Files
PaperClipAI/tests/runner-e2e/chat-flow.ts
T
DottaandPaperclip 9d19f98b50 fix: harden native chat recovery and add coordination evals (#13741)
## Thinking Path

> - Paperclip helps people manage AI agents and their work.
> - Agent chat uses native runner sessions to plan, delegate, and track
that work.
> - A user can press Stop while the native session is still starting.
> - The server can acknowledge that Stop without dispatching it, then
let the session submit a turn.
> - This leaves chat recovery waiting for an execution that the user
expected to stop.
> - This PR waits for the startup handle, dispatches cancellation, and
prevents a late startup from submitting a turn.
> - New full-stack evals check the resulting records and outputs across
Claude and Codex.
> - Those evals also exposed missing ACPX readiness fields, unbounded
polling, and an old-run identity check that rejected valid warm
handoffs.

## Linked Issues or Issue Description

**What happened?**

Stop during native startup could record an acknowledged cancellation
with `dispatched: false`. The provider could then begin work. A
subsequent `/new` stayed queued. A remote Claude follow-up also
exhausted the command journal while probing warm-session readiness: ACPX
never returned the readiness fields required by the shared transport.
Once readiness worked, attachment incorrectly compared the next run
descriptor against the old run ID. The 25 ms polling loop could issue
4,800 commands during its two-minute wait, beyond the 500-command bound.
The existing chat eval treated lifecycle logs as proof of an active
provider turn, so it did not distinguish startup cancellation from
active-turn cancellation.

**Expected behavior**

A Stop during startup must reach the pending session. A late session
must not submit a prompt after Stop. Recovery must retain control when
startup exceeds the bounded wait. Chat evals must check saved task
state, document contents, worker identity, account binding, and
duplicate effects.

**Steps to reproduce**

1. Start a native Claude or Codex chat turn.
2. Press Stop after process startup is requested but before the provider
turn starts.
3. Send `/new`, then send a fresh message.
4. On the affected base, cancellation can be acknowledged without
dispatch and the reset stays queued.

**Paperclip version or commit**

The live Claude baseline reproduced this on `29d6b3509`. The branch also
includes master commit `0f5fafe16`.

Related work: #13678, #13686, #13693, #13291, #13738. A separate runner
reliability branch also contains a startup-wait fix. Its overlap must be
reconciled before merging; this branch additionally prevents prompt
submission after a late startup.

## What Changed

- Wait for a pending native startup before acknowledging a run-scoped
Stop. Preserve the existing recovery error when that wait expires.
- Keep a Stop guard on startup. Cancel a late handle before it can
submit a provider turn.
- Add regression tests for normal handle publication and publication
after the Stop deadline.
- Back off blocked warm-attachment probes. Keep the fast two-snapshot
barrier, fail closed, and record changed blockers.
- Add red/green tests for delayed readiness, persistent blockers,
alternating readiness, and readiness near the deadline.
- Publish ACPX readiness and blockers. Preserve the old authority’s
event acknowledgement barrier; only settled sessions can proceed to
attachment.
- Bind warm ACPX descriptors to the validated next authority while
retaining old-run event correlation until activation. Preserve session
identity and provider profile checks.
- Exercise two consecutive run rotations through a qualified fake
sidecar, verifying checkpointing, provider identity, pre-activation
rejection, and new-run work admission.
- Separate startup and active-turn cancellation checkpoints in the
browser eval.
- Add 18 explicit native chat eval cells: 12 local and 6 Daytona cells
across Claude and Codex.
- Cover hiring and reuse through managed AI accounts, source-based
review, current blocked-task status, request replay after a lost HTTP
acknowledgement, server restart continuity, and Stop/reset continuity.
- Use ordinary production agent instructions. Enable API tools only for
the two coordination cases that need them.
- Calibrate the matchers with invalid records and outputs. Require
remembered context after restart and a structured status snapshot that
distinguishes the current blocker from history and task status from
active execution. Compare the public issue mutation contract and
relationships during read-only reporting. Preserve before/after source
records in failed eval evidence.
- Fix the lost-ack browser harness and verify it against a real HTTP
server. Check the chat composer after restart instead of waiting for an
unrelated document lifecycle event.
- Document the scope and limits of each case.

