Keep created agents after first-prompt failures and bound fork history (#6379)

* fix(agents): preserve creation when the first prompt fails A rejected initial prompt must leave a created agent with a turn error instead of returning the user to a draft whose creation key is already committed. Preserve receipt replay without repeating the prompt, and bound fork history to 100,000 characters before it becomes an attachment. * test(creation): isolate user agent counters from background naming Workspace naming uses internal provider sessions. Counting and rejecting those sessions made creation assertions depend on installed provider discovery timing. Keep discovery within the test adapters and scope observations to user-created agents. * test(creation): retry temporary directory cleanup on Windows Windows CI completed the rejection assertions but failed teardown with EBUSY. Match the daemon test harness bounded removal retries so transient directory locks do not fail creation tests. * fix(agents): record rejected prompts before their errors Startup rejection recorded only the system error, leaving the client to append its local prompt after it. Record the rejected submission at the terminal failure boundary so stream and fetched history agree on prompt-before-error ordering. * fix(agents): require acceptance for delivery notifications Retain rejected human submissions before their errors without projecting rejected inter-agent prompts as completed deliveries. Reproduces and fixes the notification regression exposed by merge CI. * test(agents): retain rejected submissions across retries The existing browser retry test assumed terminally rejected prompts disappeared. Assert the intended prompt-before-error history and a separate accepted retry while preserving draft, attachment, and ambiguous-steering checks. * test(creation): allow integration latency in raw RPC responses The peer inherited a one-second polling timeout despite exercising Git and provider startup. Reproduce the timeout with delayed startup and wait within an explicit integration budget while retaining all response and identity assertions.
2026-10-14 13:22:37 +00:00 · 2026-10-08 22:25:57 +01:00
parent 8c096b5e33
commit 29db2b7980
16 changed files with 404 additions and 50 deletions
@@ -345,6 +345,10 @@ join active work or reuse committed stages. A persisted resource alone cannot pr
 that a provider accepted its initial prompt. Interrupted side effects with no conclusive
 receipt return an unknown outcome instead of being repeated.

+Agent registration commits creation. Rejecting the first prompt is a turn error on that
+agent; reporting creation failure leaves an orphaned agent and invites another Create.
+Receipt replay must not resend the prompt.
+
 `packages/client/src/creation/` owns capability selection and legacy orchestration.
 Callers always pass the initial prompt to agent creation. On an older host, the client
 adapts keyed creation to the legacy create/send sequence; it cannot continue that
@@ -284,7 +284,6 @@ async function expectRejectedSubmissionRestored(
  await expectComposerDraft(page, input.prompt);
  await expectComposerEditable(page);
  await expectAttachmentPill(page, "composer-image-attachment-pill");
-  await expect(page.getByTestId("user-message").filter({ hasText: input.prompt })).toHaveCount(0);
  if (input.preservesActiveTurn) {
    await expect(page.getByTestId("turn-working-indicator")).toBeVisible();
    return;
@@ -295,13 +294,21 @@ async function expectRejectedSubmissionRestored(

 async function retryRestoredSubmission(page: Page, prompt: string): Promise<void> {
  await page.getByRole("textbox", { name: "Message agent..." }).first().press("Enter");
-  const userMessage = page.getByTestId("user-message").filter({ hasText: prompt });
-  await expect(userMessage).toHaveCount(1);
+  const attempts = page.getByTestId("user-message").filter({ hasText: prompt });
+  await expect(attempts).toHaveCount(2);
+  const userMessage = attempts.last();
  await expect(userMessage).toHaveAttribute("aria-busy", "false", { timeout: 30_000 });
  await expect(userMessage.getByRole("button", { name: "Open image attachment" })).toBeVisible();
  await expect(page.getByTestId("composer-image-attachment-pill")).toHaveCount(0);
 }

+async function expectRejectedPromptBeforeError(page: Page, prompt: string, errorMessage: string) {
+  const messages = page.getByTestId("user-message").or(page.getByTestId("assistant-message"));
+  await expect(messages).toHaveCount(2);
+  await expect(messages.nth(0)).toContainText(prompt);
+  await expect(messages.nth(1)).toContainText(`[System Error] ${errorMessage}`);
+}
+
 async function configureSteerInSettings(page: Page): Promise<void> {
  const modifier = process.platform === "darwin" ? "Meta" : "Control";
  await page.keyboard.press(`${modifier}+Comma`);
@@ -1173,13 +1180,14 @@ test.describe("Agent message submission", () => {
    await toasts.expectNeverShown("agent-updating-toast");
  });

-  test("restores a rejected submission and accepts its retry", async ({
+  test("retains a rejected submission before its error and accepts its retry", async ({
    page,
    rejectionScenario,
  }) => {
    const prompt = "Restore this rejected submission.";
    await submitMessageThatWillBeRejected(page, prompt);
    await expectRejectedSubmissionRestored(page, { prompt, ...rejectionScenario });
+    await expectRejectedPromptBeforeError(page, prompt, rejectionScenario.errorMessage);
    await retryRestoredSubmission(page, prompt);
  });

