fix(core): attribute one-shot generation requests

fix(core): raise instruction entry limit (#48350 )
Co-authored-by: nexxeln <nexxeln@users.noreply.github.com>
2026-09-10 19:06:24 +00:00 · 2026-09-10 16:33:50 +00:00 · 2026-09-10 20:59:38 +05:30 · 2026-09-10 15:47:03 +02:00 · 2026-09-10 17:27:43 +05:30 · 2026-09-10 19:24:23 +08:00
20 changed files with 441 additions and 56 deletions
@@ -18,7 +18,6 @@ const WebSocketResponseCreate = Schema.StructWithRest(Schema.Struct({ type: Sche
 ])
 const decodeMessage = ProviderShared.validateWith(Schema.decodeUnknownEffect(WebSocketResponseCreate))
 const encodeMessage = Schema.encodeSync(Schema.fromJsonString(WebSocketResponseCreate))
-const decodeEvent = Schema.decodeUnknownEffect(OpenResponses.protocol.stream.event)

 export interface Options {
  readonly id: string
@@ -27,6 +26,7 @@ export interface Options {
  readonly enabled?: (url: string) => boolean
  readonly url?: (url: string) => string
  readonly headers?: (headers: Headers.Headers) => Headers.Headers
+  readonly continuation?: OpenResponsesContinuation.Shape
 }

 export interface Prepared {
@@ -60,7 +60,7 @@ const driver = (options: Options, body: string): WebSocketChannelDriver => {
      }),
    observe: (_create, frame) =>
      Effect.gen(function* () {
-        const event = yield* decodeEvent(frame).pipe(
+        const event = yield* OpenResponses.decodeChannelEvent(frame).pipe(
          Effect.mapError((cause) =>
            ProviderShared.eventError(options.id, `Invalid ${options.name} WebSocket event`, frame, cause),
          ),
@@ -163,6 +163,7 @@ export const transport = <Body>(options: Options): Transport<Body, Prepared, str
                    request: create.request,
                    message: create.message,
                    base,
+                    continuation: options.continuation,
                  }),
                }
              })
@@ -6,7 +6,6 @@ import { OpenResponses } from "./open-responses.js"

 const PROTOCOL = "open-responses.websocket.v1"
 const VERSION = 1
-const decodeEvent = Schema.decodeUnknownEffect(OpenResponses.protocol.stream.event)

 interface CheckpointValue {
  readonly version: typeof VERSION
@@ -15,12 +14,19 @@ interface CheckpointValue {
  readonly output: ReadonlyArray<unknown>
 }

+/**
+ * Fields to send next to `previous_response_id` on an incremental step, or undefined to send the step in full.
+ * Whether omitted fields carry over from the continued response is provider behavior the route must know.
+ */
+export type Shape = (request: Readonly<Record<string, unknown>>) => Readonly<Record<string, unknown>> | undefined
+
 export interface DriverInput {
  readonly id: string
  readonly name: string
  readonly request: Readonly<Record<string, unknown>>
  readonly message: string
  readonly base: WebSocketChannelDriver
+  readonly continuation?: Shape
 }

 const checkpointValue = (checkpoint: ChannelCheckpoint | undefined): CheckpointValue | undefined => {
@@ -127,22 +133,26 @@ const rejected = (

 export const driver = (input: DriverInput): WebSocketChannelDriver => {
  const { previous_response_id: _previousResponseID, ...request } = input.request
+  const shape = input.continuation ?? ((fields: Readonly<Record<string, unknown>>) => fields)
  let output: OpenResponses.StreamItem[] = []
  return {
    create: (checkpoint) =>
      Effect.sync(() => {
        output = []
        const previous = checkpointValue(checkpoint)
-        const delta = previous ? incremental(request, previous) : undefined
-        if (!previous || !delta) return { message: ProviderShared.encodeJson(request), mode: "full" as const }
+        // Ask the route first: diffing the whole history is wasted when it declines the continuation.
+        const fields = previous ? shape(request) : undefined
+        const delta = previous && fields ? incremental(request, previous) : undefined
+        if (!previous || !fields || !delta)
+          return { message: ProviderShared.encodeJson(request), mode: "full" as const }
        return {
-          message: ProviderShared.encodeJson({ ...request, input: delta, previous_response_id: previous.responseID }),
+          message: ProviderShared.encodeJson({ ...fields, input: delta, previous_response_id: previous.responseID }),
          mode: "incremental" as const,
        }
      }),
    observe: (create, frame) =>
      Effect.gen(function* () {
-        const event = yield* decodeEvent(frame).pipe(
+        const event = yield* OpenResponses.decodeChannelEvent(frame).pipe(
          Effect.mapError((cause) =>
            ProviderShared.eventError(input.id, `Invalid ${input.name} WebSocket event`, frame, cause),
          ),
@@ -195,4 +205,4 @@ export const driver = (input: DriverInput): WebSocketChannelDriver => {
  }
 }