## Verification

- The startup regression failed on the unfixed executor and passed after
the fix.
- `pnpm test:e2e:runner:typecheck` passed.
- `pnpm test:e2e:runner:unit` passed: 424 tests in 37 files.
- `pnpm exec vitest run
server/src/services/native-runtime/native-session-executor.test.ts`
passed: 385 tests.
- [Baseline live
campaign](https://github.com/paperclipai/paperclip/actions/runs/35608208868):
Claude Stop reproduced the bug. Codex Stop and Claude hire/reuse passed.
Codex delegation was blocked by provider capacity.
- [Eval-only startup
campaign](https://github.com/paperclipai/paperclip/actions/runs/35609479786):
both providers failed as expected. Both persisted `dispatched: false`
and left `/new` queued.
- [First fixed
campaign](https://github.com/paperclipai/paperclip/actions/runs/35610533706)
on `c9e95797d`: 10/18 cells passed. Startup Stop passed for both
providers. Failed cases exposed eval harness defects and remote
continuity failures. All attempts remain available.
- [Original workflows and stronger memory
checks](https://github.com/paperclipai/paperclip/actions/runs/35611896649)
on `c04324fab`: 9/12 passed. Reassignment, local restart memory, and
startup Stop passed for both providers; Codex remote restart passed.
Claude remote restart exposed the missing readiness contract. Two Codex
planning cells hit provider capacity.
- [Unchanged-model
retry](https://github.com/paperclipai/paperclip/actions/runs/35613854548):
Codex planning and backlog creation both passed.
- [18-cell campaign with ACPX
readiness](https://github.com/paperclipai/paperclip/actions/runs/35614586963)
on `6a98ef743`: 16/18 passed, including all local/remote Stop and
committed-send cases. Claude remote continuity exposed the
next-authority check, now fixed. Codex hiring produced its checklist,
but the runner redacted the requested marker after it appeared as
“Tracking token: …”. That content-redaction policy is unchanged and
remains an explicit limitation.
- [Structured status
grading](https://github.com/paperclipai/paperclip/actions/runs/35614954725)
on `50448c228`: both providers passed on their first attempt, including
cleanup.
- [Complete read-only state
grading](https://github.com/paperclipai/paperclip/actions/runs/35616089011)
on `551e13892`: both providers passed.
- [Final ACPX handoff and hiring
retry](https://github.com/paperclipai/paperclip/actions/runs/35617045456)
on `cbd637587`: all three Claude Daytona cases passed (restart
continuity, active Stop/reset, and lost-ack replay). Codex hiring
reproduced the content-redaction failure: the saved checklist contained
`Tracking token: [REDACTED]` instead of the required business marker.
All four cases completed cleanup successfully. [Published
report](https://d1p6rlowie26tp.cloudfront.net/runner-e2e/campaigns/gha-35617045456-1/).
The only subsequent commit adds the qualified-sidecar integration test;
production code is identical to this live proof.
- `pnpm test:e2e:runner:browser-support` passed: 5 browser tests without
paid models.
- Runner TypeScript typecheck passed. All 5 warm-readiness tests pass;
two failed with the prior fixed-rate loop, and the late-readiness test
failed before the pacing correction.
- ACPX readiness and warm-identity regressions each failed before their
fixes. All 292 runner-core Rust library tests passed. The
qualified-sidecar integration test passes. Rust formatting is checked.
- Status-grader regressions for misleading historical mentions and
previously unchecked mutations each failed before tightening the oracle
and pass now.
- [Latest-head
CI](https://github.com/paperclipai/paperclip/actions/runs/35617522307)
passed on `a4093c8f1`: full build, type checks, test partitions, browser
E2E, and native runner checks. Two unrelated tests initially failed
(Sentry fixture release attribution and local-service fixture
readiness); both passed locally together (35 passed, 5 optional SDK
tests skipped) and on the failed-job retry. No changes were made to
those tests.
- Greptile reviewed `a4093c8f1` at 5/5; both earlier findings are fixed
and all review threads are resolved.
- The paid live suite is not fully green: the reproducible
content-redaction case remains red. This is separate from the passing PR
merge checks. No production content-redaction, prompt, model, or
completion-policy change is included.
- Managed-account hiring and review cases explicitly enable API tools;
these do not qualify default new-user onboarding.

## Risks

- Stop can wait up to 30 seconds for startup, then use the existing
pending-recovery path. This does not prove that remote cleanup has
finished.
- Blocked warm readiness adds up to 750 ms between later probes with the
two-minute remote budget, or about 32 ms with the default five-second
budget. Ready sessions retain the short second barrier.
- Paid evals can fail because of provider capacity or agent decisions.
Each failure needs evidence-based classification.
- The HTTP request replay case checks comment idempotency and duplicate
effects. It does not prove replay safety for an ambiguous provider tool
call.
- The new suite is opt-in. It does not increase the default paid
campaign.
- No production prompts or model selection change. Review-handoff
behavior and content-redaction policy remain separate product decisions.
The latter can remove harmless business content that looks like
credential syntax; the failing attempt is retained.

## Model Used

OpenAI Codex, GPT-6, with repository tools and code execution. The exact
deployment model ID and context window are not exposed in this task.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