@@ -1209,6 +1217,7 @@ test.describe("Agent message submission", () => {
        errorMessage: "Requested mock steer transport failure",
        preservesActiveTurn: true,
      });
+      await expect(page.getByTestId("user-message").filter({ hasText: prompt })).toHaveCount(0);

      expect(gate.getClientRequestCount("send_agent_message_request")).toBe(sendsBefore + 1);
      expect(gate.getClientRequestCount("cancel_agent_request")).toBe(cancelsBefore);
@@ -128,3 +128,46 @@ for (const scenario of [
    await creation.expectAgentCount(1);
  });
 }
+
+for (const entry of [
+  {
+    name: "agent draft",
+    open: (creation: Awaited<ReturnType<typeof createCreationScenario>>) =>
+      creation.openAgentDraft(),
+    button: "Send message",
+    settled: async (_creation: Awaited<ReturnType<typeof createCreationScenario>>) => {},
+  },
+  {
+    name: "new workspace",
+    open: (creation: Awaited<ReturnType<typeof createCreationScenario>>) =>
+      creation.openWorkspaceForm("local"),
+    button: "Create",
+    settled: (creation: Awaited<ReturnType<typeof createCreationScenario>>) =>
+      creation.expectOneCreatedWorkspace(),
+  },
+]) {
+  test(`${entry.name} becomes an agent tab when the first prompt is rejected`, async ({
+    creation,
+    promptRejection,
+    page,
+  }, testInfo) => {
+    void promptRejection;
+    const prompt =
+      "Continue the attached conversation.\n" + "Earlier conversation context.\n".repeat(100);
+    await entry.open(creation);
+    await creation.submitPrompt(prompt, entry.button);
+    await entry.settled(creation);
+    await creation.expectCreatedAgentError();
+    await creation.expectPromptBeforeError(prompt);
+    await page.screenshot({ path: testInfo.outputPath("created-agent-prompt-error.png") });
+    await creation.reloadAgent();
+    await creation.expectPromptBeforeError(prompt);
+    await creation.submitPrompt("emit 1 coalesced agent stream updates for a corrected prompt.");
+    await creation.expectPromptVisible(
+      "emit 1 coalesced agent stream updates for a corrected prompt.",
+    );
+    await creation.expectAssistantReply();
+    await creation.expectAgentCount(1);
+    await expect(page.getByText("agent_request_key_conflict", { exact: true })).toHaveCount(0);
+  });
+}
@@ -13,6 +13,7 @@ import { delayBrowserAgentCreatedStatus } from "./helpers/new-workspace";

