.getBoundingClientRect()
+                return rect.width > 0 && rect.height > 0 && rect.bottom > view.top && rect.top < view.bottom
              })
            : undefined
          const spacer = root.querySelector<HTMLElement>('[data-timeline-row="bottom-spacer"]')?.getBoundingClientRect()
          samples.push({
-            observedAtMs: performance.now() - started,
+            observedAtMs,
            destination: visible.filter((id) => destination.has(id)),
            source: visible.filter((id) => source.has(id)),
            hasVisibleRows,
@@ -97,7 +94,7 @@ async function installSessionSwitchProbe(
          })
        } else {
          samples.push({
-            observedAtMs: performance.now() - started,
+            observedAtMs,
            destination: [],
            source: [],
            hasVisibleRows: false,
@@ -110,25 +107,23 @@ async function installSessionSwitchProbe(
        requestAnimationFrame(sample)
      }, 0)
    }
-    const start = (event: MouseEvent) => {
-      if (started !== undefined || event.button !== 0) return
-      const link = event.target instanceof Element ? event.target.closest("a") : undefined
-      if (link?.getAttribute("href") !== href) return
-      started = performance.now()
-      for (const [name, selector] of Object.entries(reviewLevels)) {
-        initialReviewNodes[name] = document.querySelector(selector)
-      }
-      requestAnimationFrame(sample)
-    }
-    // Tabs activate on mousedown; click alone misses the synchronous navigation work.
-    document.addEventListener("mousedown", start, true)
-    document.addEventListener("click", start, true)
+    document.addEventListener(
+      "click",
+      (event) => {
+        const link = event.target instanceof Element ? event.target.closest("a") : undefined
+        if (link?.getAttribute("href") !== href) return
+        started = performance.now()
+        for (const [name, selector] of Object.entries(reviewLevels)) {
+          initialReviewNodes[name] = document.querySelector(selector)
+        }
+        requestAnimationFrame(sample)
+      },
+      { capture: true, once: true },
+    )
    ;(window as Window & { __sessionSwitchProbe?: SessionSwitchProbe }).__sessionSwitchProbe = {
      samples,
      stop: () => {
        running = false
-        document.removeEventListener("mousedown", start, true)
-        document.removeEventListener("click", start, true)
      },
    }
  }, input)
@@ -53,15 +53,6 @@ test("reports missing correctness without throwing", () => {
  expect(result.stableObservedMs).toBeNull()
 })

-test("does not report stability for only two correct samples", () => {
-  const result = classifySessionSwitch([
-    { observedAtMs: 16, destination: ["destination"], source: [], hasVisibleRows: true, last: true, bottomErrorPx: 0 },
-    { observedAtMs: 32, destination: ["destination"], source: [], hasVisibleRows: true, last: true, bottomErrorPx: 0 },
-  ])
-  expect(result.firstCorrectObservedMs).toBe(16)
-  expect(result.stableObservedMs).toBeNull()
-})
-
 test("requires an explicitly tracked part to be visible", () => {
  const result = classifySessionSwitch([
    {
@@ -3,7 +3,7 @@ import { base64Encode } from "@opencode-ai/util/encode"
 import { currentSession } from "../utils/mock-server"
 import { installSseTransport } from "../utils/sse-transport"

-const serverA = `http://${process.env.PLAYWRIGHT_SERVER_HOST ?? "127.0.0.1"}:${process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"}`
+const serverA = "http://127.0.0.1:4096"
 const serverB = "http://127.0.0.1:4097"
 const sessionA = session("ses_server_a", "C:/server-a", "Server A session")
 const sessionB = session("ses_server_b", "/home/server-b", "Server B session")
@@ -12,17 +12,17 @@ test("closing the active server's last tab opens the remaining server tab", asyn
  const requests: string[] = []
  await mockServers(page, requests)
  await page.addInitScript(
-    ({ serverA, serverB, sessionA, sessionB }) => {
+    ({ serverB, sessionA, sessionB }) => {
      localStorage.setItem("opencode.global.dat:server", JSON.stringify({ list: [serverB] }))
      localStorage.setItem(
        "opencode.window.browser.dat:tabs",
        JSON.stringify([
-          { type: "session", server: serverA, sessionId: sessionA },
+          { type: "session", server: "http://127.0.0.1:4096", sessionId: sessionA },
          { type: "session", server: serverB, sessionId: sessionB },
        ]),
      )
    },
-    { serverA, serverB, sessionA: sessionA.id, sessionB: sessionB.id },
+    { serverB, sessionA: sessionA.id, sessionB: sessionB.id },
  )