-export const OpenResponsesContinuation = { driver } as const
+export * as OpenResponsesContinuation from "./open-responses-continuation.js"
@@ -405,6 +405,24 @@ export const Event = Schema.StructWithRest(
 export type Event = Schema.Schema.Type<typeof Event>
 export type NormalizedEvent = Event & { readonly item?: OutputItem | null }

+const decodeEventValue = Schema.decodeUnknownEffect(Event)
+const decodeFrame = Schema.decodeUnknownEffect(ProviderShared.Json)
+
+/**
+ * Decodes one WebSocket frame. xAI answers a rejected `response.create` with `{ "error": { "message", "type" } }` and no
+ * event type; that envelope reads as an error event so the failure classifies instead of failing decoding.
+ */
+export const decodeChannelEvent = (frame: string) =>
+  decodeFrame(frame).pipe(
+    Effect.flatMap((value) =>
+      decodeEventValue(
+        ProviderShared.isRecord(value) && value.type === undefined && ProviderShared.isRecord(value.error)
+          ? { ...value, type: "error" }
+          : value,
+      ),
+    ),
+  )
+
 export interface ProviderAdapter {
  readonly id: string
  readonly name: string
@@ -41,6 +41,10 @@ const responsesRoute = Route.make({
    id: "openai-responses",
    name: "xAI Responses",
    rotateAfterMs: RESPONSES_WEBSOCKET_ROTATE_AFTER_MS,
+    // xAI continues a chain only from stored responses: with `store: false` (the route default) `previous_response_id`
+    // fails with "Response with id=… not found", so those steps are sent in full over the reused connection. It also
+    // rejects `instructions` next to `previous_response_id` and keeps the instructions of the response it continues.
+    continuation: ({ instructions: _instructions, ...request }) => (request.store === false ? undefined : request),
  }),
  defaults: { providerOptions: { store: false, include: ["reasoning.encrypted_content"] } },
 })
@@ -90,7 +90,11 @@ const classifyingChannelDriver = (message: string): WebSocketChannelDriver => {
  }
 }

-const continuationDriver = (request: Readonly<Record<string, unknown>>, base = baseChannelDriver) => {
+const continuationDriver = (
+  request: Readonly<Record<string, unknown>>,
+  base = baseChannelDriver,
+  continuation?: OpenResponsesContinuation.Shape,
+) => {
  const message = ProviderShared.encodeJson(request)
  return OpenResponsesContinuation.driver({
    id: "openai-responses",
@@ -98,6 +102,7 @@ const continuationDriver = (request: Readonly<Record<string, unknown>>, base = b
    request,
    message,
    base: base(message),
+    continuation,
  })
 }

@@ -921,6 +926,58 @@ describe("OpenAI Responses route", () => {
        type: "provider-failure",
        error: { reason: { _tag: "InvalidRequest", classification: "context-overflow" } },
      })
+
+      // A retryable failure stays one: the runner retries it, and the transport has already dropped the
+      // checkpoint, so that retry is a full send. xAI reports every rejection this way.
+      const internal = ProviderShared.encodeJson({
+        type: "error",
+        error: { type: "api_error", message: "gRPC error: Response with id=resp_1 not found" },
+      })
+      expect(yield* second.observe(yield* second.create(saved), internal)).toMatchObject({
+        type: "provider-failure",
+        error: { reason: { _tag: "ProviderInternal" } },
+      })
+    }),
+  )
+
+  it.effect("shapes the incremental send with the route continuation", () =>
+    Effect.gen(function* () {
+      const firstRequest = {
+        type: "response.create",
+        model: "grok-4.6",
+        store: true,
+        instructions: "You are terse.",
+        input: [{ role: "user", content: [{ type: "input_text", text: "First" }] }],
+      }
+      const secondRequest = {
+        ...firstRequest,
+        input: [...firstRequest.input, { role: "user", content: [{ type: "input_text", text: "Second" }] }],
+      }
+      const saved = checkpoint(
+        yield* continuationDriver(firstRequest).observe(
+          yield* continuationDriver(firstRequest).create(undefined),
+          ProviderShared.encodeJson({ type: "response.completed", response: { id: "resp_1" } }),
+        ),
+      )
+
+      const trimmed = yield* continuationDriver(
+        secondRequest,
+        baseChannelDriver,
+        ({ instructions: _, ...rest }) => rest,
+      ).create(saved)
+      expect(trimmed.mode).toBe("incremental")
+      expect(JSON.parse(trimmed.message)).toEqual({
+        type: "response.create",
+        model: "grok-4.6",
+        store: true,
+        previous_response_id: "resp_1",
+        input: [{ role: "user", content: [{ type: "input_text", text: "Second" }] }],
+      })
+
+      // Declining the continuation sends the step in full and never sends a previous_response_id.
+      const declined = yield* continuationDriver(secondRequest, baseChannelDriver, () => undefined).create(saved)
+      expect(declined.mode).toBe("full")
+      expect(JSON.parse(declined.message)).toEqual(secondRequest)
    }),
  )