 export const test = base.extend<{
  startup: { release(): Promise<void>; fail(): Promise<void> };
+  promptRejection: void;
  creation: Awaited<ReturnType<typeof createCreationScenario>>;
  promptRetry: ReturnType<typeof createPromptRetryScenario>;
  agentRetries: Awaited<ReturnType<typeof retryNextAgentCreation>>;
@@ -65,6 +66,37 @@ export default function contribute(server) {
      await client.close();
    }
  },
+  promptRejection: async ({ page }, provide) => {
+    void page;
+    const client = await connectNewWorkspaceDaemonClient();
+    const plugin = await copyPluginExample("lifecycle-logger");
+    const previous = await client.getDaemonConfig();
+    try {
+      await writeFile(
+        path.join(plugin.directory, "index.server.ts"),
+        `
+export default function contribute(server) {
+  return server.before("agent.create", ({ request }) => ({
+    ...request,
+    config: {
+      ...request.config,
+      model: "e2e-fast-stream",
+      featureValues: { ...request.config.featureValues, mockPromptRejections: 1 },
+    },
+  }));
+}
+`,
+      );
+      await client.patchDaemonConfig({ pluginsEnabled: true });
+      await client.installDirectoryPlugin(plugin.directory);
+      await provide();
+    } finally {
+      await client.removePlugin("lifecycle-logger");
+      await client.patchDaemonConfig({ pluginsEnabled: previous.config.pluginsEnabled });
+      await plugin.cleanup();
+      await client.close();
+    }
+  },
  creation: async ({ page }, provide) => {
    const creation = await createCreationScenario(page);
    try {
@@ -182,6 +182,47 @@ export async function createCreationScenario(page: Page) {
      await this.expectWorkspaceReadyBeforeAgentCompletion();
      expect((await project.client.fetchAgents()).entries).toHaveLength(0);
    },
+    async expectCreatedAgentError() {
+      await requests.settled();
+      await expect(
+        page
+          .getByTestId("assistant-message")
+          .filter({ hasText: "[System Error] Requested mock prompt rejection", visible: true }),
+      ).toBeVisible();
+      await expect(
+        page
+          .locator('[data-testid^="workspace-tab-agent_"][aria-selected="true"]')
+          .filter({ visible: true }),
+      ).toBeVisible();
+      await expect(
+        page.locator('[data-testid^="workspace-tab-draft_"]').filter({ visible: true }),
+      ).toHaveCount(0);
+      await expect(page.getByRole("textbox", { name: "Message agent..." }).first()).toBeEditable();
+      await expect(page.getByTestId("turn-working-indicator")).toHaveCount(0);
+    },
+    async expectPromptBeforeError(prompt: string) {
+      const messages = page
+        .getByTestId("user-message")
+        .or(page.getByTestId("assistant-message"))
+        .filter({ visible: true });
+      await expect(messages).toHaveCount(2);
+      await expect(messages.nth(0)).toContainText(prompt);
+      await expect(messages.nth(1)).toContainText("[System Error]");
+      await expect(messages.nth(1)).toBeInViewport();
+    },
+    async reloadAgent() {
+      await page.reload();
+      await expect(page.getByText("Updating messages", { exact: true })).toHaveCount(0);
+      await this.expectCreatedAgentError();
+    },
+    async expectAssistantReply() {
+      await expect(
+        page
+          .getByTestId("assistant-message")
+          .filter({ visible: true, hasNotText: "[System Error]" })
+          .first(),
+      ).toBeVisible();
+    },
    async expectStartupFailure() {
      await expect(page.getByText(/Creation startup failed for test/).first()).toBeVisible();
    },
@@ -89,7 +89,7 @@
    "speech:transcribe:local": "tsx scripts/transcribe-local-wav.ts",
    "test": "npm run test:unit && npm run test:integration",
    "test:unit": "vitest run --fileParallelism --exclude \"**/*.e2e.test.ts\"",
-    "test:integration": "vitest run --maxWorkers=1 src/server/daemon-e2e/terminal-workspace-sdk.e2e.test.ts src/server/daemon-e2e/models.e2e.test.ts src/server/daemon-e2e/live-preferences.e2e.test.ts src/server/agent/model-catalog.e2e.test.ts src/server/plugins/plugin-paseo-api.e2e.test.ts && vitest run --maxWorkers=1 src/server/session.create-agent-worktree-autoarchive.e2e.test.ts -t \"creates a worktree and auto-archives both\" && vitest run --maxWorkers=1 src/server/agent/providers/opencode-bridge.local.e2e.test.ts -t \"versioned runtime\"",
+    "test:integration": "vitest run --maxWorkers=1 src/server/creation/creation.e2e.test.ts src/server/daemon-e2e/terminal-workspace-sdk.e2e.test.ts src/server/daemon-e2e/models.e2e.test.ts src/server/daemon-e2e/live-preferences.e2e.test.ts src/server/agent/model-catalog.e2e.test.ts src/server/plugins/plugin-paseo-api.e2e.test.ts && vitest run --maxWorkers=1 src/server/session.create-agent-worktree-autoarchive.e2e.test.ts -t \"creates a worktree and auto-archives both\" && vitest run --maxWorkers=1 src/server/agent/providers/opencode-bridge.local.e2e.test.ts -t \"versioned runtime\"",
    "test:hub-cli-contract": "vitest run --maxWorkers=1 src/server/hub/hub-cli-contract.test.ts",
    "test:integration:all": "npm run test:e2e",
    "test:integration:real": "vitest run real.e2e.test.ts",
@@ -438,3 +438,49 @@ second line'`,
    ).toThrow("Selected assistant message is no longer available.");
  });
 });
+
+it("bounds fork history to 100,000 characters while preserving the newest messages", () => {
+  const result = buildAgentForkContextAttachment({
+    rows: [
+      row(1, { type: "user_message", text: "old-message " + "x".repeat(1_100_000) }),
+      row(2, { type: "assistant_message", messageId: "checkpoint", text: "Latest answer." }),
+      row(3, { type: "user_message", text: "Beyond the selected checkpoint." }),
+    ],
+    boundaryMessageId: "checkpoint",
+  });
+  expect(result.attachment.text.length).toBeLessThanOrEqual(100_000);
+  expect(result.attachment.text).toContain("[Earlier chat history omitted]");
+  expect(result.attachment.text).toContain("[Assistant] Latest answer.");
+  expect(result.attachment.text).not.toContain("old-message");
+  expect(result.attachment.text).not.toContain("Beyond the selected checkpoint.");
+  expect(result.attachment.text).toMatch(/^<chat-history-summary>\n/);
+  expect(result.attachment.text).toMatch(/\n<\/chat-history-summary>$/);
+});
+
+it("bounds one oversized fork message and metadata without losing its beginning and end", () => {
+  const result = buildAgentForkContextAttachment({
+    rows: [
+      row(1, {
+        type: "assistant_message",
+        text: "Beginning. " + "x".repeat(1_100_000) + " The end.",
+      }),
+    ],
+    agentTitle: "T".repeat(200_000),
+    cwd: "/" + "p".repeat(200_000),
+  });
+  expect(result.attachment.text.length).toBe(100_000);
+  expect(result.attachment.text).toContain("[Assistant] Beginning.");
+  expect(result.attachment.text).toContain("[Message truncated]");
+  expect(result.attachment.text).toContain("The end.");
+  expect(result.attachment.text).toMatch(/\n<\/chat-history-summary>$/);
+});
+
+it("preserves every message in fork history below the character budget", () => {
+  const text = "Complete conversation " + "x".repeat(90_000);
+  const result = buildAgentForkContextAttachment({
+    rows: [row(1, { type: "user_message", text })],
+  });
+  expect(result.attachment.text).toBe(
+    `<chat-history-summary>\nChat history from a previous Paseo agent.\n\n[User] ${text}\n</chat-history-summary>`,
+  );
+});
@@ -8,6 +8,9 @@ import { projectTimelineRows } from "./timeline-projection.js";
 const DEFAULT_MAX_ITEMS = 0;
 const MAX_TOOL_INPUT_CHARS = 400;
 const MAX_TOOL_SUMMARY_CHARS = 200;
+const MAX_FORK_CONTEXT_CHARS = 100_000;
+const MAX_FORK_METADATA_CHARS = 1_000;
+const FORK_HISTORY_OMITTED = "[Earlier chat history omitted]\n";