  const hrefA = `/server/${base64Encode(serverA)}/session/${sessionA.id}`
@@ -55,7 +55,7 @@ function session(id: string, directory: string, title: string) {
 async function mockServers(page: Page, requests: string[]) {
  await installSseTransport(page, { server: serverA })
  await installSseTransport(page, { server: serverB })
-  await page.route("**/api/**", async (route) => {
+  await page.route("**/*", async (route) => {
    const url = new URL(route.request().url())
    if (url.origin !== serverA && url.origin !== serverB) return route.fallback()
    requests.push(url.toString())
@@ -1,103 +0,0 @@
-import { expect, test } from "@playwright/test"
-import { base64Encode } from "@opencode-ai/util/encode"
-import { mockOpenCodeServer } from "../utils/mock-server"
-
-const directory = "C:/Projects/extensions-demo"
-const server = `http://${process.env.PLAYWRIGHT_SERVER_HOST ?? "127.0.0.1"}:${process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"}`
-const session = {
-  id: "ses_project_extensions",
-  title: "Existing session",
-  directory,
-  projectID: "proj_extensions_demo",
-  time: { created: 1700000000000, updated: 1700000000000 },
-}
-
-test.use({ viewport: { width: 1440, height: 1000 }, colorScheme: "dark" })
-
-test("project Extensions stays inside settings while plugins load", async ({ page }) => {
-  await mockOpenCodeServer(page, {
-    directory,
-    project: {
-      id: session.projectID,
-      canonical: directory,
-      name: "Extensions demo",
-      vcs: "git",
-      time: session.time,
-      sandboxes: [],
-    },
-    provider: { all: [], connected: [], default: {} },
-    sessions: [session],
-    pageMessages: () => ({ items: [] }),
-  })
-  await page.addInitScript(
-    ({ server, sessionID, directory }) => {
-      localStorage.setItem(
-        "opencode.global.dat:server",
-        JSON.stringify({ projects: { local: [{ worktree: directory, expanded: true }] } }),
-      )
-      localStorage.setItem(
-        "opencode.window.browser.dat:tabs",
-        JSON.stringify([{ type: "session", server, sessionId: sessionID }]),
-      )
-    },
-    { server, sessionID: session.id, directory },
-  )
-  const href = `/server/${base64Encode(server)}/session/${session.id}`
-  await page.goto(href)
-  await expect(page.getByRole("heading", { name: session.title, exact: true })).toBeVisible()
-  await page.keyboard.press("Control+,")
-  const settings = page.getByTestId("settings-screen")
-  await settings.getByRole("tab", { name: "Projects", exact: true }).click()
-  await settings.getByText("Extensions demo", { exact: true }).click()
-  const dialog = page.getByRole("dialog")
-  await expect(dialog.getByRole("textbox", { name: "Name", exact: true })).toBeFocused()
-
-  const globalPlugins = Promise.withResolvers<void>()
-  const projectPlugins = Promise.withResolvers<void>()
-  await page.route(
-    (url) => url.pathname === "/api/plugin",
-    async (route) => {
-      const project = new URL(route.request().url()).searchParams.get("location[directory]")
-      await (project ? projectPlugins : globalPlugins).promise
-      await route.fulfill({
-        json: {
-          location: project ? { directory: project } : {},
-          data: (project ? ["shared-plugin", "project-plugin"] : ["shared-plugin"]).map((id) => ({
-            id,
-            source: { type: "package", package: id },
-            status: "active",
-            tui: false,
-          })),
-        },
-      })
-    },
-  )
-  const requested = page.waitForRequest((request) => {
-    const url = new URL(request.url())
-    return url.pathname === "/api/plugin" && url.searchParams.get("location[directory]") === directory
-  })
-  await dialog.getByRole("tab", { name: "Extensions", exact: true }).click()
-  await requested
-  await expect(page).toHaveURL(href)
-  await expect(dialog.getByRole("heading", { name: "Extensions", exact: true })).toBeVisible()
-  await expect(settings).toBeVisible()
-  await expect(page.getByRole("heading", { name: session.title, exact: true, includeHidden: true })).toBeHidden()
-  await dialog.getByRole("tab", { name: "Plugins", exact: true }).click()
-  await expect(dialog.getByRole("tab", { name: "Plugins", exact: true })).toHaveAttribute("aria-selected", "true")
-
-  globalPlugins.resolve()
-  await dialog.getByRole("tab", { name: "Scripts", exact: true }).click()
-  await expect(dialog.getByRole("heading", { name: "Scripts", exact: true })).toBeVisible()
-  await dialog.getByRole("tab", { name: "Extensions", exact: true }).click()
-  projectPlugins.resolve()
-  await dialog.getByRole("tab", { name: "Plugins", exact: true }).click()
-  await expect(dialog.getByText("project-plugin", { exact: true })).toBeVisible()
-  await dialog.getByRole("button", { name: "Shared with all projects 1", exact: true }).click()
-  await expect(dialog.getByText("shared-plugin", { exact: true })).toBeVisible()
-  await expect(page).toHaveURL(href)
-
-  await page.keyboard.press("Escape")
-  await expect(dialog).toBeHidden()
-  await expect(settings.getByRole("tab", { name: "Projects", exact: true })).toHaveAttribute("aria-selected", "true")
-  await expect(page.getByRole("heading", { name: session.title, exact: true, includeHidden: true })).toBeHidden()
-})
@@ -3,7 +3,7 @@ import { expect, test, type Page, type Route } from "@playwright/test"
 import { installSseTransport } from "../utils/sse-transport"
 import { currentSession } from "../utils/mock-server"