---------

Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-09-21 10:35:38 -05:00

963 lines
41 KiB
TypeScript

import { expect, type Page } from "@playwright/test";
import { pollUntil, type RunnerApi } from "./api.js";
import type {
AskUserQuestionsPayload,
PaperclipQuestionSetPayload,
} from "../../packages/shared/src/types/issue.js";
import type { LiveFixtureValues } from "./live-fixtures.js";
import type { MatrixExecution } from "./types.js";
import { isBlockedUnstartedWake } from "./non-execution-wake.js";
import { chatMarker } from "./chat-cases.js";
import { assertChatRememberedAfterRestart, assertChatStartupStopped, isChatStopReady, runChatHardeningFlow } from "./chat-hardening.js";
// Public API observations only: this driver never fabricates provider results or writes DB state.
export interface ChatIssue {
id: string;
companyId: string;
title: string;
status: string;
identifier?: string;
conversationState?: string;
conversationSessionGeneration?: number;
conversationBoundaryCommentId?: string;
parentId?: string | null;
projectId?: string | null;
assigneeAgentId?: string | null;
scheduledRetry?: unknown;
activeRecoveryAction?: unknown;
}
export interface ChatRun {
id: string;
companyId: string;
agentId: string;
status: string;
error?: string | null;
errorCode?: string | null;
runtimeMode?: string;
contextSnapshot?: Record<string, unknown>;
resultJson?: Record<string, unknown>;
sessionIdBefore?: string | null;
sessionIdAfter?: string | null;
startedAt?: string | null;
}
type Comment = {
id: string;
body: string;
authorAgentId?: string;
createdByRunId?: string;
conversationSessionGeneration?: number;
};
type Plan = { body: string; latestRevisionId: string; updatedAt: string };
type ChatOutputDocument = Plan & { id: string; issueId: string; key: string };
export function assertChatBacklogCreation(input: {
tasks: ChatIssue[];
runs: ChatRun[];
ownerId: string;
plan: Plan;
marker: string;
activity: Array<{ action: string; details?: Record<string, unknown> }>;
}) {
expect(input.tasks).toHaveLength(1);
expect(input.tasks[0]).toMatchObject({ status: "backlog", parentId: null, assigneeAgentId: input.ownerId });
expect(input.runs.filter(run => run.contextSnapshot?.issueId === input.tasks[0]!.id)).toHaveLength(0);
expect(input.plan.body).toContain(input.marker);
expect(input.plan.latestRevisionId).toBeTruthy();
const created = input.activity.filter(row => row.action === "issue.created");
expect(created).toHaveLength(1);
// Correcting todo after creation still permits an unauthorized start race.
expect(created[0]!.details).toMatchObject({ status: "backlog", source: "paperclip_runner_protocol" });
}
/** Clarification may request information imperatively rather than end in a question mark. */
export function isChatClarificationReply(body: string): boolean {
if (body.includes("?")) return true;
const request = body.match(
/\b(?:please\s+(?:share|provide|clarify|confirm)|tell me|let me know)\b([\s\S]*)/i,
) ?? body.match(
/\b(?:I|we)(?:'ll|\s+will)?\s+need\s+(?:(?:a|some|the|your|more|following|compact|short|few|additional)\s+){0,4}(?:brief|details|information|context|clarification)\b([\s\S]*)/i,
);
return Boolean(request && /[\p{L}\p{N}]/u.test(request[1]));
}
export function assertChatExecutionOutput(
body: string,
marker: string,
supersededMarker?: string,
): void {
expect(body).toContain(marker);
if (supersededMarker) expect(body).not.toContain(supersededMarker);
}
/** A requested output document may have a descriptive key; a copied plan is not output. */
export async function readChatOutputDocument(
api: Pick<RunnerApi, "get">,
issueId: string,
marker: string,
): Promise<ChatOutputDocument> {
const summaries = await api.get<Array<{ key: string }>>(
`/api/issues/${issueId}/documents`,
);
const documents = await Promise.all(
summaries
.filter((document) => document.key !== "plan")
.map((document) =>
api.get<ChatOutputDocument>(
`/api/issues/${issueId}/documents/${encodeURIComponent(document.key)}`,
),
),
);
const output = documents.find(
(document) => document.issueId === issueId && document.body.includes(marker),
);
if (!output)
throw new Error(
`Execution task ${issueId} has no non-plan output document containing ${marker}; document keys: ${summaries.map((document) => document.key).join(", ") || "none"}`,
);
expect(output.id).toBeTruthy();
expect(output.latestRevisionId).toBeTruthy();
return output;
}
export const isResetRun = (run: ChatRun) =>
run.contextSnapshot?.conversationReset === true ||
run.resultJson?.conversationReset === true;
export function chatRunFailure(
runs: ChatRun[],
allowCancelled = false,
): string | undefined {
const failed = runs.find(
(run) =>
["failed", "timed_out"].includes(run.status) ||
(!allowCancelled && run.status === "cancelled"),
);
return failed
? `run ${failed.id} ${failed.status}${failed.errorCode ? ` (${failed.errorCode})` : ""}${failed.error ? `: ${failed.error}` : ""}`
: undefined;
}
export function chatTaskCompletionFailure(
task: ChatIssue,
runs: ChatRun[],
): string | undefined {
if (
task.scheduledRetry ||
task.activeRecoveryAction ||
runs.some((run) => ["queued", "running"].includes(run.status))
)
return undefined;
return chatRunFailure(runs);
}
/** Require two settled observations so an in-flight finalization is not a failure. */
export function createChatIdleFailureDetector(minimumProviderRuns: number) {
let priorInconsistentState: string | undefined;
return (state: {
resolved: boolean;
status?: string;
conversationState?: string;
providerRunCount: number;
activeRuns: string[];
}): string | undefined => {
const inconsistentState =
state.resolved &&
state.providerRunCount >= minimumProviderRuns &&
state.activeRuns.length === 0 &&
state.conversationState === "waiting" &&
["blocked", "done", "cancelled"].includes(state.status ?? "")
? `${state.status}:${state.providerRunCount}`
: undefined;
