saeeol 1.2.9 → 1.3.1

package/test/saeeol/session-prompt-queue.test.ts

@@ -0,0 +1,815 @@
+import path from "path"
+import { describe, expect, test } from "bun:test"
+import { Effect } from "effect"
+import { Bus } from "../../src/bus"
+import { SaeeolSessionPromptQueue } from "@/saeeol/session/prompt-queue"
+import { Suggestion } from "../../src/overlay/suggestion"
+import { Question } from "../../src/question"
+import { ModelID, ProviderID } from "../../src/provider/schema"
+import { Instance } from "../../src/project/instance"
+import { Session } from "../../src/session/session"
+import { MessageV2 } from "../../src/session/message-v2"
+import { SessionCompaction } from "../../src/session/compaction"
+import { SessionPrompt } from "../../src/session/prompt"
+import { MessageID, SessionID } from "../../src/session/schema"
+import * as Log from "@saeeol/core/util/log"
+import { tmpdir } from "../fixture/fixture"
+
+Log.init({ print: false })
+
+function line(input: unknown) {
+  return `data: ${JSON.stringify(input)}\n\n`
+}
+
+function chunk(input: { delta?: Record<string, unknown>; finish?: string }) {
+  return {
+    id: "chatcmpl-queue-test",
+    object: "chat.completion.chunk",
+    choices: [
+      {
+        delta: input.delta ?? {},
+        ...(input.finish ? { finish_reason: input.finish } : {}),
+      },
+    ],
+  }
+}
+
+function reply(input: { text: string; ready?: () => void; wait?: Promise<unknown> }) {
+  const enc = new TextEncoder()
+  const head = line(chunk({ delta: { role: "assistant" } }))
+  const tail = [
+    line(chunk({ delta: { content: input.text } })),
+    line(chunk({ finish: "stop" })),
+    "data: [DONE]\n\n",
+  ].join("")
+
+  return new ReadableStream<Uint8Array>({
+    start(ctrl) {
+      ctrl.enqueue(enc.encode(head))
+      input.ready?.()
+      const done = () => {
+        ctrl.enqueue(enc.encode(tail))
+        ctrl.close()
+      }
+      if (input.wait) {
+        void input.wait.then(done)
+        return
+      }
+      done()
+    },
+  })
+}
+
+function hasText(msg: Awaited<ReturnType<typeof SessionPrompt.prompt>>, text: string) {
+  return msg.parts.some((part) => part.type === "text" && part.text.includes(text))
+}
+
+// Find the last non-system message in an OpenAI-compatible request body. Kept
+// tolerant: we only care about role invariants, not the exact content shape,
+// because providers may serialize `content` as a string or as a parts array.
+function lastConversational(body: Record<string, unknown>): { role: string; content: unknown } | undefined {
+  const msgs = Array.isArray(body.messages) ? (body.messages as Array<Record<string, unknown>>) : []
+  for (let i = msgs.length - 1; i >= 0; i--) {
+    const msg = msgs[i]
+    if (!msg || typeof msg !== "object") continue
+    const role = typeof msg.role === "string" ? msg.role : undefined
+    if (!role || role === "system") continue
+    return { role, content: msg.content }
+  }
+  return undefined
+}
+
+function user(sessionID: SessionID, id: MessageID): MessageV2.WithParts {
+  return {
+    info: {
+      id,
+      sessionID,
+      role: "user",
+      time: { created: 1 },
+      agent: "code",
+      model: { providerID: ProviderID.make("test"), modelID: ModelID.make("model") },
+    },
+    parts: [],
+  }
+}
+
+function assistant(sessionID: SessionID, id: MessageID, parentID: MessageID): MessageV2.WithParts {
+  return {
+    info: {
+      id,
+      sessionID,
+      role: "assistant",
+      time: { created: 1, completed: 2 },
+      parentID,
+      modelID: ModelID.make("model"),
+      providerID: ProviderID.make("test"),
+      mode: "code",
+      agent: "code",
+      path: { cwd: "/tmp", root: "/tmp" },
+      cost: 0,
+      tokens: { input: 0, output: 0, reasoning: 0, cache: { read: 0, write: 0 } },
+      finish: "stop",
+    },
+    parts: [],
+  }
+}
+
+describe("session prompt queue", () => {