-const serverA = `http://${process.env.PLAYWRIGHT_SERVER_HOST ?? "127.0.0.1"}:${process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"}`
+const serverA = `http://127.0.0.1:${process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"}`
 const serverB = "http://127.0.0.1:4097"
 const directoryA = "C:/server-a"
 const directoryB = "/home/server-b"
@@ -24,17 +24,11 @@ test("session settings use the remote server context", async ({ page }) => {
  await configureServers(page)

  await page.goto(`/server/${base64Encode(serverB)}/session/${sessionB.id}`)
-  const sessionHeading = page.getByRole("heading", { name: sessionB.title, exact: true, includeHidden: true })
-  await expect(sessionHeading).toBeVisible()
+  await expect(page.getByRole("heading", { name: sessionB.title, exact: true })).toBeVisible()
  await page.keyboard.press("Control+,")

-  const settings = page.getByTestId("settings-screen")
-  await expect(settings).toBeVisible()
-  await expect(page.getByRole("dialog")).toHaveCount(0)
-  await expect(settings.getByRole("tablist")).toHaveCSS("width", "328px")
-  await expect(sessionHeading).toBeAttached()
-  await expect(sessionHeading).toBeHidden()
-  const autoAccept = settings.locator('[data-action="settings-auto-accept-permissions"]')
+  const dialog = page.locator(".settings-dialog")
+  const autoAccept = dialog.locator('[data-action="settings-auto-accept-permissions"]')
  const input = autoAccept.getByRole("switch")
  await expect(autoAccept).toBeVisible()
  await expect(input).toBeEnabled()
@@ -61,12 +55,9 @@ test("session settings use the remote server context", async ({ page }) => {
      },
    ])

-  await settings.getByRole("tab", { name: "Models" }).click()
-  await expect(settings.getByRole("switch", { name: "Server B Model" })).toBeEnabled()
-  await expect(settings.getByRole("switch", { name: "Server A Model" })).toHaveCount(0)
-  await settings.getByRole("button", { name: "Back to app" }).click()
-  await expect(settings).toBeHidden()
-  await expect(sessionHeading).toBeVisible()
+  await dialog.getByRole("tab", { name: "Models" }).click()
+  await expect(dialog.getByRole("switch", { name: "Server B Model" })).toBeEnabled()
+  await expect(dialog.getByRole("switch", { name: "Server A Model" })).toHaveCount(0)
 })