const stable =
inconsistentState !== undefined &&
inconsistentState === priorInconsistentState;
priorInconsistentState = inconsistentState;
return stable
? `chat_idle_state_invariant: conversation is ${state.status} while waiting after ${state.providerRunCount} settled provider runs`
: undefined;
};
}
/** Match the shared question form's durable/native presentation, including custom labels. */
export function chatQuestionPresentation(
payload: AskUserQuestionsPayload,
): PaperclipQuestionSetPayload {
if (payload.questionSet) return payload.questionSet;
return {
schema: "paperclip.question_set.v1",
...(payload.submitLabel ? { submitLabel: payload.submitLabel } : {}),
questions: payload.questions.map((question) => {
const freeText = question.options.find((option) => option.freeText);
return {
id: question.id,
prompt: question.prompt,
required: question.required === true,
answerMode:
question.selectionMode === "multi" ? "multi_select" : "single_select",
...(freeText
? { customAnswer: { enabled: true as const, label: freeText.label } }
: {}),
};
}),
};
}
export function assertChatTaskHandoff(
task: ChatIssue,
runs: ChatRun[],
source: ChatIssue,
) {
expect(task.parentId).toBeNull();
expect(task.projectId).toBeTruthy();
expect(task.assigneeAgentId).toBe(source.assigneeAgentId);
expect(runs.length).toBeGreaterThan(0);
}
/** A running row can precede creation of its log file. Only that expected 404 is retryable. */
export async function readRunningChatLog(
api: Pick<RunnerApi, "request">,
runId: string,
): Promise<string | undefined> {
const response = await api.request.get(
`/api/heartbeat-runs/${runId}/log?limitBytes=65536`,
);
if (response.status() === 404) return undefined;
if (!response.ok())
throw new Error(`Run ${runId} log returned ${response.status()}`);
return ((await response.json()) as { content?: string }).content;
}
/** Synthetic reset runs have durable events but never start a provider log. */
export async function collectChatRunEvidence(
api: Pick<RunnerApi, "get">,
run: ChatRun,
) {
return {
runId: run.id,
log: isResetRun(run) || isBlockedUnstartedWake({ ...run })
? null
: await api.get(`/api/heartbeat-runs/${run.id}/log?limitBytes=1048576`),
events: await api.get(`/api/heartbeat-runs/${run.id}/events?limit=1000`),
};
}
export function assertChatHandoff(
task: ChatIssue,
plan: Plan,
runs: ChatRun[],
source: ChatIssue,
) {
assertChatTaskHandoff(task, runs, source);
expect(plan.body.trim()).not.toBe("");
for (const run of runs) {
expect(Date.parse(plan.updatedAt)).toBeLessThanOrEqual(
Date.parse(run.startedAt!),
);
}
}
export async function sendChatMessage(page: Page, message: string) {
const composer = page.getByTestId("task-chat-composer-input").last();
const takeover = page.getByTestId("task-chat-composer-takeover").last();
await expect(composer.or(takeover).first()).toBeVisible();
if (await takeover.isVisible()) {
// Close the card's composer view without answering, skipping, or approving it.
await takeover.getByRole("button", { name: /^Dismiss / }).click();
}
await composer
.locator('[contenteditable="true"], textarea')
.first()
.fill(message);
await page.getByTestId("task-chat-composer-send").last().click();
}
/** Independent durable oracle: prose alone cannot make a reassignment pass. */
export function assertChatReassignment(input: {
readyId: string; queuedId: string; teammateId: string; tasks: ChatIssue[]; runs: ChatRun[];
audit: Array<{ action: string; details?: Record<string, unknown> }>; outputBody: string; marker: string;
}) {
expect(input.tasks.map(task => task.id).sort()).toEqual([input.readyId, input.queuedId].sort());
expect(input.tasks.find(task => task.id === input.readyId)).toMatchObject({ status: "done", assigneeAgentId: input.teammateId });
expect(input.tasks.find(task => task.id === input.queuedId)).toMatchObject({ status: "backlog", assigneeAgentId: input.teammateId });
const successor = input.runs.filter(run => run.contextSnapshot?.issueId === input.readyId);
expect(successor).toHaveLength(1);
expect(successor[0]).toMatchObject({ agentId: input.teammateId, status: "succeeded", runtimeMode: "native" });
expect(input.runs.filter(run => run.contextSnapshot?.issueId === input.queuedId)).toHaveLength(0);
expect(input.audit.filter(row => row.action === "issue.reassigned" && row.details?.source === "paperclip_runner_protocol")).toHaveLength(1);
expect(input.outputBody).toContain(input.marker);
}
export interface ChatFlowInput {
page: Page;
api: RunnerApi;
fixtures: LiveFixtureValues;
execution: MatrixExecution;
nonce: string;
workspacePath: string;
restart: () => Promise<void>;
observe: (issue: ChatIssue, runs: ChatRun[]) => void;
capture: (id: string, label: string, file: string) => Promise<void>;
evidence: (name: string, data: unknown) => Promise<void>;
}
export async function runChatFlow(input: ChatFlowInput) {
const { page, api, fixtures: f, execution, nonce } = input;
const chatPath = `/api/companies/${f.company.id}/chats/${f.agent.id}`;
const route = `/${f.company.issuePrefix}/chats/${f.agent.id}`;
const marker = execution.task.buildVisibleMarker(nonce);
const draftMarker = chatMarker("DRAFT", nonce);
const caseId = execution.task.id;
const stopCase = ["stop-new-resume", "stop-startup-new-resume"].includes(caseId);
let issue: ChatIssue;
let runs: ChatRun[] = [];
const settings = await api.get<Record<string, unknown>>(
"/api/instance/settings/experimental",
);
const allRuns = async () => {
const rows = await api.get<ChatRun[]>(
`/api/companies/${f.company.id}/heartbeat-runs?limit=100`,
);
return Promise.all(
rows.map((row) => api.get<ChatRun>(`/api/heartbeat-runs/${row.id}`)),
);
};
const tasks = () =>
api.get<ChatIssue[]>(`/api/companies/${f.company.id}/issues`);
const comments = async () =>
(
await api.get<Array<Comment & { createdAt: string }>>(
`/api/issues/${issue.id}/comments?order=asc`,
)
).sort(
(a, b) =>