+  test("scopes queued turns without moving prior assistant history", async () => {
+    const sessionID = SessionID.make("session_scope")
+    const one = MessageID.make("message_01")
+    const ans = MessageID.make("message_02")
+    const two = MessageID.make("message_03")
+    const three = MessageID.make("message_04")
+    const messages = [
+      user(sessionID, one),
+      assistant(sessionID, ans, one),
+      user(sessionID, two),
+      user(sessionID, three),
+    ]
+
+    const ids = await Effect.runPromise(
+      SaeeolSessionPromptQueue.enqueue(
+        sessionID,
+        two,
+        Effect.sync(() => SaeeolSessionPromptQueue.scope(sessionID, messages).map((item) => item.info.id)),
+        Effect.succeed([]),
+      ),
+    )
+
+    expect(ids).toEqual([one, ans, two])
+  })
+
+  test("moves queued target to the end when prior-turn messages come after it", async () => {
+    // Regression: when a user queues a prompt while a turn is still running,
+    // the queued message's time_created falls before later assistant steps of
+    // that turn. Ordering by time_created alone would leave the queued prompt
+    // in the middle of the prior turn's messages, ending the next model request
+    // with an assistant message and tripping Anthropic's prefill rejection.
+    const sessionID = SessionID.make("session_queue_mid_turn")
+    const m1 = MessageID.make("message_10")
+    const a1 = MessageID.make("message_20")
+    const m2 = MessageID.make("message_30")
+    const a2step1 = MessageID.make("message_40")
+    const m3 = MessageID.make("message_50") // queued mid-turn
+    const a2step2 = MessageID.make("message_60")
+    const a2final = MessageID.make("message_70")
+    const messages = [
+      user(sessionID, m1),
+      assistant(sessionID, a1, m1),
+      user(sessionID, m2),
+      assistant(sessionID, a2step1, m2),
+      user(sessionID, m3),
+      assistant(sessionID, a2step2, m2),
+      assistant(sessionID, a2final, m2),
+    ]
+
+    const ids = await Effect.runPromise(
+      SaeeolSessionPromptQueue.enqueue(
+        sessionID,
+        m3,
+        Effect.sync(() => SaeeolSessionPromptQueue.scope(sessionID, messages).map((item) => item.info.id)),
+        Effect.succeed([]),
+      ),
+    )
+
+    expect(ids).toEqual([m1, a1, m2, a2step1, a2step2, a2final, m3])
+    expect(ids[ids.length - 1]).toBe(m3)
+  })
+
+  test("keeps the target turn's own assistant steps grouped at the end", async () => {
+    // After the first step of a queued turn has produced an assistant message,
+    // subsequent scope() calls should keep the target user together with its
+    // own turn's assistants (not interleaved with a prior turn's tail).
+    const sessionID = SessionID.make("session_queue_step_two")
+    const m1 = MessageID.make("message_01a")
+    const a1 = MessageID.make("message_02a")
+    const m2 = MessageID.make("message_03a") // queued mid-turn
+    const a1tail = MessageID.make("message_04a")
+    const a2step1 = MessageID.make("message_05a")
+    const messages = [
+      user(sessionID, m1),
+      assistant(sessionID, a1, m1),
+      user(sessionID, m2),
+      assistant(sessionID, a1tail, m1), // prior turn's tail was written after m2
+      assistant(sessionID, a2step1, m2),
+    ]
+
+    const ids = await Effect.runPromise(
+      SaeeolSessionPromptQueue.enqueue(
+        sessionID,
+        m2,
+        Effect.sync(() => SaeeolSessionPromptQueue.scope(sessionID, messages).map((item) => item.info.id)),
+        Effect.succeed([]),
+      ),
+    )
+
+    expect(ids).toEqual([m1, a1, a1tail, m2, a2step1])
+  })
+
+  test("retarget keeps older queued prompts hidden", async () => {
+    // Regression: retargeting used to move the visible-message boundary forward,
+    // which unhid any user prompts queued between the base and the injected
+    // follow-up. Exempt the follow-up without reopening the boundary.