 test("auto-accept responds for an unfocused server session", async ({ page }) => {
@@ -87,7 +78,7 @@ test("auto-accept responds for an unfocused server session", async ({ page }) =>
  await page.goto(`/server/${base64Encode(serverA)}/session/${sessionA.id}`)
  await expect(page.getByRole("heading", { name: sessionA.title, exact: true })).toBeVisible()
  await page.keyboard.press("Control+,")
-  const autoAccept = page.getByTestId("settings-screen").locator('[data-action="settings-auto-accept-permissions"]')
+  const autoAccept = page.locator(".settings-dialog").locator('[data-action="settings-auto-accept-permissions"]')
  await autoAccept.locator('[data-slot="switch-control"]').click()
  await expect(autoAccept.getByRole("switch")).toBeChecked()
  await expect
@@ -187,7 +178,7 @@ test("auto-accept sweeps again after a reconnect", async ({ page }) => {
  const first = await transport.waitForConnection()

  await page.keyboard.press("Control+,")
-  const autoAccept = page.getByTestId("settings-screen").locator('[data-action="settings-auto-accept-permissions"]')
+  const autoAccept = page.locator(".settings-dialog").locator('[data-action="settings-auto-accept-permissions"]')
  await autoAccept.locator('[data-slot="switch-control"]').click()
  await expect(autoAccept.getByRole("switch")).toBeChecked()
  await expect
@@ -243,7 +234,7 @@ test("auto-accept approves a request discovered by opening a session", async ({
  await expect(page.getByRole("heading", { name: sessionA.title, exact: true })).toBeVisible()

  await page.keyboard.press("Control+,")
-  const autoAccept = page.getByTestId("settings-screen").locator('[data-action="settings-auto-accept-permissions"]')
+  const autoAccept = page.locator(".settings-dialog").locator('[data-action="settings-auto-accept-permissions"]')
  await autoAccept.locator('[data-slot="switch-control"]').click()
  await expect(autoAccept.getByRole("switch")).toBeChecked()

@@ -308,7 +299,7 @@ async function mockServers(
  permissionResponses: PermissionResponse[] = [],
  options: MockServerOptions = {},
 ) {
-  await page.route("**/api/**", async (route) => {
+  await page.route("**/*", async (route) => {
    const url = new URL(route.request().url())
    if (url.origin !== serverA && url.origin !== serverB) return route.fallback()
    const remote = url.origin === serverB
@@ -2,7 +2,7 @@ import { expect, test, type Page, type Route } from "@playwright/test"
 import { base64Encode } from "@opencode-ai/util/encode"
 import { currentSession } from "../utils/mock-server"

-const serverA = `http://${process.env.PLAYWRIGHT_SERVER_HOST ?? "127.0.0.1"}:${process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"}`
+const serverA = "http://127.0.0.1:4096"
 const serverB = "http://127.0.0.1:4097"
 const sessionA = session("ses_server_a", "C:/server-a", "Server A session")
 const sessionB = session("ses_server_b", "/home/server-b", "Server B session")
@@ -51,7 +51,7 @@ function session(id: string, directory: string, title: string) {
 }

 async function mockServers(page: Page) {
-  await page.route("**/api/**", async (route) => {
+  await page.route("**/*", async (route) => {
    const url = new URL(route.request().url())
    if (url.origin !== serverA && url.origin !== serverB) return route.fallback()
    const current = url.origin === serverA ? sessionA : sessionB
@@ -12,6 +12,54 @@ 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")
@@ -2,7 +2,7 @@ import { expect, test, type Route } from "@playwright/test"

 const server = "http://127.0.0.1:4097"

-test("server dialog keeps focus above fullscreen settings", async ({ page }) => {
+test("nested server dialog keeps focus inside the top layer", async ({ page }) => {
  await page.addInitScript((server) => {
    localStorage.setItem("opencode.global.dat:server", JSON.stringify({ list: [server] }))
  }, server)
@@ -24,9 +24,8 @@ test("server dialog keeps focus above fullscreen settings", async ({ page }) =>

  await page.goto("/")
  await page.keyboard.press("Control+,")
-  const settings = page.getByTestId("settings-screen")
+  const settings = page.locator(".settings-dialog")
  await expect(settings).toBeVisible()
-  await expect(page.getByRole("dialog")).toHaveCount(0)
  await settings.getByRole("tab", { name: "Servers" }).click()
  await settings.getByRole("button", { name: "Add server" }).click()