a.createdAt.localeCompare(b.createdAt) || a.id.localeCompare(b.id),
);
const idle = async (minimumProviderRuns: number) => {
const inconsistentIdle = createChatIdleFailureDetector(minimumProviderRuns);
await pollUntil({
label: "chat turn settles to waiting",
deadlineAt: Date.now() + 240_000,
intervalMs: 1000,
load: async () => {
const resolved = await api.get<ChatIssue | null>(chatPath);
if (resolved) issue = resolved;
runs = await allRuns();
if (resolved) input.observe(resolved, runs);
return {
resolved: Boolean(resolved),
status: resolved?.status,
conversationState: resolved?.conversationState,
providerRunCount: runs.filter((run) => !isResetRun(run)).length,
activeRuns: runs
.filter((run) => ["queued", "running"].includes(run.status))
.map((run) => run.id),
failure: chatRunFailure(runs, stopCase),
};
},
reject: (state) => state.failure ?? inconsistentIdle(state),
accept: (state) =>
!state.failure &&
state.resolved &&
state.providerRunCount >= minimumProviderRuns &&
state.activeRuns.length === 0 &&
state.status === "in_review" &&
state.conversationState === "waiting",
});
};
const turn = async (text: string, count: number) => {
await sendChatMessage(page, text);
await idle(count);
};
const noTasks = async () => expect(await tasks()).toHaveLength(0);
try {
await api.patch("/api/instance/settings/experimental", {
enableAgentChat: true,
enableClassicTaskInterface: false,
});
expect(await api.get(chatPath)).toBeNull();
// Cold Vite startup can keep unrelated assets loading after the chat is
// interactive. The composer assertion below verifies actual UI readiness.
await page.goto(route, { waitUntil: "domcontentloaded", timeout: 60_000 });
await expect(page.getByTestId("task-chat-composer-input")).toBeVisible();
expect(await api.get(chatPath)).toBeNull();
expect(await allRuns()).toHaveLength(0);
if (
["continuity-restart", "new-session", "stop-new-resume", "stop-startup-new-resume"].includes(caseId)
) {
const secret = chatMarker("OLDCONTEXT", nonce);
await turn(
`For this conversation only, remember the phrase ${secret}. Just acknowledge briefly; no project or task is needed.`,
1,
);
const initialId = issue!.id;
const before = runs.filter((run) => !isResetRun(run))[0]!;
await noTasks();
await page.reload({ waitUntil: "domcontentloaded", timeout: 60_000 });
if (caseId === "continuity-restart") {
await turn(
"What phrase did I just ask you to remember? Reply with the phrase only.",
2,
);
expect(
(await comments()).filter((c) => c.authorAgentId).at(-1)?.body,
).toContain(secret);
const count = runs.length;
await input.restart();
// Re-enter the canonical route after the server replaces its browser
// transport; reloading the stale document can target a detached page.
// The restarted dev server can leave DOMContentLoaded pending after the
// app is interactive. Require the actual chat composer after navigation.
await page.goto(route, { waitUntil: "commit", timeout: 60_000 });
await expect(page.getByTestId("task-chat-composer-input")).toBeVisible({ timeout: 60_000 });
await idle(2);
expect(runs).toHaveLength(count);
await turn(
`What phrase did I ask you to remember earlier? Reply with that remembered phrase followed by ${marker}; no further work.`,
3,
);
assertChatRememberedAfterRestart(
(await comments()).filter((comment) => comment.authorAgentId).at(-1)?.body ?? "",
secret,
marker,
);
} else {
let cancelledId: string | undefined;
if (stopCase) {
await sendChatMessage(
page,
"Explain the history of gardening at length here, in 100 numbered paragraphs. This is discussion only; do not create work.",
);
await expect
.poll(
async () => {
runs = await allRuns();
const active = runs.find(
(run) => run.status === "running" && run.id !== before.id,
);
if (!active) return false;
const events = await api.get<Array<Record<string, unknown>>>(
`/api/heartbeat-runs/${active.id}/events?limit=1000`,
);
if (execution.profile.generation === "native") {
if (!isChatStopReady(events, caseId === "stop-startup-new-resume" ? "startup" : "active")) return false;
} else if (!(events.length || (await readRunningChatLog(api, active.id))?.length)) return false;
cancelledId = active.id;
await input.evidence("chat-stop-boundary.json", {
phase: caseId === "stop-startup-new-resume" ? "startup" : "active",
runId: active.id, events,
});
return true;
},
{ timeout: 120_000 },
)
.toBe(true);
await page.getByTestId("task-chat-composer-stop").click();
await expect
.poll(
async () =>
(await api.get<ChatRun>(`/api/heartbeat-runs/${cancelledId}`))
.status,
{ timeout: 30_000 },
)
.toBe("cancelled");
}
const oldComments = await comments();
await sendChatMessage(page, "/new");
await expect
.poll(
async () =>
(await api.get<ChatIssue>(chatPath))
.conversationSessionGeneration,
{ timeout: 30_000 },
)
.toBe(1);
await expect
.poll(async () => (await allRuns()).some(isResetRun))
.toBe(true);
await turn(
`Without reading older history or files, if you have a remembered phrase in your current context return it; otherwise reply exactly ${marker}. Do not look it up.`,
caseId === "new-session" ? 2 : 3,
);
const fresh = runs
.filter((run) => !isResetRun(run) && run.status === "succeeded")
.sort((a, b) => Date.parse(a.startedAt!) - Date.parse(b.startedAt!))
.at(-1)!;
expect(fresh.contextSnapshot?.conversationSessionGeneration).toBe(1);
expect(fresh.sessionIdBefore).toBeFalsy();
expect(
String(fresh.contextSnapshot?.paperclipTaskMarkdown ?? ""),
).not.toContain(secret);
if (before.sessionIdAfter && fresh.sessionIdAfter)
expect(fresh.sessionIdAfter).not.toBe(before.sessionIdAfter);
const replies = (await comments()).filter(
(c) => c.createdByRunId === fresh.id && c.authorAgentId,
);
expect(replies.map((c) => c.body).join("\n")).toContain(marker);
expect(replies.map((c) => c.body).join("\n")).not.toContain(secret);
const reset = runs.filter(isResetRun);
expect(reset).toHaveLength(1);
expect(
(await comments()).filter(