+    const sessionID = SessionID.make("session_retarget_hide")
+    const base = MessageID.make("message_b1")
+    const ans = MessageID.make("message_b2")
+    const queued = MessageID.make("message_b3") // queued while base was running
+    const injected = MessageID.make("message_b4") // injected follow-up
+    const messages = [
+      user(sessionID, base),
+      assistant(sessionID, ans, base),
+      user(sessionID, queued),
+      user(sessionID, injected),
+    ]
+
+    const ids = await Effect.runPromise(
+      SaeeolSessionPromptQueue.enqueue(
+        sessionID,
+        base,
+        Effect.sync(() => {
+          SaeeolSessionPromptQueue.retarget(sessionID, injected)
+          return SaeeolSessionPromptQueue.scope(sessionID, messages).map((item) => item.info.id)
+        }),
+        Effect.succeed([]),
+      ),
+    )
+
+    expect(ids).not.toContain(queued)
+    expect(ids).toContain(injected)
+    expect(ids[ids.length - 1]).toBe(injected)
+  })
+
+  test("keeps auto-compaction markers created during a queued turn visible", async () => {
+    // Regression for ses_20d25cccbffeVAw7Y9XGfL9p5O: an overflow inside a queued
+    // turn creates an auto-compaction user message after the queued prompt. If
+    // scope() hides that marker, runLoop never processes the compaction task and
+    // instead retries the same oversized request until compaction is exhausted.
+    await using tmp = await tmpdir({ git: true })
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await Session.create({ title: "Queued compaction regression" })
+        const first = MessageID.ascending()
+        const ans = MessageID.ascending()
+        const queued = MessageID.ascending()
+
+        await Session.updateMessage(user(session.id, first).info)
+        await Session.updateMessage(assistant(session.id, ans, first).info)
+        await Session.updateMessage(user(session.id, queued).info)
+
+        const result = await Effect.runPromise(
+          SaeeolSessionPromptQueue.enqueue(
+            session.id,
+            queued,
+            Effect.promise(async () => {
+              await SessionCompaction.create({
+                sessionID: session.id,
+                agent: "code",
+                model: { providerID: ProviderID.make("test"), modelID: ModelID.make("model") },
+                auto: true,
+                overflow: true,
+              })
+              const messages = await Session.messages({ sessionID: session.id })
+              const compact = messages.find((msg) => msg.parts.some((part) => part.type === "compaction"))?.info.id
+              return { compact, ids: SaeeolSessionPromptQueue.scope(session.id, messages).map((item) => item.info.id) }
+            }),
+            Effect.succeed({ compact: undefined, ids: [] }),
+          ),
+        )
+
+        if (!result.compact) throw new Error("missing compaction marker")
+        expect(result.ids).toEqual([first, ans, queued, result.compact])
+        expect(result.ids[result.ids.length - 1]).toBe(result.compact)
+      },
+    })
+  })
+
+  test("hasFollowup reports true only for prompts enqueued after the active slot started", async () => {
+    const sessionID = SessionID.make("session_followup_semantics")
+    const observed: Array<{ where: string; value: boolean }> = []
+    const firstStarted = Promise.withResolvers<void>()
+    const firstReleased = Promise.withResolvers<void>()
+    const secondStarted = Promise.withResolvers<void>()
+    const secondReleased = Promise.withResolvers<void>()
+
+    const first = Effect.runPromise(
+      SaeeolSessionPromptQueue.enqueue(
+        sessionID,
+        MessageID.make("message_followup_1"),
+        Effect.gen(function* () {
+          observed.push({ where: "first:start", value: SaeeolSessionPromptQueue.hasFollowup(sessionID) })
+          firstStarted.resolve()
+          yield* Effect.promise(() => firstReleased.promise)
+          observed.push({ where: "first:end", value: SaeeolSessionPromptQueue.hasFollowup(sessionID) })
+          return "first"
+        }),
+        Effect.succeed("first-cancelled"),
+      ),
+    )
+
+    await firstStarted.promise
+    // msg1 is alone — nothing newer has arrived yet.