@@ -42,9 +41,6 @@ test("server dialog keeps focus above fullscreen settings", async ({ page }) =>
  await expect(password).toBeFocused()
  await password.fill("secret")
  await expect(password).toHaveValue("secret")
-  await page.keyboard.press("Escape")
-  await expect(editor).toBeHidden()
-  await expect(settings).toBeVisible()
 })

 function json(route: Route, body: unknown, status = 200) {
@@ -1,5 +1,5 @@
 import { expect, test, type Page } from "@playwright/test"
-import type { OpenCodeEvent, SessionMessageInfo } from "@opencode-ai/client/promise"
+import type { OpenCodeEvent } from "@opencode-ai/client/promise"
 import { base64Encode } from "@opencode-ai/util/encode"
 import { mockOpenCodeServer } from "../utils/mock-server"
 import { expectAppVisible } from "../utils/waits"
@@ -18,7 +18,7 @@ type InboxRow = {
  delivery: "steer" | "queue"
 }

-function createQueueMock(seed: string[], messages: SessionMessageInfo[] = []) {
+function createQueueMock(seed: string[]) {
  const rows: InboxRow[] = seed.map((text, index) => ({
    id: `inb_seed_${index + 1}`,
    sessionID,
@@ -32,16 +32,13 @@ function createQueueMock(seed: string[], messages: SessionMessageInfo[] = []) {
  const changes: { inboxID: string; action: "cancel" | "steer" }[] = []
  const log: string[] = []
  let sequence = 0
-  const emit = <Type extends OpenCodeEvent["type"]>(
-    type: Type,
-    data: Extract<OpenCodeEvent, { type: Type }>["data"],
-  ) => {
+  const emit = (type: OpenCodeEvent["type"], data: OpenCodeEvent["data"]) => {
    sequence += 1
    events.push({
      id: `evt_queue_${sequence}`,
      type,
      created: Date.now(),
-      durable: { aggregateID: sessionID, seq: sequence, version: type === "session.tool.success" ? 2 : 1 },
+      durable: { aggregateID: sessionID, seq: sequence, version: 1 },
      data,
    } as OpenCodeEvent)
  }
@@ -50,8 +47,6 @@ function createQueueMock(seed: string[], messages: SessionMessageInfo[] = []) {
    prompts,
    changes,
    log,
-    messages,
-    emit,
    events: () => events.splice(0),
    onPrompt: (input: { sessionID: string; body: Record<string, unknown> }) => {
      prompts.push(input.body)
@@ -131,11 +126,10 @@ async function openSession(page: Page, mock: ReturnType<typeof createQueueMock>,
        directory,
        title: "Session queue regression",
        version: "dev",
-        model: { id: "queue-model", providerID: "opencode" },
        time: { created: 1700000000000, updated: 1700000000000 },
      },
    ],
-    pageMessages: () => ({ items: mock.messages }),
+    pageMessages: () => ({ items: [] }),
    sessionStatus: () => ({ [sessionID]: { type: "running" } }),
    inbox: () => mock.rows.map((row) => ({ ...row, payload: { ...row.payload } })),
    onPrompt: mock.onPrompt,
@@ -207,15 +201,12 @@ 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([
@@ -233,130 +224,3 @@ test("editing restores the existing draft and replaces only the original queue p
  expect(mock.changes.map((change) => change.action)).toEqual(["cancel", "cancel", "cancel"])
  expect(mock.log[0]).toBe("prompt:queue")
 })
-
-for (const delivery of ["steer", "queue"] as const) {
-  test(`keeps finished tools above a pending ${delivery === "queue" ? "queue-to-steer" : "steer"} follow-up`, async ({
-    page,
-  }, testInfo) => {
-    const model = { id: "queue-model", providerID: "opencode" }
-    const userID = "msg_queue_initial_user"
-    const assistantID = "msg_queue_continued_assistant"
-    const followUp = "U2: Also check the retry path."
-    const mock = createQueueMock(
-      [],
-      [
-        { id: userID, type: "user", text: "U1: Inspect the queue ordering.", time: { created: 1700000000000 } },
-        {
-          id: "msg_queue_initial_assistant",
-          type: "assistant",
-          agent: "build",
-          model,
-          content: [{ type: "text", text: "A1: I will inspect the current implementation." }],
-          finish: "tool-calls",
-          time: { created: 1700000000001, completed: 1700000000002 },
-        },
-      ],
-    )
-    const view = await openSession(page, mock, delivery)
-    const transcript = page.locator("[data-timeline-virtual-content]")
-    const thinking = transcript.locator('[data-timeline-row="Thinking"]')
-    await expect(transcript.getByText("A1: I will inspect the current implementation.", { exact: true })).toBeVisible()
-    await expect(thinking).toBeVisible()
-    await expect(view.input).toBeEditable()
-    await view.input.fill(followUp)
-    await view.input.press("Enter")
-    await expect.poll(() => mock.rows.map((row) => row.delivery)).toEqual([delivery])
-    await expect(view.input).toHaveText("")
-
-    const inboxID = mock.rows[0].id
-    const pending = transcript.locator(`[data-timeline-row="UserMessage"][data-message-id="${inboxID}"]`)
-    if (delivery === "queue") {
-      const queued = view.rows.filter({ hasText: followUp })
-      await expect(queued).toBeVisible()
-      await expect(pending).toHaveCount(0)
-      await queued.hover()