@@ -1,11 +1,18 @@
 import { describe, expect } from "bun:test"
-import { Effect } from "effect"
+import { Effect, Layer, Stream } from "effect"
 import { LLM, LLMEvent, Message } from "../../src/index.js"
 import { XAI } from "../../src/providers.js"
 import { OpenResponses } from "../../src/protocols/open-responses.js"
 import { OpenAIResponses } from "../../src/protocols/openai-responses.js"
+import * as ProviderShared from "../../src/protocols/shared.js"
 import { XAIResponses } from "../../src/protocols/xai-responses.js"
-import { LLMClient } from "../../src/route.js"
+import {
+  LLMClient,
+  RequestExecutor,
+  WebSocketTransport,
+  type ChannelCheckpoint,
+  type WebSocketChannelDriver,
+} from "../../src/route.js"
 import { compileRequest } from "../../src/route/client.js"
 import { it } from "../lib/effect.js"
 import { fixedResponse } from "../lib/http.js"
@@ -13,6 +20,35 @@ import { sseEvents } from "../lib/sse.js"

 const model = XAI.configure({ apiKey: "test", baseURL: "https://api.x.ai/v1" }).responses("grok-4.6")

+/** Runs a request through the WebSocket transport and hands back its channel driver; the HTTP fallback answers. */
+const channelDriver = (request: ReturnType<typeof LLM.request>) =>
+  Effect.gen(function* () {
+    let driver: WebSocketChannelDriver | undefined
+    yield* LLMClient.generate(request, {
+      webSocket: {
+        execute: (exchange) =>
+          Effect.sync(() => {
+            driver = exchange.driver
+            return { frames: exchange.fallback(), complete: Effect.void }
+          }),
+      },
+    }).pipe(Effect.provide(fixedResponse(sseEvents({ type: "response.completed", response: { id: "http" } }))))
+    if (!driver) throw new Error("Expected a WebSocket channel driver")
+    return driver
+  })
+
+const completed = (driver: WebSocketChannelDriver, id: string) =>
+  Effect.gen(function* () {
+    const create = yield* driver.create(undefined)
+    yield* driver.observe(create, ProviderShared.encodeJson({ type: "response.created", response: { id } }))
+    const observation = yield* driver.observe(
+      create,
+      ProviderShared.encodeJson({ type: "response.completed", response: { id } }),
+    )
+    if (observation.type !== "completed" || !observation.checkpoint) throw new Error("Expected a checkpoint")
+    return observation.checkpoint
+  })
+
 describe("xAI Responses route", () => {
  it.effect("composes the Open Responses baseline with xAI extensions", () =>
    Effect.gen(function* () {
@@ -162,6 +198,78 @@ describe("xAI Responses route", () => {
    }),
  )