+    expect(observed[0]?.value).toBe(false)
+
+    const second = Effect.runPromise(
+      SaeeolSessionPromptQueue.enqueue(
+        sessionID,
+        MessageID.make("message_followup_2"),
+        Effect.gen(function* () {
+          observed.push({ where: "second:start", value: SaeeolSessionPromptQueue.hasFollowup(sessionID) })
+          secondStarted.resolve()
+          yield* Effect.promise(() => secondReleased.promise)
+          return "second"
+        }),
+        Effect.succeed("second-cancelled"),
+      ),
+    )
+
+    // Enqueueing msg2 while msg1 is still running must flip hasFollowup to true
+    // for msg1's running slot.
+    await new Promise((resolve) => setTimeout(resolve, 10))
+    expect(SaeeolSessionPromptQueue.hasFollowup(sessionID)).toBe(true)
+
+    const third = Effect.runPromise(
+      SaeeolSessionPromptQueue.enqueue(
+        sessionID,
+        MessageID.make("message_followup_3"),
+        Effect.sync(() => {
+          observed.push({ where: "third:start", value: SaeeolSessionPromptQueue.hasFollowup(sessionID) })
+          return "third"
+        }),
+        Effect.succeed("third-cancelled"),
+      ),
+    )
+
+    // Let msg1 finish.
+    firstReleased.resolve()
+    await first
+    await secondStarted.promise
+
+    // msg2 started after msg3 was enqueued, so hasFollowup should be false for
+    // msg2 — everything waiting is older than msg2's activeSince snapshot.
+    expect(SaeeolSessionPromptQueue.hasFollowup(sessionID)).toBe(false)
+    secondReleased.resolve()
+
+    expect(await second).toBe("second")
+    expect(await third).toBe("third")
+
+    const events = observed.map((item) => `${item.where}=${item.value}`)
+    expect(events).toEqual(["first:start=false", "first:end=true", "second:start=false", "third:start=false"])
+  })
+
+  test("processes queued prompts without aborting the in-flight stream", async () => {
+    const ready = Promise.withResolvers<void>()
+    const injected = Promise.withResolvers<void>()
+    const calls: number[] = []
+    const bodies: Array<Record<string, unknown>> = []
+    const server = Bun.serve({
+      port: 0,
+      async fetch(req) {
+        const url = new URL(req.url)
+        if (!url.pathname.endsWith("/chat/completions")) return new Response("not found", { status: 404 })
+
+        const body = (await req.json().catch(() => ({}))) as Record<string, unknown>
+        bodies.push(body)
+        calls.push(Date.now())
+        const stream =
+          calls.length === 1
+            ? reply({ text: "first reply", ready: ready.resolve })
+            : reply({ text: "second reply", ready: injected.resolve })
+        return new Response(stream, {
+          status: 200,
+          headers: { "Content-Type": "text/event-stream" },
+        })
+      },
+    })
+
+    try {
+      await using tmp = await tmpdir({
+        git: true,
+        init: async (dir) => {
+          await Bun.write(
+            path.join(dir, "saeeol.json"),
+            JSON.stringify({
+              $schema: "https://saeeol.ai/config.json",
+              enabled_providers: ["alibaba"],
+              provider: {
+                alibaba: {
+                  options: {
+                    apiKey: "test-key",
+                    baseURL: `${server.url.origin}/v1`,
+                  },
+                },
+              },
+              agent: {
+                code: {
+                  model: "alibaba/qwen-plus",
+                },
+              },
+            }),
+          )
+        },
+      })
+
+      await Instance.provide({
+        directory: tmp.path,
+        fn: async () => {
+          const session = await Session.create({ title: "Queued prompt regression" })
+          const first = SessionPrompt.prompt({
+            sessionID: session.id,
+            agent: "code",
+            parts: [{ type: "text", text: "first prompt" }],
+          })
+
+          await ready.promise
+
+          const second = SessionPrompt.prompt({
+            sessionID: session.id,
+            agent: "code",
+            parts: [{ type: "text", text: "second prompt" }],
+          })
+
+          const one = await first
+          await injected.promise
+          const two = await second
+
+          expect(calls).toHaveLength(2)
+
+          // The in-flight stream must complete; no aborted error on msg1's reply.