-      await queued.getByRole("button", { name: "Steer", exact: true }).click()
-      await expect.poll(() => mock.changes).toEqual([{ inboxID, action: "steer" }])
-    }
-    await expect(view.rows).toHaveCount(0)
-    await expect(pending).toContainText(followUp)
-
-    // The next assistant step still belongs to U1: U2 has been admitted, not delivered.
-    mock.emit("session.step.started", { sessionID, assistantMessageID: assistantID, agent: "build", model })
-    for (const tool of [
-      { id: "tool_queue_read", name: "read", input: { path: "src/queue.ts" } },
-      { id: "tool_queue_grep", name: "grep", input: { pattern: "retry", path: "src" } },
-    ]) {
-      const ref = { sessionID, assistantMessageID: assistantID, id: tool.id }
-      mock.emit("session.tool.input.started", { ...ref, name: tool.name })
-      mock.emit("session.tool.input.ended", { ...ref, text: JSON.stringify(tool.input) })
-      mock.emit("session.tool.called", { ...ref, input: tool.input, executed: true })
-      mock.emit("session.tool.success", {
-        ...ref,
-        content: [{ type: "text", text: "Inspection complete." }],
-        executed: true,
-      })
-    }
-    mock.emit("session.step.ended", {
-      sessionID,
-      assistantMessageID: assistantID,
-      finish: "tool-calls",
-      cost: 0,
-      tokens: { input: 100, output: 20, reasoning: 0, cache: { read: 0, write: 0 } },
-    })
-    const tools = page.locator('[data-timeline-part-ids="tool_queue_read,tool_queue_grep"]')
-    await expect(tools).toBeVisible()
-    await expect(tools).toContainText(/Used\s*Read, Grep/)
-    await expect(tools.locator('[data-component="tag"]')).toHaveText("2")
-    await expect(thinking).toBeVisible()
-    await expect(pending).toBeVisible()
-    expect(mock.rows.map((row) => ({ id: row.id, delivery: row.delivery }))).toEqual([
-      { id: inboxID, delivery: "steer" },
-    ])
-    await transcript.screenshot({ path: testInfo.outputPath("pending-steer.png") })
-
-    // Soft assertions let delivery run too, even when the pending ordering regresses.
-    await expect
-      .soft(tools.or(thinking).or(pending))
-      .toHaveText([/Used\s*Read, Grep/, /Thinking/, /U2: Also check the retry path\./])
-    await expect
-      .soft(transcript.locator('[data-timeline-row="AssistantPart"]').filter({ has: tools }))
-      .toHaveAttribute("data-message-id", userID)
-    await expect
-      .configure({ soft: true })
-      .poll(async () => {
-        const boxes = await Promise.all([tools.boundingBox(), thinking.boundingBox(), pending.boundingBox()])
-        return (
-          boxes.every((box) => box !== null) &&
-          boxes[0]!.y + boxes[0]!.height <= boxes[1]!.y &&
-          boxes[1]!.y + boxes[1]!.height <= boxes[2]!.y
-        )
-      })
-      .toBe(true)
-
-    mock.rows.splice(0, 1)
-    mock.emit("session.inbox.delivered", { sessionID, inboxID })
-    await expect(thinking).toHaveAttribute("data-message-id", inboxID)
-    await expect(pending).toHaveCount(1)
-    await expect(transcript.locator('[data-timeline-row="UserMessage"]')).toHaveCount(2)
-    await expect(transcript.locator('[data-timeline-row="AssistantPart"]').filter({ has: tools })).toHaveAttribute(
-      "data-message-id",
-      userID,
-    )
-
-    const later = { sessionID, assistantMessageID: "msg_queue_follow_up_assistant" }
-    mock.emit("session.step.started", { ...later, agent: "build", model })
-    mock.emit("session.text.started", { ...later, ordinal: 0 })
-    mock.emit("session.text.ended", { ...later, ordinal: 0, text: "A3: Now checking the retry path for U2." })
-    const response = transcript
-      .locator('[data-timeline-row="AssistantPart"]')
-      .filter({ hasText: "A3: Now checking the retry path for U2." })
-    await expect(response).toHaveAttribute("data-message-id", inboxID)
-    await expect(tools.or(pending).or(response).or(thinking)).toHaveText([
-      /Used\s*Read, Grep/,
-      /U2: Also check the retry path\./,
-      /A3: Now checking the retry path for U2\./,
-      /Thinking/,
-    ])
-  })
-}
@@ -1,43 +1,16 @@
 import { expect, test } from "@playwright/test"
-import {
-  assistantMessage,
-  setupTimeline,
-  shell,
-  textPart,
-  userMessage,
-} from "../performance/timeline-stability/fixture"
+import { assistantMessage, setupTimeline, shell, 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)),
-        textPart(
-          "prt_space_following",
-          "Following content leaves room to focus the command away from the bottom. ".repeat(40),
-        ),
-      ]),
-    ],
+    messages: [userMessage(), assistantMessage([shell(shellID, "completed", lines(5))])],
    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()
+  const trigger = page.locator(`[data-timeline-part-id="${shellID}"] [data-slot="collapsible-trigger"]`)
  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")
@@ -40,7 +40,7 @@ test.describe("regression: session timeline context group resize", () => {
    expect(samples.at(-1)?.expanded).toBe("true")
  })