+  it.effect("classifies xAI's untyped WebSocket error envelope", () =>
+    Effect.gen(function* () {
+      // xAI answers a rejected response.create with an error envelope that carries no event type.
+      const envelope = ProviderShared.encodeJson({
+        error: {
+          message:
+            'Request validation error: {"code":"400","error":"Argument not supported: instructions and previous_response_id together"}',
+          type: "api_error",
+        },
+      })
+      const webSocket = WebSocketTransport.makeDirect({
+        open: () =>
+          Effect.succeed({ sendText: () => Effect.void, messages: Stream.make(envelope), close: Effect.void }),
+      })
+      const error = yield* LLMClient.generate(LLM.request({ model, prompt: "Hello" }), { webSocket }).pipe(
+        Effect.provide(
+          LLMClient.layer.pipe(
+            Layer.provide(
+              Layer.succeed(
+                RequestExecutor.Service,
+                RequestExecutor.Service.of({ execute: () => Effect.die("unexpected HTTP request") }),
+              ),
+            ),
+          ),
+        ),
+        Effect.flip,
+      )
+
+      expect(error.reason._tag).toBe("ProviderInternal")
+      expect(error.message).toContain("Argument not supported: instructions and previous_response_id together")
+      expect(error.reason.body).toBe(envelope)
+    }),
+  )
+
+  it.effect("continues stored responses without instructions and sends unstored steps in full", () =>
+    Effect.gen(function* () {
+      const step = (store: boolean, ...prompts: string[]) =>
+        LLM.request({
+          model,
+          system: "You are terse.",
+          messages: prompts.map((prompt) => Message.user(prompt)),
+          providerOptions: { store },
+        })
+      const send = (store: boolean, checkpoint: ChannelCheckpoint) =>
+        channelDriver(step(store, "First", "Second")).pipe(Effect.flatMap((driver) => driver.create(checkpoint)))
+
+      const stored = yield* send(true, yield* completed(yield* channelDriver(step(true, "First")), "resp_1"))
+      expect(stored.mode).toBe("incremental")
+      expect(JSON.parse(stored.message)).toEqual({
+        type: "response.create",
+        model: "grok-4.6",
+        store: true,
+        include: ["reasoning.encrypted_content"],
+        previous_response_id: "resp_1",
+        input: [{ role: "user", content: [{ type: "input_text", text: "Second" }] }],
+      })
+
+      // The connection cache only serves stored responses, so the default store: false never chains.
+      const unstored = yield* send(false, yield* completed(yield* channelDriver(step(false, "First")), "resp_1"))
+      expect(unstored.mode).toBe("full")
+      expect(JSON.parse(unstored.message)).toMatchObject({
+        instructions: "You are terse.",
+        store: false,
+        input: [
+          { role: "user", content: [{ type: "input_text", text: "First" }] },
+          { role: "user", content: [{ type: "input_text", text: "Second" }] },
+        ],
+      })
+      expect(JSON.parse(unstored.message).previous_response_id).toBeUndefined()
+    }),
+  )
+
  it.effect("parses xAI hosted tool items", () =>
    Effect.gen(function* () {
      const item = { type: "x_search_call", id: "x_search_1", status: "completed", action: { query: "news" } }
@@ -0,0 +1,74 @@
+import { expect, test } from "@playwright/test"
+import { sessionID, setupTimeline, userMessage } from "../performance/timeline-stability/fixture"
+
+test("keeps a submitted prompt in place while its optimistic rows are measured", async ({ page }) => {
+  await setupTimeline(page, { messages: [userMessage()], seedHistory: true })
+  const release = Promise.withResolvers<void>()
+  await page.route(`**/api/session/${sessionID}/prompt`, async (route) => {
+    if (route.request().method() !== "POST") return route.fallback()
+    await release.promise
+    return route.fallback()
+  })
+
+  const editor = page.locator('[data-component="composer"]').getByRole("textbox")
+  await expect(editor).toBeEditable()
+  await editor.fill("Observe optimistic prompt spacing.")
+  await expect
+    .poll(() =>
+      page.locator("[data-timeline-virtual-content]").evaluate((element) => {
+        const root = element.parentElement!
+        return root.scrollHeight - root.clientHeight - root.scrollTop
+      }),
+    )
+    .toBe(0)
+
+  const observation = await page.evaluateHandle(() => {
+    const frames: { prompt?: number; working: boolean }[] = []
+    let frame = 0
+    const sample = () => {
+      const prompt = [...document.querySelectorAll<HTMLElement>('[data-timeline-row="UserMessage"]')].find((row) =>
+        row.textContent?.includes("Observe optimistic prompt spacing."),
+      )
+      frames.push({
+        ...(prompt ? { prompt: prompt.getBoundingClientRect().y } : {}),
+        working: !!document.querySelector('[data-component="session-working"]'),
+      })
+      frame = requestAnimationFrame(sample)
+    }
+    frame = requestAnimationFrame(sample)
+    return {
+      stop: () => {
+        cancelAnimationFrame(frame)
+        return frames
+      },
+    }
+  })
+  const requested = page.waitForRequest(
+    (request) => request.method() === "POST" && new URL(request.url()).pathname === `/api/session/${sessionID}/prompt`,
+  )
+  try {
+    await editor.press("Enter")
+    await requested
+    const prompt = page
+      .locator('[data-timeline-row="UserMessage"]')
+      .filter({ hasText: "Observe optimistic prompt spacing." })
+    await expect(prompt).toBeInViewport()
+    await expect(page.locator('[data-component="session-working"]')).toBeVisible()
+    await expect
+      .poll(() =>
+        page.locator("[data-timeline-virtual-content]").evaluate((element) => {
+          const root = element.parentElement!
+          return root.scrollHeight - root.clientHeight - root.scrollTop
+        }),
+      )
+      .toBe(0)
+    const frames = await observation.evaluate((value) => value.stop())
+    expect(frames.some((frame) => frame.working && frame.prompt === undefined)).toBe(false)
+    const positions = frames.flatMap((frame) => (frame.prompt === undefined ? [] : [frame.prompt]))
+    expect(positions.length).toBeGreaterThan(0)
+    expect(new Set(positions).size).toBe(1)
+  } finally {
+    release.resolve()
+    await observation.dispose()
+  }
+})
@@ -122,7 +122,15 @@ test("renders compaction progress, summary, and outcome in order", async ({ page
  )
  await expect(compaction.getByRole("heading", { name: "Checkpoint" })).toBeVisible()
  await expect(compaction).toContainText("Streamed implementation details.")
-  await expect(compaction.getByRole("status").getByLabel("Compacting", { exact: true })).toBeVisible()
+  const running = compaction.getByRole("status").getByLabel("Compacting", { exact: true })
+  await expect(running).toBeVisible()
+  await expect
+    .poll(async () => {
+      const summary = await compaction.locator('[data-component="text-part"]').boundingBox()
+      const status = await running.boundingBox()
+      return !!summary && !!status && status.y >= summary.y + summary.height
+    })
+    .toBe(true)
  await expect(compaction.getByText("Session compacted", { exact: true })).toHaveCount(0)