+          expect(one.info.role).toBe("assistant")
+          if (one.info.role === "assistant") expect(one.info.error).toBeUndefined()
+          expect(hasText(one, "first reply")).toBe(true)
+          expect(hasText(two, "second reply")).toBe(true)
+
+          const msgs = await Session.messages({ sessionID: session.id })
+          const users = msgs.filter((msg) => msg.info.role === "user")
+          const assistants = msgs.filter((msg) => msg.info.role === "assistant")
+          const prompts = users.flatMap((msg) =>
+            msg.parts.filter((part) => part.type === "text").map((part) => part.text),
+          )
+          const text = assistants.flatMap((msg) =>
+            msg.parts.filter((part) => part.type === "text").map((part) => part.text),
+          )
+          expect(users).toHaveLength(2)
+          expect(assistants).toHaveLength(2)
+          expect(prompts).toContain("first prompt")
+          expect(prompts).toContain("second prompt")
+          expect(text).toContain("first reply")
+          expect(text).toContain("second reply")
+
+          const firstUser = users.find((msg) => hasText(msg, "first prompt"))
+          const secondUser = users.find((msg) => hasText(msg, "second prompt"))
+          const firstReply = assistants.find((msg) => hasText(msg, "first reply"))
+          const secondReply = assistants.find((msg) => hasText(msg, "second reply"))
+          if (
+            firstUser?.info.role !== "user" ||
+            secondUser?.info.role !== "user" ||
+            firstReply?.info.role !== "assistant" ||
+            secondReply?.info.role !== "assistant"
+          ) {
+            throw new Error("missing expected messages")
+          }
+          expect(firstReply.info.parentID).toBe(firstUser.info.id)
+          expect(secondReply.info.parentID).toBe(secondUser.info.id)
+
+          // Regression for #9492: the second LLM request must end with the
+          // queued user prompt, not an assistant tail from the prior turn.
+          // Anthropic's API rejects requests whose final message is assistant
+          // (prefill), and scope() is supposed to partition the queued target
+          // turn to the end before the model request is built.
+          expect(bodies).toHaveLength(2)
+          const second2 = bodies[1]
+          expect(JSON.stringify(second2)).toContain("second prompt")
+          const tail = lastConversational(second2)
+          expect(tail?.role).toBe("user")
+          expect(JSON.stringify(tail?.content)).toContain("second prompt")
+        },
+      })
+    } finally {
+      server.stop(true)
+    }
+  })
+
+  test("cancel drops queued prompts and resets internal state", async () => {
+    const ready = Promise.withResolvers<void>()
+    const calls: number[] = []
+    const server = Bun.serve({
+      port: 0,
+      fetch(req) {
+        const url = new URL(req.url)
+        if (!url.pathname.endsWith("/chat/completions")) return new Response("not found", { status: 404 })
+
+        calls.push(Date.now())
+        const body = reply({ text: "first reply", ready: ready.resolve, wait: new Promise(() => {}) })
+        return new Response(body, {
+          status: 200,
+          headers: { "Content-Type": "text/event-stream" },
+        })
+      },
+    })
+
+    try {
+      await using tmp = await tmpdir({
+        git: true,
+        init: async (dir) => {
+          await Bun.write(
+            path.join(dir, "saeeol.json"),
+            JSON.stringify({
+              $schema: "https://saeeol.ai/config.json",
+              enabled_providers: ["alibaba"],
+              provider: {
+                alibaba: {
+                  options: { apiKey: "test-key", baseURL: `${server.url.origin}/v1` },
+                },
+              },
+              agent: { code: { model: "alibaba/qwen-plus" } },
+            }),
+          )
+        },
+      })
+
+      await Instance.provide({
+        directory: tmp.path,
+        fn: async () => {
+          const session = await Session.create({ title: "Queued cancel regression" })
+          const first = SessionPrompt.prompt({
+            sessionID: session.id,
+            agent: "code",
+            parts: [{ type: "text", text: "first prompt" }],
+          })
+          await ready.promise
+
+          const second = SessionPrompt.prompt({
+            sessionID: session.id,
+            agent: "code",
+            parts: [{ type: "text", text: "second prompt" }],
+          })
+          const third = SessionPrompt.prompt({
+            sessionID: session.id,
+            agent: "code",
+            parts: [{ type: "text", text: "third prompt" }],
+          })
+
+          // Let msg2/msg3's enqueue capture the current version before cancel bumps it.