-  test("keeps a grouped tool summary stable as its calls complete", async ({ page }) => {
+  test("paints a stable exploring to explored transition", async ({ page }) => {
    const events: OpenCodeEvent[] = []
    await page.setViewportSize({ width: 1400, height: 900 })
    await mockServer(page, events, [
@@ -55,12 +55,13 @@ test.describe("regression: session timeline context group resize", () => {
    await devtools.send("Emulation.setCPUThrottlingRate", { rate: 4 })
    const context = page.locator(`[data-timeline-part-ids="${contextIDs.join(",")}"]`).first()
    await expectAppVisible(context)
-    await expect(context.getByRole("button")).toHaveAccessibleName("Used Read, Glob, Grep, List")
+    await expect(context.locator('[data-component="tool-status-title"]')).toHaveAttribute("aria-label", "Exploring")

    const contextSelector = `[data-timeline-part-ids="${contextIDs.join(",")}"]`
    const regions = defineVisualRegions({
      status: {
-        selector: `${contextSelector} [data-component="context-tool-group-trigger"]`,
+        selector: `${contextSelector} [data-component="tool-status-title"]`,
+        opacitySelectors: ['[data-slot="tool-status-active"]', '[data-slot="tool-status-done"]'],
      },
      context: { selector: contextSelector, closest: '[data-timeline-row="AssistantPart"]' },
      following: {
@@ -88,7 +89,7 @@ test.describe("regression: session timeline context group resize", () => {
      await page.waitForTimeout(delay)
    }