  await timeline.send(
@@ -88,8 +88,12 @@ export function createComposerSubmit(input: ComposerSubmitInput) {
      if (value.mode === "normal" && !command) {
        session.handoff?.set(handoffMessage(value))
        const optimisticBusy = !input.adapter.working()
-        if (optimisticBusy) session.data.session.setStatus(session.id, "running")
-        const sending = sendPrompt(session, value, input.adapter.controls().model.selection.trackSessionCommit).then(
+        if (optimisticBusy && input.adapter.kind === "new-session")
+          session.data.session.setStatus(session.id, "running")
+        const sending = sendPrompt(session, value, input.adapter.controls().model.selection.trackSessionCommit, () => {
+          if (optimisticBusy && input.adapter.kind === "active-session")
+            session.data.session.setStatus(session.id, "running")
+        }).then(
          () => ({ ok: true as const }),
          (error) => ({ ok: false as const, error }),
        )
@@ -122,15 +126,9 @@ export function createComposerSubmit(input: ComposerSubmitInput) {

      if (command) {
        clearSubmission(input, submission)
-        // Commands always steer: the server applies a command's configured
-        // agent and model immediately at admission, so queueing one would
-        // reconfigure the turn it is supposed to wait behind.
-        void sendCommand(
-          session,
-          { ...value, delivery: "steer" },
-          command,
-          input.adapter.controls().model.selection.trackSessionCommit,
-        ).catch((error) => failSubmission(input, session, "command", error, restore, value.id))
+        void sendCommand(session, value, command, input.adapter.controls().model.selection.trackSessionCommit).catch(
+          (error) => failSubmission(input, session, "command", error, restore, value.id),
+        )
        return
      }
    } finally {
@@ -322,7 +320,8 @@ async function sendCommand(
  track?: ModelSelection["trackSessionCommit"],
 ) {
  const request = await buildSubmissionRequest(session, value)
-  await applySelection(session, value.selection, track)
+  // Like queued prompts, queued commands must not apply the composer's selection to active work.
+  if (value.delivery === "steer") await applySelection(session, value.selection, track)
  await session.api.command({
    sessionID: session.id,
    command: command.command,
@@ -359,7 +358,8 @@ async function applySelection(
 async function sendPrompt(
  session: ComposerSession,
  value: ComposerSubmission,
-  track?: ModelSelection["trackSessionCommit"],
+  track: ModelSelection["trackSessionCommit"] | undefined,
+  onAdmit: () => void,
 ) {
  const request = await buildSubmissionRequest(session, value)
  // Switching agent or model reconfigures the session immediately, and with it
@@ -389,7 +389,9 @@ async function sendPrompt(
      },
    },
  }
-  await session.data.session.prompt(admission).catch(() => session.data.session.prompt(admission))
+  const sending = session.data.session.prompt(admission).catch(() => session.data.session.prompt(admission))
+  onAdmit()
+  await sending
 }