(c) => c.authorAgentId && c.createdByRunId === reset[0]!.id,
),
).toHaveLength(0);
expect(
(await comments()).filter((c) =>
oldComments.some((old) => old.id === c.id),
),
).toHaveLength(oldComments.length);
if (cancelledId)
expect(
(await comments()).filter((c) => c.createdByRunId === cancelledId),
).toEqual(
oldComments.filter((c) => c.createdByRunId === cancelledId),
);
if (caseId === "stop-startup-new-resume" && cancelledId) {
const stopped = await api.get<ChatRun>(`/api/heartbeat-runs/${cancelledId}`);
const events = await api.get<Array<{ eventType?: unknown; createdAt?: string }>>(`/api/heartbeat-runs/${cancelledId}/events?limit=1000`);
await input.evidence("chat-startup-stop-result.json", { stopped, events });
assertChatStartupStopped(stopped, events);
}
await page.reload({ waitUntil: "domcontentloaded", timeout: 60_000 });
await expect(
page.getByText("New session", { exact: true }),
).toHaveCount(1);
}
expect(issue!.id).toBe(initialId);
await noTasks();
} else if (["hire-delegate-reuse", "blocked-status-review", "committed-send-retry"].includes(caseId)) {
await runChatHardeningFlow({ input, marker, issue: () => issue!, turn, idle, tasks, allRuns, comments });
} else if (caseId === "create-backlog") {
const project = await api.post<{ id: string }>(`/api/companies/${f.company.id}/projects`, {
name: `Later planning ${nonce}`, description: "Repository-free plans to save for later.",
});
await turn(`Create exactly one task titled Later checklist ${nonce} in the existing Later planning ${nonce} project. Assign it to yourself but keep it in backlog: do not start or execute it. Save a concise three-step initial plan that contains ${marker}. Reply with its identifier and status. Do not create a replacement task or change other tasks.`, 1);
const saved = (await tasks())[0]!;
expect(saved).toBeTruthy();
const plan = await api.get<Plan>(`/api/issues/${saved.id}/documents/plan`);
await turn(`What are the current owner and status of ${saved.identifier}? Just report its saved state. Do not execute it, change its status, or create another task.`, 2);
const observedTasks = await tasks();
// Issue activity also links the conversation run that created the task.
// Inspect actual run bindings to distinguish creation from execution.
const observedRuns = await allRuns();
const activity = await api.get<Array<{ action: string; details?: Record<string, unknown> }>>(`/api/issues/${saved.id}/activity`);
const persistedPlan = await api.get<Plan>(`/api/issues/${saved.id}/documents/plan`);
assertChatBacklogCreation({ tasks: observedTasks, runs: observedRuns, ownerId: f.agent.id, plan: persistedPlan, marker, activity });
expect(observedTasks[0]).toMatchObject({ id: saved.id, projectId: project.id });
expect(persistedPlan).toEqual(plan);
expect((await comments()).filter(c => c.authorAgentId).at(-1)?.body).toMatch(/backlog/i);
await page.reload({ waitUntil: "domcontentloaded", timeout: 60_000 });
await expect(page.getByTestId("task-chat-composer-input")).toBeVisible();
await expect(page.getByRole("link", { name: saved.identifier! }).first()).toBeVisible();
await input.capture("chat-backlog", "Planned backlog task without execution", "chat-backlog.png");
await input.evidence("chat-backlog.json", { tasks: observedTasks, runs: observedRuns, activity, plan: persistedPlan });
} else if (caseId === "reassign-task") {
const config = execution.profile.buildAgent({
environmentId: f.environment.id, environmentFixtureId: execution.environment.id,
workspacePath: input.workspacePath, secretRefs: f.secretRefs, executionId: nonce,
});
const teammate = await api.post<{ id: string }>(`/api/companies/${f.company.id}/agents`, {
...config, name: "Riley Reassignment", role: "engineer", reportsTo: f.agent.id,
instructionsBundle: { entryFile: "AGENTS.md", files: { "AGENTS.md": "Complete the assigned work and save the requested Paperclip document." } },
});
const queued = await api.post<ChatIssue>(`/api/companies/${f.company.id}/issues`, {
title: `Later checklist ${nonce}`, description: `Preserve this deferred scope ${draftMarker}.`,
status: "backlog", assigneeAgentId: f.agent.id,
});
const ready = await api.post<ChatIssue>(`/api/companies/${f.company.id}/issues`, {
title: `Ready checklist ${nonce}`, status: "todo",
description: `Write a short launch checklist as a Paperclip document attached to this task, including ${marker}. Complete this task after saving it.`,
});
await turn(`Assign the existing Ready checklist ${nonce} task to Riley Reassignment so Riley completes it. Also move the existing Later checklist ${nonce} task from you to Riley, keeping it in backlog. Preserve both tasks and their descriptions. Explain the handoff briefly here. Do not create replacement tasks or start the backlog work.`, 2);
const observedTasks = await tasks();
const readyAfter = await api.get<ChatIssue>(`/api/issues/${ready.id}`);
const queuedAfter = await api.get<ChatIssue>(`/api/issues/${queued.id}`);
const output = await readChatOutputDocument(api, ready.id, marker);
const activity = await api.get<Array<{ action: string; details?: Record<string, unknown> }>>(`/api/issues/${ready.id}/activity`);
assertChatReassignment({ readyId: ready.id, queuedId: queued.id, teammateId: teammate.id, tasks: observedTasks, runs,
audit: activity, outputBody: output.body, marker });
expect(queuedAfter).toMatchObject({ assigneeAgentId: teammate.id, status: "backlog", description: `Preserve this deferred scope ${draftMarker}.` });
expect(readyAfter).toMatchObject({ assigneeAgentId: teammate.id, status: "done" });
expect(issue!).toMatchObject({ assigneeAgentId: f.agent.id, conversationState: "waiting" });
await page.reload({ waitUntil: "domcontentloaded", timeout: 60_000 });
await expect(page.getByTestId("task-chat-composer-input")).toBeVisible();
await expect(page.getByRole("link", { name: readyAfter.identifier! }).first()).toBeVisible();