 interface ActivityCuratorOptions {
  maxItems?: number;
@@ -276,7 +279,7 @@ function selectForkContextRows(input: {

 function trimContextMetadata(value: string | null | undefined): string | null {
  const trimmed = value?.trim();
-  return trimmed ? trimmed : null;
+  return trimmed ? trimmed.slice(0, MAX_FORK_METADATA_CHARS) : null;
 }

 function buildForkContextText(input: {
@@ -296,6 +299,30 @@ function buildForkContextText(input: {
  return `<chat-history-summary>\n${header.join("\n")}\n\n${input.body}\n</chat-history-summary>`;
 }

+function fitForkContextEntries(entries: ActivityEntry[], budget: number): string {
+  const length = entries.reduce((total, entry) => total + entry.text.length + 1, -1);
+  if (length <= budget) return entries.map((entry) => entry.text).join("\n");
+
+  const selected: string[] = [];
+  let remaining = budget - FORK_HISTORY_OMITTED.length;
+  for (let index = entries.length - 1; index >= 0; index--) {
+    const text = entries[index]!.text;
+    const size = text.length + (selected.length > 0 ? 1 : 0);
+    if (size > remaining) {
+      if (selected.length === 0) {
+        const marker = "\n[Message truncated]\n";
+        const available = remaining - marker.length;
+        const head = Math.ceil(available / 2);
+        selected.push(text.slice(0, head) + marker + text.slice(-(available - head)));
+      }
+      break;
+    }
+    selected.push(text);
+    remaining -= size;
+  }
+  return FORK_HISTORY_OMITTED + selected.toReversed().join("\n");
+}
+
 export function buildAgentForkContextAttachment(input: {
  rows: readonly AgentTimelineRow[];
  cursorBoundary?: ForkCursorBoundary | null;
@@ -319,10 +346,10 @@ export function buildAgentForkContextAttachment(input: {
    includeKinds: ["user_message", "assistant_message", "tool_call"],
    includeExternalToolInput: false,
  });
+  const metadata = { agentTitle: input.agentTitle, cwd: input.cwd };
+  const budget = MAX_FORK_CONTEXT_CHARS - buildForkContextText({ body: "", ...metadata }).length;
  const body =
-    entries.length > 0
-      ? entries.map((entry) => entry.text).join("\n")
-      : "No chat history to display.";
+    entries.length > 0 ? fitForkContextEntries(entries, budget) : "No chat history to display.";
  return {
    attachment: {
      type: "text",
@@ -3954,12 +3954,23 @@ test("a prompt after provider replacement reopens the stale session", async () =
      clients: { [provider]: replacement },
    });

-    const dispatch = await startAgentRun(manager, created.id, "continue after reload", logger);
+    const dispatch = await startAgentRun(manager, created.id, "continue after reload", logger, {
+      runOptions: { clientMessageId: "stale-session-retry" },
+    });
    expect(dispatch.disposition).toBe("turn_started");
    await replacement.waitForRetryStart();
    const result = await manager.waitForAgentEvent(created.id);
    expect(result.status).toBe("idle");
    expect(replacement.resumeOverrides).toHaveLength(1);
+    const timeline = manager.getTimeline(created.id);
+    expect(timeline.filter((item) => item.type === "user_message")).toMatchObject([
+      { text: "continue after reload", clientMessageId: "stale-session-retry" },
+    ]);
+    expect(
+      timeline.some(
+        (item) => item.type === "assistant_message" && item.text.includes("[System Error]"),
+      ),
+    ).toBe(false);
  } finally {
    await manager.closeAgent(created.id).catch(() => undefined);
    await manager.flush().catch(() => undefined);
@@ -8141,9 +8152,20 @@ test("streamAgent clears pending run when startTurn fails before a turn id exist
    { workspaceId: undefined },
  );