 async function buildSubmissionRequest(session: ComposerSession, value: ComposerSubmission) {
@@ -515,6 +515,19 @@ export function createTimelineVirtualizer(input: Input) {
          <div
            ref={(value) => {
              element = value
+              if (row()._tag !== "UserMessage" || !addedKeys.has(rowProps.rowKey) || !input.pinned() || coldPending)
+                return
+              // The optimistic row can paint before ResizeObserver corrects the tail estimates.
+              // Measure the mounted tail and pin it in this render's microtask instead.
+              queueMicrotask(() => {
+                if (!input.pinned() || !virtualContent?.isConnected) return
+                virtualizer.elementsCache.forEach((item) => {
+                  if (item.isConnected) virtualizer.resizeItem(virtualizer.indexFromElement(item), item.offsetHeight)
+                })
+                virtualizer.resizeItem(item().index, element.offsetHeight)
+                virtualContent.style.height = `${virtualizer.getTotalSize()}px`
+                virtualizer.scrollToEnd()
+              })
            }}
            data-index={item().index}
            style={{ "min-height": ready() ? undefined : `${initialItem.size}px` }}
@@ -1,6 +1,7 @@
 export * as Generate from "./generate.js"

 import { LLM, LLMClient, AIError } from "@opencode/ai"
+import { SessionID } from "@opencode/schema/session-id"
 import { Context, Effect, Layer, Schema } from "effect"
 import { makeLocationNode } from "@opencode/util/effect/app-node"
 import { llmClient } from "./effect/app-node-platform.js"
@@ -58,15 +59,24 @@ export const layer = Layer.effect(
            ? `Model unavailable: ${input.model.providerID}/${input.model.id}`
            : "No model specified and no supported model is available",
        })
-      const response = yield* llm.generate(LLM.request({ model: resolved.model, prompt: input.prompt })).pipe(
-        Effect.mapError(
-          (error: AIError) =>
-            new UnavailableError({
-              message: error.message,
-              service: resolved.ref.providerID,
-            }),
-        ),
-      )
+      const response = yield* llm
+        .generate(
+          LLM.request({
+            model: resolved.model,
+            prompt: input.prompt,
+            // Gateways require session attribution even for a stateless call; no Session is stored.
+            http: { headers: { "x-opencode-session": SessionID.create() } },
+          }),
+        )
+        .pipe(
+          Effect.mapError(
+            (error: AIError) =>
+              new UnavailableError({
+                message: error.message,
+                service: resolved.ref.providerID,
+              }),
+          ),
+        )
      return response.text
    })

@@ -101,6 +101,7 @@ export const XAIPlugin = define({
      for (const model of provider.models.values()) {
        catalog.model.update(providerID, model.id, (draft) => {
          draft.capabilities.responsesWebsockets = true
+          draft.websocket = true
        })
      }
    })
@@ -1,5 +1,6 @@
 import { expect } from "bun:test"
-import { LanguageModel } from "@opencode/ai"
+import { LanguageModel, LLMClient } from "@opencode/ai"
+import { RequestExecutor } from "@opencode/ai/route"
 import { OpenAIChat } from "@opencode/ai/protocols"
 import { TestLLM } from "@opencode/ai/testing"
 import { AISDK } from "@opencode/core/aisdk"
@@ -11,6 +12,7 @@ import { ID, Info, Ref } from "@opencode/core/model"
 import { Provider } from "@opencode/core/provider"
 import { Npm } from "@opencode/util/npm"
 import { Effect, Layer } from "effect"
+import { HttpClient, HttpClientResponse } from "effect/unstable/http"
 import { testEffect } from "./lib/effect"