await input.capture("chat-reassignment", "Reassignment completed in Agent Chat", "chat-reassignment.png");
await input.evidence("chat-reassignment.json", { tasks: observedTasks, runs, activity, output, teammateId: teammate.id });
} else {
let existingProject: { id: string; name: string } | undefined;
let acceptedPlan: Plan | undefined;
let repositoryCatalogBefore:
{ repositories: Array<{ id: string; url: string }> } | undefined;
const expectedRepositoryUrls = [
"https://github.com/octocat/Hello-World",
"https://github.com/octocat/Spoon-Knife",
];
if (caseId === "clarify-reuse") {
existingProject = await api.post(
`/api/companies/${f.company.id}/projects`,
{
name: `Garden ${nonce}`,
description: "Garden club welcome notes and event announcements.",
},
);
await turn(
"I need a welcome note for our club. Help me clarify what information you need before assigning the work.",
1,
);
await noTasks();
const questions = await api.get<
Array<{
status: string;
kind: string;
payload: AskUserQuestionsPayload;
}>
>(`/api/issues/${issue!.id}/interactions`);
const pendingInteraction = questions.find(
(row) =>
row.status === "pending" && row.kind === "ask_user_questions",
);
const questionSet = pendingInteraction
? chatQuestionPresentation(pendingInteraction.payload)
: undefined;
const pendingQuestions = questionSet?.questions;
expect(
Boolean(pendingQuestions?.length) ||
(await comments()).some(
(c) => c.authorAgentId && isChatClarificationReply(c.body),
),
).toBe(true);
const clarification = `It is the garden club; use the existing Garden ${nonce} project. Make one assigned task for yourself to write a two-sentence welcome note. Include ${marker} in that note, save it as a Paperclip document attached to that execution task, and finish that execution task. Please get it started now.`;
if (pendingQuestions?.length) {
for (const [index, question] of pendingQuestions.entries()) {
const textInput = page
.getByTestId("question-text-answer-composer")
.last();
if (await textInput.isVisible()) {
await textInput
.locator('[contenteditable="true"],textarea')
.first()
.fill(clarification);
} else {
await page
.getByRole(
question.answerMode === "multi_select" ? "checkbox" : "radio",
{
name: question.customAnswer?.label ?? "Other",
exact: true,
},
)
.last()
.click();
await page
.getByTestId("question-other-answer-composer")
.last()
.locator('[contenteditable="true"],textarea')
.first()
.fill(clarification);
}
await page
.getByRole("button", {
name:
index === pendingQuestions.length - 1
? (questionSet?.submitLabel ?? "Submit answers")
: "Next",
exact: true,
})
.last()
.click();
}
await idle(3);
} else await turn(clarification, 3);
} else if (caseId === "plan-handoff") {
await page.getByTestId("task-chat-composer-mode").click();
await page
.getByTestId("task-chat-composer-mode-menu")
.getByText("Plan mode", { exact: true })
.click();
await turn(
`Let's plan a two-sentence garden club welcome note. The finished welcome note itself must contain the exact phrase ${draftMarker}. Write a plan in the plan panel that includes this requirement, and present it for approval. When I approve the final revision, create a suitable repository-free project and an assigned task for yourself, copy the plan into that task, and have it save the note as a Paperclip document attached to that execution task and finish. Do not create the project or task before approval.`,
1,
);
const draft = await api.get<Plan>(
`/api/issues/${issue!.id}/documents/plan`,
);
expect(draft.body).toContain(draftMarker);
await noTasks();
await input.capture(
"chat-plan-draft",
"Draft plan in the conversation",
"chat-plan-draft.png",
);
const initialInteractions = await api.get<
Array<{
status: string;
kind: string;
payload?: {
target?: { revisionId?: string };
rejectLabel?: string;
};
}>
>(`/api/issues/${issue!.id}/interactions`);
const initialApproval = initialInteractions.find(
(row) =>
row.status === "pending" &&
row.kind === "request_confirmation" &&
row.payload?.target?.revisionId === draft.latestRevisionId,
);
expect(
initialApproval,
"draft has a revision-bound approval",
).toBeTruthy();
const reviseButton = page
.getByRole("button", {
name: initialApproval!.payload?.rejectLabel ?? "Reject",
exact: true,
})
.last();
await reviseButton.click();
await page
.getByTestId("plan-revision-composer")
.last()
.locator('[contenteditable="true"],textarea')
.first()
.fill(
`Revise the plan: the finished welcome note itself must contain the exact phrase ${marker} instead of ${draftMarker}. Include that requirement in the revised plan. The execution task should save that welcome note as a Paperclip document attached to that task. Present this revised plan for approval; wait for that approval before handing it off as agreed.`,
);
await reviseButton.click();
await idle(2);
const revised = await api.get<Plan>(
`/api/issues/${issue!.id}/documents/plan`,
);
expect(revised.body).toContain(marker);
// A revision-history section may quote the superseded requirement.
// The executed output below must use only the accepted requirement.
expect(revised.latestRevisionId).not.toBe(draft.latestRevisionId);
acceptedPlan = revised;
await noTasks();
const interactions = await api.get<
Array<{
id: string;
status: string;
kind: string;
payload?: {
target?: { revisionId?: string };
acceptLabel?: string;
};
}>
>(`/api/issues/${issue!.id}/interactions`);
const approval = interactions.find(
(row) =>
row.status === "pending" &&
row.kind === "request_confirmation" &&
row.payload?.target?.revisionId === revised.latestRevisionId,
);
expect(approval, "approval targets the revised plan").toBeTruthy();
await input.capture(
"chat-plan-revised",
"Revised plan before handoff",
"chat-plan-revised.png",
);
await page
.getByRole("button", {
name: approval!.payload?.acceptLabel ?? "Approve",
exact: true,
})
.last()
.click();
await idle(4);