-  await expect(manager.runAgent(agent.id, "fail before turn id")).rejects.toThrow(
-    "Invalid request: missing field `text`",
-  );
+  await expect(
+    manager.runAgent(agent.id, "fail before turn id", {
+      clientMessageId: "rejected-prompt",
+    }),
+  ).rejects.toThrow("Invalid request: missing field `text`");
+  expect(manager.getTimeline(agent.id)).toEqual([
+    {
+      type: "user_message",
+      text: "fail before turn id",
+      clientMessageId: "rejected-prompt",
+      messageId: "rejected-prompt",
+    },
+    { type: "assistant_message", text: "[System Error] Invalid request: missing field `text`" },
+  ]);

  await expect(manager.runAgent(agent.id, "second turn")).resolves.toEqual(
    expect.objectContaining({
@@ -2360,7 +2360,7 @@ export class AgentManager {
      return false;
    }
    if (options?.clientMessageId) {
-      this.recordSubmittedPrompt(agent, prompt, options.clientMessageId);
+      this.recordSubmittedPrompt(agent, prompt, options.clientMessageId, "accepted");
      this.emitState(agent);
    }
    const dispatch = (event: AgentStreamEvent): void => {
@@ -2456,6 +2456,12 @@ export class AgentManager {
      agent.pendingReplacement = false;
      const errorMsg = error instanceof Error ? error.message : "Failed to start turn";
      pendingRun.start = { status: "failed", error: errorMsg };
+      // A terminal rejection belongs after the submitted prompt even though no provider turn exists.
+      if (options?.clientMessageId) {
+        this.recordSubmittedPrompt(agent, prompt, options.clientMessageId, "rejected", {
+          messageId: options.clientMessageId,
+        });
+      }
      await this.handleStreamEvent(agent, {
        type: "turn_failed",
        provider: agent.provider,
@@ -2545,7 +2551,7 @@ export class AgentManager {
          )
        : undefined;
      if (options?.clientMessageId) {
-        this.recordSubmittedPrompt(agent, prompt, options.clientMessageId, {
+        this.recordSubmittedPrompt(agent, prompt, options.clientMessageId, "accepted", {
          messageId: options.clientMessageId,
          turnId,
          providerMessageId:
@@ -2851,7 +2857,7 @@ export class AgentManager {
    if (!clientMessageId) {
      return;
    }
-    this.recordSubmittedPrompt(agent, prompt, clientMessageId, {
+    this.recordSubmittedPrompt(agent, prompt, clientMessageId, "accepted", {
      messageId: clientMessageId,
      turnId: expectedTurnId,
    });
@@ -4696,6 +4702,7 @@ export class AgentManager {
    agent: ActiveManagedAgent,
    prompt: AgentPromptInput,
    clientMessageId: string,
+    outcome: "accepted" | "rejected",
    options?: { messageId?: string; providerMessageId?: string; turnId?: string },
  ): void {
    const item = projectAgentMessage({
@@ -4704,7 +4711,10 @@ export class AgentManager {
      clientMessageId,
      ...(options?.messageId ? { messageId: options.messageId } : {}),
    });
-    if (item) this.recordSubmittedPromptItem(agent, item, options);
+    if (!item) return;
+    // Human attempts stay in history on rejection; delivery notifications require acceptance.
+    if (outcome === "rejected" && item.type !== "user_message") return;
+    this.recordSubmittedPromptItem(agent, item, options);
  }

  private recordSubmittedPromptItem(
@@ -300,7 +300,8 @@ async function resolveSessionCreateAgent(
    runOptions,
    setupContinuation,
    background: true,
-    promptFailure: "throw",
+    // Registration commits creation. First-turn failures belong to the created agent.
+    promptFailure: "return-error",
    promptLogger: dependencies.logger.child({
      clientMessageId: resolveClientMessageId(input.clientMessageId),
    }),
@@ -481,7 +482,11 @@ async function sendInitialPrompt(
      throw error;
    }
    if (resolved.promptFailure === "return-error") {
-      return { started: false, liveSnapshot: snapshot, error };
+      return {
+        started: false,
+        liveSnapshot: dependencies.agentManager.getAgent(snapshot.id) ?? snapshot,
+        error,
+      };
    }
    dependencies.logger.error({ err: error, agentId: snapshot.id }, "Failed to run initial prompt");
    return { started: false, liveSnapshot: snapshot };
@@ -1,5 +1,6 @@
 import { execFileSync } from "node:child_process";
 import type { z } from "zod";
+import { BUILTIN_PROVIDER_IDS } from "@getpaseo/protocol/provider-manifest";
 import { WebSocket } from "ws";
 import { SessionInboundMessageSchema, WSOutboundMessageSchema } from "@getpaseo/protocol/messages";
 import { randomUUID } from "node:crypto";
@@ -9,10 +10,23 @@ import { join } from "node:path";
 import { expect, test, vi } from "vitest";
 import pino from "pino";
 import type { CreationSnapshot, SessionOutboundMessage } from "@getpaseo/protocol/messages";
+import type { AgentTimelineItem } from "../agent/agent-sdk-types.js";
 import { DaemonClient } from "../test-utils/daemon-client.js";
 import { createTestAgentClients } from "../test-utils/fake-agent-client.js";
 import { createTestPaseoDaemon } from "../test-utils/paseo-daemon.js";