+          await Bun.sleep(20)
+          expect(calls).toHaveLength(1)
+
+          await SessionPrompt.cancel(session.id)
+          await Promise.all([first, second, third])
+
+          // The queued prompts must never reach the LLM once cancel flushes the queue.
+          expect(calls).toHaveLength(1)
+          const msgs = await Session.messages({ sessionID: session.id })
+          const assistants = msgs.filter((msg) => msg.info.role === "assistant")
+          expect(assistants).toHaveLength(1)
+          expect(msgs.filter((msg) => msg.info.role === "user")).toHaveLength(3)
+
+          // Internal state should have no lingering tail/version/target entries after the last release.
+          const ids = await Effect.runPromise(
+            SaeeolSessionPromptQueue.enqueue(
+              session.id,
+              MessageID.make("message_probe"),
+              Effect.succeed(SaeeolSessionPromptQueue.scope(session.id, []).map((item) => item.info.id)),
+              Effect.succeed([]),
+            ),
+          )
+          expect(ids).toEqual([])
+          expect(SaeeolSessionPromptQueue.hasFollowup(session.id)).toBe(false)
+        },
+      })
+    } finally {
+      server.stop(true)
+    }
+  })
+
+  test("new prompt dismisses a pending suggestion", async () => {
+    const shown = Promise.withResolvers<void>()
+    const dismissed = Promise.withResolvers<void>()
+    await using tmp = await tmpdir({ git: true })
+
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await Session.create({ title: "Suggestion unblock regression" })
+        const offShown = Bus.subscribe(Suggestion.Event.Shown, (event) => {
+          if (event.properties.sessionID === session.id) shown.resolve()
+        })
+        const offDismissed = Bus.subscribe(Suggestion.Event.Dismissed, (event) => {
+          if (event.properties.sessionID === session.id) dismissed.resolve()
+        })
+
+        try {
+          const base = Suggestion.show({
+            sessionID: session.id,
+            text: "Run review?",
+            actions: [{ label: "Review", prompt: "/local-review-uncommitted" }],
+          }).catch((err) => {
+            if (err instanceof Suggestion.DismissedError) return "dismissed"
+            throw err
+          })
+
+          await shown.promise
+          await SessionPrompt.prompt({
+            sessionID: session.id,
+            agent: "code",
+            parts: [{ type: "text", text: "replacement prompt" }],
+            noReply: true,
+          })
+          await dismissed.promise
+
+          expect(await base).toBe("dismissed")
+          expect(await Suggestion.list()).toEqual([])
+        } finally {
+          offShown()
+          offDismissed()
+        }
+      },
+    })
+  })
+
+  test("new prompt dismisses a pending question", async () => {
+    const asked = Promise.withResolvers<void>()
+    const rejected = Promise.withResolvers<void>()
+    await using tmp = await tmpdir({ git: true })
+
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await Session.create({ title: "Question unblock regression" })
+        const offAsked = Bus.subscribe(Question.Event.Asked, (event) => {
+          if (event.properties.sessionID === session.id) asked.resolve()
+        })
+        const offRejected = Bus.subscribe(Question.Event.Rejected, (event) => {
+          if (event.properties.sessionID === session.id) rejected.resolve()
+        })
+
+        try {
+          const pending = Question.ask({
+            sessionID: session.id,
+            questions: [
+              {
+                header: "Continue?",
+                question: "Should I continue?",
+                options: [
+                  { label: "Yes", description: "Go ahead" },
+                  { label: "No", description: "Stop" },
+                ],
+              },
+            ],
+          }).catch((err) => {
+            if (err instanceof Question.RejectedError) return "rejected"
+            throw err
+          })
+
+          await asked.promise
+          await SessionPrompt.prompt({
+            sessionID: session.id,
+            agent: "code",
+            parts: [{ type: "text", text: "replacement prompt" }],
+            noReply: true,
+          })
+          await rejected.promise
+
+          expect(await pending).toBe("rejected")
+          expect(await Question.list()).toEqual([])
+        } finally {
+          offAsked()
+          offRejected()
+        }
+      },
+    })
+  })
+
+  test("auto-dismisses a suggestion shown after a queued prompt", async () => {
+    // Reverse ordering of the "new prompt dismisses a pending suggestion" test:
+    // queue the follow-up first, then open the blocker. Suggestion.show must see
+    // hasFollowup=true and reject synchronously, before any pending entry or
+    // Shown event is published.