-    await expect(context.getByRole("button")).toHaveAccessibleName("Used Read, Glob, Grep, List")
+    await expect(context.locator('[data-component="tool-status-title"]')).toHaveAttribute("aria-label", "Explored")
    await page.waitForTimeout(700)
    const trace = await stopVisualProbe<keyof typeof regions>(page)
    const labels = trace.samples
@@ -107,7 +108,7 @@ test.describe("regression: session timeline context group resize", () => {
      ]),
    )

-    expect(labels).toEqual(["Used Read, Glob, Grep, List"])
+    expect(labels).toEqual(["Exploring", "Explored"])
    expect(issues, JSON.stringify(trace.samples, null, 2)).toEqual([])
  })
 })
@@ -208,7 +209,13 @@ function turn(index: number, target: boolean, status: "running" | "completed" =
  const content: SessionMessageAssistant["content"] = target
    ? [
        toolContent(
-          contextTool(contextIDs[0]!, assistantID, "read", { path: "src/recent-a.ts", offset: 0, limit: 120 }, status),
+          contextTool(
+            contextIDs[0]!,
+            assistantID,
+            "read",
+            { path: "src/recent-a.ts", offset: 0, limit: 120 },
+            status,
+          ),
        ),
        toolContent(contextTool(contextIDs[1]!, assistantID, "glob", { path: directory, pattern: "**/*.ts" }, status)),
        toolContent(
@@ -8,6 +8,21 @@ 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="${id}"] [data-component="write-content"]`)).toBeVisible()
+})
+
 test("renders a completed single-file patch", async ({ page }) => {
  const id = "prt_file_projection_single_patch"
  await setupTimeline(page, {
@@ -68,7 +83,6 @@ test("keeps an expanded file diff header at the same viewport position", async (
  const before = Array.from({ length: 80 }, (_, index) => `export const value${index} = ${index}\n`).join("")
  const after = before.replaceAll(" = ", " = compute(").replaceAll("\n", ")\n")
  await setupTimeline(page, {
-    settings: { editToolPartsExpanded: true },
    messages: [
      userMessage([userText("Preceding context ".repeat(120))]),
      assistantMessage([
@@ -0,0 +1,53 @@
+import { expect, test } from "@playwright/test"
+import { assistantMessage, setupTimeline, toolPart, userMessage } from "../performance/timeline-stability/fixture"
+import { createTwoFilesPatch } from "diff"
+
+test("keeps patch file disclosures independent", async ({ page }) => {
+  const patchID = "prt_nested_patch"
+  const files = [patchFile("src/a.ts", "modified"), patchFile("src/b.ts", "added"), patchFile("src/old.ts", "deleted")]
+  await setupTimeline(page, {
+    messages: [
+      userMessage(),
+      assistantMessage([
+        toolPart(
+          patchID,
+          "patch",
+          "completed",
+          { patchText: "Update three files" },
+          { metadata: { files } },
+        ),
+      ]),
+    ],
+    settings: { editToolPartsExpanded: true },
+  })
+  const wrapper = page.locator(`[data-timeline-part-id="${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(
+    "",
+  )
+}
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Kit Langton	4f6ea3e8ef	refactor(tui): simplify plugin update controls	2026-08-25 22:02:54 -04:00
Kit Langton	f74d7972eb	fix(tui): refine plugin update states	2026-08-25 22:02:54 -04:00
Kit Langton	0395985b58	docs(cli): document plugin update controls	2026-08-25 22:02:54 -04:00
Kit Langton	7a0944cdd6	refactor(tui): use generated plugin update client	2026-08-25 22:02:54 -04:00
Kit Langton	fe2583cd26	fix(tui): harden plugin update controls	2026-08-25 22:02:54 -04:00
Kit Langton	b63167558c	feat(tui): add plugin update controls	2026-08-25 22:02:54 -04:00
Kit Langton	88ad378d83	refactor(core): simplify plugin updates	2026-08-25 22:02:53 -04:00
Kit Langton	f19ad6fa31	fix(core): harden explicit plugin updates	2026-08-25 22:02:53 -04:00
Kit Langton	1b7332467e	chore(sdk): regenerate types	2026-08-25 22:02:53 -04:00
Kit Langton	e6b6f30914	feat(core): support explicit plugin updates	2026-08-25 22:02:53 -04:00