+function createCreationDaemon(options: Parameters<typeof createTestPaseoDaemon>[0] = {}) {
+  const agentClients = options.agentClients ?? createTestAgentClients();
+  return createTestPaseoDaemon({
+    ...options,
+    agentClients,
+    // Installed provider CLIs must not change when background naming starts.
+    providerOverrides: Object.fromEntries(
+      BUILTIN_PROVIDER_IDS.map((id) => [id, { enabled: id in agentClients }]),
+    ),
+  });
+}
+
 function deferred<T>() {
  let resolve!: (value: T) => void;
  const promise = new Promise<T>((done) => {
@@ -29,7 +43,7 @@ test("creation progresses before agent readiness and continues after the disconn
  const directory = await mkdtemp(join(tmpdir(), "creation-wire-"));
  let agents = 0;
  let prompts = 0;
-  const daemon = await createTestPaseoDaemon({
+  const daemon = await createCreationDaemon({
    logger: pino(
      { level: "trace" },
      {
@@ -41,11 +55,13 @@ test("creation progresses before agent readiness and continues after the disconn
      },
    ),
    agentClients: createTestAgentClients({
-      beforeCreateSession: async () => {
+      beforeCreateSession: async (config) => {
+        if (config.internal) return;
        agents++;
        await provider.promise;
      },
-      onStartTurn: () => {
+      onStartTurn: (_prompt, config) => {
+        if (config.internal) return;
        prompts++;
      },
    }),
@@ -122,11 +138,12 @@ test("creation progresses before agent readiness and continues after the disconn
    await client.close();
    await observer.close();
    await daemon.close();
-    await rm(directory, { recursive: true, force: true });
+    await rm(directory, { recursive: true, force: true, maxRetries: 3, retryDelay: 100 });
  }
 }, 60000);