 const selected = Info.make({
@@ -98,3 +100,56 @@ resolverIt.effect("resolves dynamic models with their catalog metadata", () =>
    })
  }),
 )
+
+testEffect(Layer.empty).effect("attributes each stateless completion without creating a stored session", () =>
+  Effect.gen(function* () {
+    const sessions: string[] = []
+    const http = Layer.succeed(
+      HttpClient.HttpClient,
+      HttpClient.make((request) =>
+        Effect.sync(() => {
+          const session = request.headers["x-opencode-session"]
+          if (!session)
+            return HttpClientResponse.fromWeb(
+              request,
+              Response.json(
+                {
+                  error: { type: "MissingSessionID", message: "Session ID is required" },
+                },
+                { status: 400 },
+              ),
+            )
+          sessions.push(session)
+          return HttpClientResponse.fromWeb(
+            request,
+            new Response(
+              `data: ${JSON.stringify({
+                id: "completion",
+                object: "chat.completion.chunk",
+                created: 1,
+                model: "gemini",
+                choices: [{ index: 0, delta: { content: "OK" }, finish_reason: "stop" }],
+              })}\n\ndata: [DONE]\n\n`,
+              { headers: { "content-type": "text/event-stream" } },
+            ),
+          )
+        }),
+      ),
+    )
+    const native = LLMClient.layer.pipe(Layer.provide(RequestExecutor.layer.pipe(Layer.provide(http))))
+    yield* Effect.gen(function* () {
+      const generate = yield* Generate.Service
+      for (let index = 0; index < 2; index++) {
+        expect(
+          yield* generate.text({
+            prompt: "Return exactly OK",
+            model: Ref.make({ providerID: selected.providerID, id: selected.id }),
+          }),
+        ).toBe("OK")
+      }
+    }).pipe(Effect.provide(Generate.layer.pipe(Layer.provide(Layer.merge(resolver, native)))))
+    expect(sessions).toHaveLength(2)
+    expect(sessions[0]).toStartWith("ses_")
+    expect(sessions[1]).not.toBe(sessions[0])
+  }),
+)
@@ -69,7 +69,7 @@ describe("XAIPlugin", () => {
    }),
  )

-  it.effect("keeps xAI Responses WebSockets opt-in", () =>
+  it.effect("enables xAI Responses WebSockets", () =>
    Effect.gen(function* () {
      const catalog = yield* Catalog.Service
      const providerID = Provider.ID.make("xai")
@@ -84,7 +84,7 @@ describe("XAIPlugin", () => {

      const model = yield* catalog.model.get(providerID, Model.ID.make("grok-4.6"))
      expect(model?.capabilities.responsesWebsockets).toBe(true)
-      expect(model?.websocket).toBeUndefined()
+      expect(model?.websocket).toBe(true)
    }),
  )
 })
@@ -1944,7 +1944,16 @@ describe("SessionRunnerLLM", () => {
    expect(yield* entries.list(sessionID)).toEqual([{ key: "nullable", value: null }])
  })

-  scenario("rejects API instruction entries larger than 8KB", function* () {
+  scenario("accepts API instruction entries up to 256 KiB", function* () {
+    const entries = yield* InstructionEntry.Service
+    const value = "x".repeat(InstructionEntry.MaxValueBytes - 2)
+
+    yield* entries.put({ sessionID, key: "large", value })
+
+    expect(yield* entries.list(sessionID)).toEqual([{ key: "large", value }])
+  })
+
+  scenario("rejects API instruction entries larger than 256 KiB", function* () {
    const entries = yield* InstructionEntry.Service

    const exit = yield* entries
@@ -27,7 +27,7 @@ export const Snapshot = Schema.Array(
 ).annotate({ identifier: "InstructionEntry.Snapshot" })
 export type Snapshot = typeof Snapshot.Type