+    await using tmp = await tmpdir({ git: true })
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const sessionID = SessionID.make("ses_auto_suggestion")
+        const started = Promise.withResolvers<void>()
+        const release = Promise.withResolvers<void>()
+
+        // Slot 1: active, activeSince snapshots latest=1.
+        const first = Effect.runPromise(
+          SaeeolSessionPromptQueue.enqueue(
+            sessionID,
+            MessageID.make("message_auto_sug_1"),
+            Effect.gen(function* () {
+              started.resolve()
+              yield* Effect.promise(() => release.promise)
+              return "first" as const
+            }),
+            Effect.succeed("first-cancelled" as const),
+          ),
+        )
+        await started.promise
+
+        // Slot 2: enqueued while slot 1 is active → latest=2 > activeSince=1.
+        const second = Effect.runPromise(
+          SaeeolSessionPromptQueue.enqueue(
+            sessionID,
+            MessageID.make("message_auto_sug_2"),
+            Effect.succeed("second" as const),
+            Effect.succeed("second-cancelled" as const),
+          ),
+        )
+        await Bun.sleep(10)
+        expect(SaeeolSessionPromptQueue.hasFollowup(sessionID)).toBe(true)
+
+        let shown = 0
+        const offShown = Bus.subscribe(Suggestion.Event.Shown, (event) => {
+          if (event.properties.sessionID === sessionID) shown++
+        })
+        try {
+          await expect(
+            Suggestion.show({
+              sessionID,
+              text: "Run review?",
+              actions: [{ label: "Review", prompt: "/local-review-uncommitted" }],
+            }),
+          ).rejects.toBeInstanceOf(Suggestion.DismissedError)
+        } finally {
+          offShown()
+        }
+        expect(shown).toBe(0)
+        expect(await Suggestion.list()).toEqual([])
+
+        release.resolve()
+        expect(await first).toBe("first")
+        expect(await second).toBe("second")
+      },
+    })
+  })
+
+  test("auto-dismisses a question shown after a queued prompt", async () => {
+    await using tmp = await tmpdir({ git: true })
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const sessionID = SessionID.make("ses_auto_question")
+        const started = Promise.withResolvers<void>()
+        const release = Promise.withResolvers<void>()
+
+        const first = Effect.runPromise(
+          SaeeolSessionPromptQueue.enqueue(
+            sessionID,
+            MessageID.make("message_auto_q_1"),
+            Effect.gen(function* () {
+              started.resolve()
+              yield* Effect.promise(() => release.promise)
+              return "first" as const
+            }),
+            Effect.succeed("first-cancelled" as const),
+          ),
+        )
+        await started.promise
+
+        const second = Effect.runPromise(
+          SaeeolSessionPromptQueue.enqueue(
+            sessionID,
+            MessageID.make("message_auto_q_2"),
+            Effect.succeed("second" as const),
+            Effect.succeed("second-cancelled" as const),
+          ),
+        )
+        await Bun.sleep(10)
+        expect(SaeeolSessionPromptQueue.hasFollowup(sessionID)).toBe(true)
+
+        let asked = 0
+        const offAsked = Bus.subscribe(Question.Event.Asked, (event) => {
+          if (event.properties.sessionID === sessionID) asked++
+        })
+        try {
+          await expect(
+            Question.ask({
+              sessionID,
+              questions: [
+                {
+                  header: "Continue?",
+                  question: "Should I continue?",
+                  options: [
+                    { label: "Yes", description: "Go ahead" },
+                    { label: "No", description: "Stop" },
+                  ],
+                },
+              ],
+            }),
+          ).rejects.toBeInstanceOf(Question.RejectedError)
+        } finally {
+          offAsked()
+        }
+        expect(asked).toBe(0)
+        expect(await Question.list()).toEqual([])
+
+        release.resolve()
+        expect(await first).toBe("first")
+        expect(await second).toBe("second")
+      },
+    })
+  })
+})