 async function connectCreationPeer(port: number) {
+  const timeout = 30_000;
  const socket = new WebSocket(`ws://127.0.0.1:${port}/ws`);
  const frames: SessionOutboundMessage[] = [];
  socket.on("message", (data) => {
@@ -147,7 +164,9 @@ async function connectCreationPeer(port: number) {
    }),
  );
  await expect
-    .poll(() => frames.some((m) => m.type === "status" && m.payload.status === "server_info"))
+    .poll(() => frames.some((m) => m.type === "status" && m.payload.status === "server_info"), {
+      timeout,
+    })
    .toBe(true);
  return {
    close: () => socket.close(),
@@ -160,7 +179,7 @@ async function connectCreationPeer(port: number) {
          (m) =>
            "payload" in m && "requestId" in m.payload && m.payload.requestId === message.requestId,
        );
-      await expect.poll(response).toBeDefined();
+      await expect.poll(response, { timeout }).toBeDefined();
      return response()!;
    },
  };
@@ -170,7 +189,7 @@ test.each([false, true])(
  "workspace identity does not depend on subscribing (first subscribe=%s)",
  async (subscribe) => {
    const directory = await mkdtemp(join(tmpdir(), "creation-identity-"));
-    const daemon = await createTestPaseoDaemon();
+    const daemon = await createCreationDaemon();
    const peer = await connectCreationPeer(daemon.port);
    try {
      const request = {
@@ -188,7 +207,7 @@ test.each([false, true])(
    } finally {
      peer.close();
      await daemon.close();
-      await rm(directory, { recursive: true, force: true });
+      await rm(directory, { recursive: true, force: true, maxRetries: 3, retryDelay: 100 });
    }
  },
  60000,
@@ -199,10 +218,13 @@ test.each(["create_agent_request", "agent.create.request"] as const)(
  async (type) => {
    const directory = await mkdtemp(join(tmpdir(), "creation-agent-identity-"));
    let creations = 0;
-    const daemon = await createTestPaseoDaemon({
+    const daemon = await createCreationDaemon({
      agentClients: createTestAgentClients({
-        beforeCreateSession: async () => {
+        beforeCreateSession: async (config) => {
+          if (config.internal) return;
          creations++;
+          // A real provider or Git checkout may take longer than Vitest's default 1s poll budget.
+          await new Promise((resolve) => setTimeout(resolve, 1_200));
        },
      }),
    });
@@ -241,7 +263,7 @@ test.each(["create_agent_request", "agent.create.request"] as const)(
    } finally {
      peer.close();
      await daemon.close();
-      await rm(directory, { recursive: true, force: true });
+      await rm(directory, { recursive: true, force: true, maxRetries: 3, retryDelay: 100 });
    }
  },
  60000,
@@ -264,7 +286,7 @@ test("legacy keyed creation preserves checkout error codes", async () => {
    ],
    { cwd: directory, stdio: "pipe" },
  );
-  const daemon = await createTestPaseoDaemon({ agentClients: createTestAgentClients() });
+  const daemon = await createCreationDaemon({ agentClients: createTestAgentClients() });
  const peer = await connectCreationPeer(daemon.port);
  try {
    const result = await peer.request({
@@ -281,6 +303,95 @@ test("legacy keyed creation preserves checkout error codes", async () => {
  } finally {
    peer.close();
    await daemon.close();
-    await rm(directory, { recursive: true, force: true });
+    await rm(directory, { recursive: true, force: true, maxRetries: 3, retryDelay: 100 });
  }
 }, 60000);
+
+test.each(["agent", "workspace"] as const)(
+  "%s creation succeeds when the provider rejects its initial prompt",
+  async (kind) => {
+    const directory = await mkdtemp(join(tmpdir(), "creation-prompt-rejected-"));
+    const rejection = "Input exceeds the maximum length of 1048576 characters.";
+    let sessions = 0;
+    let prompts = 0;
+    const daemon = await createCreationDaemon({
+      agentClients: createTestAgentClients({
+        beforeCreateSession: async (config) => {
+          if (config.internal) return;
+          sessions++;
+        },
+        onStartTurn: (_prompt, config) => {
+          if (config.internal) return;
+          prompts++;
+          throw new Error(rejection);
+        },
+      }),
+    });
+    const client = new DaemonClient({
+      url: `ws://127.0.0.1:${daemon.port}/ws`,
+      appVersion: "0.11.1",
+    });
+    const snapshots: CreationSnapshot[] = [];
+    const agent = {
+      config: { provider: "codex", cwd: directory },
+      initialPrompt: "Continue the attached conversation.",
+      clientMessageId: "draft:initial-message",
+    };
+    const streamed: AgentTimelineItem[] = [];
+    const unsubscribe = daemon.daemon.agentManager.subscribe((event) => {
+      if (event.type === "agent_stream" && event.event.type === "timeline") {
+        streamed.push(event.event.item);
+      }
+    });
+    try {
+      await client.connect();
+      const submit = async () => {
+        const common = {
+          idempotencyKey: "rejected-first-prompt",
+          onEvent: (snapshot: CreationSnapshot) => snapshots.push(snapshot),
+        };
+        if (kind === "agent") return client.createAgent({ ...common, ...agent });
+        const result = await client.createWorkspace({
+          ...common,
+          source: { kind: "directory", path: directory },
+          agent,
+        });
+        expect(result.error).toBeNull();
+        return result.agent!;
+      };
+      const created = await submit();
+      expect(created).toMatchObject({ status: "error", lastError: rejection });
+      expect(snapshots.map((snapshot) => snapshot.phase)).not.toContain("prompt_started");
+      expect(snapshots.at(-1)).toMatchObject({
+        phase: "completed",
+        error: null,
+        agent: { id: created.id },
+      });
+      const expectedTimeline = [
+        {
+          type: "user_message",
+          text: agent.initialPrompt,
+          clientMessageId: agent.clientMessageId,
+          messageId: agent.clientMessageId,
+        },
+        { type: "assistant_message", text: `[System Error] ${rejection}` },
+      ];
+      expect(
+        (await client.fetchAgentTimeline(created.id)).entries.map((entry) => entry.item),
+      ).toEqual(expectedTimeline);
+      expect(streamed).toEqual(expectedTimeline);
+      expect((await submit()).id).toBe(created.id);
+      expect(
+        (await client.fetchAgentTimeline(created.id)).entries.map((entry) => entry.item),
+      ).toEqual(expectedTimeline);
+      expect((await client.fetchAgents()).entries).toHaveLength(1);
+      expect({ sessions, prompts }).toEqual({ sessions: 1, prompts: 1 });
+    } finally {
+      unsubscribe();
+      await client.close();
+      await daemon.close();
+      await rm(directory, { recursive: true, force: true, maxRetries: 3, retryDelay: 100 });
+    }
+  },
+  60000,
+);
@@ -93,7 +93,7 @@ async function fixture() {
      calls.push("agent");
      await onReady(agent);
      calls.push("prompt");
-      return agent;
+      return { agent, initialPromptStarted: true };
    },
  };
  const updates: CreationSnapshot[] = [];
@@ -129,7 +129,7 @@ test("a failed agent startup retries only that stage with the reserved IDs", asy
    expect(readyWorkspace).toEqual(workspace);
    if (attempts === 1) throw new Error("provider unavailable");
    await onReady(agent);
-    return agent;
+    return { agent, initialPromptStarted: true };
  };
  expect(await f.service.create(f.input)).toMatchObject({
    phase: "failed",
@@ -210,7 +210,7 @@ test("omitted resource IDs are generated once and preserved on replay", async ()
  f.input.createAgent = async (id, createdWorkspace, onReady) => {
    const created = { ...agent, id, workspaceId: createdWorkspace!.id };
    await onReady(created);
-    return created;
+    return { agent: created, initialPromptStarted: true };
  };
  const result = await f.service.create(f.input);
  expect(result.workspaceId).toMatch(/^wks_[a-f0-9]{16}$/);
@@ -231,7 +231,7 @@ test("restart after a committed agent without a prompt completes without recreat
    await onReady(agent);
    // Capture the durable state at the exact milestone a process could stop after.
    await cp(f.directory, recoveredDirectory, { recursive: true });
-    return agent;
+    return { agent, initialPromptStarted: false };
  };
  await f.service.create(f.input);
  const recovered = await new CreationService(recoveredDirectory, silentLogger).create(f.input);
@@ -13,6 +13,10 @@ import { writeFileAtomic } from "../atomic-file.js";
 import { generateWorkspaceId } from "../workspace-registry-model.js";