-export const MaxValueBytes = 8 * 1024
+export const MaxValueBytes = 256 * 1024

 export class ValueTooLargeError extends Schema.TaggedError<ValueTooLargeError>()(
  "InstructionEntryValueTooLargeError",
@@ -427,17 +427,6 @@ export function SessionCompactionMessage(props: { message: SessionMessageCompact
      <div class="py-2">
        <TimelineSeparator label={i18n.t("ui.messagePart.compaction.started")} />
      </div>
-      <Show when={props.message.status === "running"}>
-        <div role="status" class="py-2">
-          <BasicTool
-            icon="archive"
-            trigger={{ title: i18n.t("ui.messagePart.compaction.running") }}
-            status="running"
-            locked
-            hideDetails
-          />
-        </div>
-      </Show>
      <Show when={summary().trim()}>
        <div data-component="text-part" data-timeline-part-id={props.message.id}>
          <div data-slot="text-part-body">
@@ -449,6 +438,17 @@ export function SessionCompactionMessage(props: { message: SessionMessageCompact
          </div>
        </div>
      </Show>
+      <Show when={props.message.status === "running"}>
+        <div role="status" class="py-2">
+          <BasicTool
+            icon="archive"
+            trigger={{ title: i18n.t("ui.messagePart.compaction.running") }}
+            status="running"
+            locked
+            hideDetails
+          />
+        </div>
+      </Show>
      <Show when={props.message.status !== "running"}>
        <div class="py-2">
          <TimelineSeparator label={label()} />
@@ -51,7 +51,8 @@ export function createClipboardAdapter(clipboard: CoreClipboardService): OwnedCl
      throw new Error(`Unexpected clipboard MIME type: ${result.representation.mimeType}`)
    },
    async write(text) {
-      const result = await clipboard.writeText(text, {
+      // OpenTUI rejects NUL before any destination; host clipboard text cannot contain it.
+      const result = await clipboard.writeText(text.replaceAll("\0", ""), {
        destination: "all-available",
        selection: "clipboard",
      })
@@ -102,6 +102,19 @@ test("uses all available routes but skips the process host remotely", async () =
  expect(writes).toEqual({ host: 0, terminal: 1 })
 })

+test("removes NUL characters before writing", async () => {
+  const writes: string[] = []
+  const clipboard = createClipboardAdapter(
+    coreClipboard({
+      onWrite: (text) => writes.push(text),
+    }),
+  )
+
+  expect(await clipboard.write("before\0after")).toBeUndefined()
+  expect(await clipboard.write("clean")).toBeUndefined()
+  expect(writes).toEqual(["beforeafter", "clean"])
+})
+
 test("rejects only when no clipboard route accepted the write", async () => {
  const writes: [string, ClipboardWriteOptions][] = []
  const failure = new Error("native clipboard failed")
@@ -110,10 +110,11 @@ role for other Foundry models. If a request fails because the token belongs to a

 ## WebSocket transport

-OpenAI and supported Azure Responses models keep one WebSocket connection open per session and send each step over it
-instead of opening a new HTTP request. While the request prefix is unchanged, consecutive steps only transmit what was
-added since the previous response, which cuts upload volume on long sessions. Provider compaction runs over the same
-connection.
+OpenAI, xAI, and supported Azure Responses models keep one WebSocket connection open per session and send each step
+over it instead of opening a new HTTP request. While the request prefix is unchanged, consecutive steps only transmit
+what was added since the previous response, which cuts upload volume on long sessions. OpenAI provider compaction runs
+over the same connection. xAI continues a chain only from stored responses, so with its default `store: false` each step
+is sent in full over the reused connection.

 The connection is transparent. When the provider closes the socket, the next step reconnects; when a connection cannot
 be opened at all, the session continues over HTTP. Plugins that register `http.request` or `http.response` hooks for a
Author	SHA1	Message	Date
OpenCode Agent	37b2c91937	fix(core): attribute one-shot generation requests	2026-09-10 16:33:50 +00:00
opencode-agent[bot]andnexxeln	9e153ce7b3	fix(core): raise instruction entry limit (#48350 ) Co-authored-by: nexxeln <nexxeln@users.noreply.github.com>	2026-09-10 20:59:38 +05:30
Simon Klee	eb357f17cf	fix(tui): strip NUL characters before clipboard writes (#48337 )	2026-09-10 15:47:03 +02:00
Shoubhit Dash	573d76933f	fix(ai): make xAI Responses websockets work and enable them (#48318 )	2026-09-10 17:27:43 +05:30
usrnk1andBrendonovich	bb8194395a	feat(desktop): respect follow-up behavior for slash commands (#48169 ) Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>	2026-09-10 19:24:23 +08:00
opencode-agent[bot]andHona	2e8ed86658	fix(app): move compaction status below summary (#48262 ) Co-authored-by: Hona <10430890+Hona@users.noreply.github.com>	2026-09-10 18:27:08 +08:00
Brendan Allan	2695607fbc	fix(app): stabilize optimistic prompt position (#46723 )	2026-09-10 17:49:55 +08:00