 type Observer = (snapshot: CreationSnapshot) => void;
+export interface CreatedAgent {
+  agent: AgentSnapshotPayload;
+  initialPromptStarted: boolean;
+}
 interface CreationRequest {
  key: string;
  request: { [key: string]: unknown };
@@ -28,7 +32,7 @@ interface CreationRequest {
    agentId: string,
    workspace: WorkspaceDescriptorPayload | undefined,
    onReady: (agent: AgentSnapshotPayload) => Promise<void>,
-  ) => Promise<AgentSnapshotPayload>;
+  ) => Promise<CreatedAgent>;
 }
 export type CreationInput = CreationRequest &
  (
@@ -167,7 +171,7 @@ export class CreationService {
      if (input.hasAgent && input.createAgent && !record.snapshot.agent) {
        record.inFlight = "agent";
        await this.write(identity, record);
-        const agent = await input.createAgent(
+        const { agent, initialPromptStarted } = await input.createAgent(
          record.snapshot.agentId!,
          record.snapshot.workspace,
          async (readyAgent) => {
@@ -180,7 +184,7 @@ export class CreationService {
          },
        );
        record.inFlight = null;
-        if (input.hasPrompt)
+        if (initialPromptStarted)
          await this.publish(identity, record, { phase: "prompt_started", agent });
        else record.snapshot = { ...record.snapshot, agent };
      }
@@ -8,7 +8,7 @@ import type {
 } from "@getpaseo/protocol/messages";
 import { relative } from "node:path";
 import { isAbsolute } from "node:path";
-import { CreationService } from "./creation/index.js";
+import { CreationService, type CreatedAgent } from "./creation/index.js";
 import type { CreationSnapshot, AgentCreateRequest } from "@getpaseo/protocol/messages";
 import type { MessageReceipts } from "./message-receipts/index.js";
 import equal from "fast-deep-equal";
@@ -4173,7 +4173,7 @@ export class Session {
        if (!record) throw new Error("Previously created agent no longer exists");
        agent = this.buildStoredAgentPayload(record);
      } else {
-        agent = await this.createSessionAgent(msg);
+        ({ agent } = await this.createSessionAgent(msg));
      }
      this.emit({
        type: "status",
@@ -4212,7 +4212,7 @@ export class Session {
    msg: CreateAgentRequestMessage,
    agentId?: string,
    onAgentReady?: (agent: AgentSnapshotPayload) => Promise<void>,
-  ): Promise<AgentSnapshotPayload> {
+  ): Promise<CreatedAgent> {
    const {
      config,
      worktreeName,
@@ -4270,7 +4270,7 @@ export class Session {
        throw new Error(`Working directory does not exist or is not a directory: ${resolvedCwd}`);
      }

-      const { snapshot, liveSnapshot } = await createAgentCommand(
+      const { snapshot, liveSnapshot, initialPromptStarted } = await createAgentCommand(
        {
          agentManager: this.agentManager,
          agentStorage: this.agentStorage,
@@ -4327,7 +4327,7 @@ export class Session {
        { agentId: snapshot.id, provider: snapshot.provider },
        "Created agent",
      );
-      return this.buildAgentPayload(liveSnapshot);
+      return { agent: await this.buildAgentPayload(liveSnapshot), initialPromptStarted };
    } catch (error) {
      await this.createAgentLifecycleDispatch.cleanupCreatedWorktreeAfterFailedAgentCreate({
        createdWorktree: createdWorktreeForCleanup,
@@ -58,13 +58,13 @@ interface FakeAgentSessionOptions {
  sessionId?: string;
  memoryMarker?: string | null;
  closeSession?: () => Promise<void>;
-  onStartTurn?: (prompt: AgentPromptInput) => void;
+  onStartTurn?: (prompt: AgentPromptInput, config: AgentSessionConfig) => void;
 }

 export interface TestAgentClientOptions {
-  beforeCreateSession?: () => Promise<void>;
+  beforeCreateSession?: (config: AgentSessionConfig) => Promise<void>;
  closeSession?: () => Promise<void>;
-  onStartTurn?: (prompt: AgentPromptInput) => void;
+  onStartTurn?: (prompt: AgentPromptInput, config: AgentSessionConfig) => void;
  supportsMcpServers?: boolean;
 }

@@ -337,7 +337,7 @@ class FakeAgentSession implements AgentSession {
  private activeForegroundTurnId: string | null = null;

  private readonly closeSession: (() => Promise<void>) | undefined;
-  private readonly onStartTurn: ((prompt: AgentPromptInput) => void) | undefined;
+  private readonly onStartTurn: TestAgentClientOptions["onStartTurn"];

  constructor(options: FakeAgentSessionOptions) {
    this.capabilities = {
@@ -440,7 +440,7 @@ class FakeAgentSession implements AgentSession {

    const turnId = `fake-turn-${this.nextTurnOrdinal++}`;
    this.activeForegroundTurnId = turnId;
-    this.onStartTurn?.(prompt);
+    this.onStartTurn?.(prompt, this.config);

    void this.emitTurnEvents(prompt);

@@ -1213,7 +1213,7 @@ class FakeAgentClient implements AgentClient {
    config: AgentSessionConfig,
    _launchContext?: AgentLaunchContext,
  ): Promise<AgentSession> {
-    await this.options.beforeCreateSession?.();
+    await this.options.beforeCreateSession?.(config);
    return new FakeAgentSession({
      providerName: this.provider,
      config: { ...config },