await input.evidence("chat-plan-revisions.json", {
draft,
revised,
approval,
source: await api.get(`/api/issues/${issue!.id}/documents/plan`),
});
} else {
repositoryCatalogBefore = await api.get(
`/api/companies/${f.company.id}/project-repositories`,
);
expect(
repositoryCatalogBefore!.repositories.filter((repository) =>
expectedRepositoryUrls.includes(repository.url),
),
).toHaveLength(0);
await turn(
`Create a project called Repository Discussion ${nonce} for work spanning https://github.com/octocat/Hello-World and https://github.com/octocat/Spoon-Knife. These existing public repositories are not in our catalog; register both URLs. Then make one assigned task for yourself to write a two-sentence description of the intended project as a Paperclip document attached to that execution task, containing ${marker}, and complete that task. No code changes or remote repository creation are needed.`,
2,
);
}
const children = await tasks();
expect(children).toHaveLength(1);
const child = children[0]!;
await pollUntil({
label: `execution task ${child.id} completes`,
deadlineAt: Date.now() + 240_000,
intervalMs: 1000,
load: async () => ({
task: await api.get<ChatIssue>(`/api/issues/${child.id}`),
runs: await api.get<ChatRun[]>(`/api/issues/${child.id}/runs`),
}),
accept: (state) => state.task.status === "done",
reject: (state) => chatTaskCompletionFailure(state.task, state.runs),
});
runs = await allRuns();
input.observe(issue!, runs);
const plan =
caseId === "plan-handoff"
? await api.get<Plan>(`/api/issues/${child.id}/documents/plan`)
: null;
const taskRuns = runs.filter(
(run) => run.contextSnapshot?.issueId === child.id,
);
if (plan) {
assertChatHandoff(child, plan, taskRuns, issue!);
expect(plan.body).toContain(marker);
const sourcePlan = await api.get<Plan>(
`/api/issues/${issue!.id}/documents/plan`,
);
expect(sourcePlan.body).toBe(acceptedPlan!.body);
expect(sourcePlan.latestRevisionId).toBe(
acceptedPlan!.latestRevisionId,
);
} else assertChatTaskHandoff(child, taskRuns, issue!);
const output = await readChatOutputDocument(api, child.id, marker);
assertChatExecutionOutput(
output.body,
marker,
caseId === "plan-handoff" ? draftMarker : undefined,
);
await input.evidence("chat-execution-output.json", {
taskId: child.id,
document: output,
revisions: await api.get(
`/api/issues/${child.id}/documents/${encodeURIComponent(output.key)}/revisions`,
),
executionRunIds: taskRuns.map((run) => run.id),
});
expect(
(await comments())
.filter((c) => c.authorAgentId)
.map((c) => c.body)
.join("\n"),
).toMatch(new RegExp(`${child.id}|${child.identifier}`));
const projects = await api.get<
Array<{
id: string;
name: string;
workspaces: Array<{ id: string; name: string; repoUrl?: string }>;
}>
>(`/api/companies/${f.company.id}/projects`);
expect(projects).toHaveLength(1);
if (existingProject) {
expect(child.projectId).toBe(existingProject.id);
await expect(
page.getByRole("article", { name: /Project created:/ }),
).toHaveCount(0);
} else {
await expect(
page.getByRole("article", { name: /Project created:/ }),
).toHaveCount(1);
if (caseId === "multi-repository") {
expect(projects[0]!.workspaces.map((w) => w.repoUrl).sort()).toEqual(
[...expectedRepositoryUrls].sort(),
);
// URL registration is persisted as project workspaces; the discovery
// catalog continues to reflect authorized GitHub connections only.
expect(
projects[0]!.workspaces.every((workspace) => Boolean(workspace.id)),
).toBe(true);
expect(
new Set(projects[0]!.workspaces.map((workspace) => workspace.id))
.size,
).toBe(2);
const persistedProject = await api.get<{
workspaces: Array<{ id: string; repoUrl?: string }>;
}>(`/api/projects/${projects[0]!.id}`);
expect(
persistedProject.workspaces.map(({ id, repoUrl }) => ({
id,
repoUrl,
})),
).toEqual(
projects[0]!.workspaces.map(({ id, repoUrl }) => ({ id, repoUrl })),
);
await input.evidence("chat-repository-registration.json", {
catalogBefore: repositoryCatalogBefore,
projectId: projects[0]!.id,
registeredWorkspaces: persistedProject.workspaces,
});
const projectCard = page.getByRole("article", {
name: /Project created:/,
});
// Repository labels may be customized; verify the actual destinations.
for (const repositoryUrl of expectedRepositoryUrls) {
await expect(
projectCard.locator(`a[href="${repositoryUrl}"]`),
).toBeVisible();
}
} else
expect(projects[0]!.workspaces.filter((w) => w.repoUrl)).toHaveLength(
0,
);
await page.reload({ waitUntil: "domcontentloaded", timeout: 60_000 });
await expect(
page.getByRole("article", { name: /Project created:/ }),
).toHaveCount(1);
}
await input.evidence("chat-handoff.json", {
source: issue!,
task: child,
plan,
output,
projects,
});
}
await idle(execution.task.expectedRunCount);
expect(runs.filter((run) => !isResetRun(run))).toHaveLength(
execution.task.expectedRunCount,
);
for (const run of runs.filter((run) => !isResetRun(run))) {
expect(run.runtimeMode).toBe(execution.profile.expectedRuntimeMode);
expect(run.status).toBe(
stopCase && run.status === "cancelled"
? "cancelled"
: "succeeded",
);
}
await input.evidence("api-state.json", {
issue: issue!,
runs,
runGroups: {
resets: runs.filter(isResetRun).map((run) => run.id),
cancelled: runs
.filter((run) => run.status === "cancelled")
.map((run) => run.id),
conversation: runs
.filter(
(run) =>
!isResetRun(run) && run.contextSnapshot?.issueId === issue!.id,
)
.map((run) => run.id),
handoff: runs
.filter((run) => run.contextSnapshot?.issueId !== issue!.id)
.map((run) => run.id),
},
comments: await comments(),
activity: await api.get(`/api/issues/${issue!.id}/activity`),
runEvidence: await Promise.all(
runs.map((run) => collectChatRunEvidence(api, run)),
),
});
await input.capture(
"final-state",
"Chat waiting after its verified workflow",
"final-state.png",
);
return { issue: issue!, runs };
} finally {
await api.patch("/api/instance/settings/experimental", {
enableAgentChat: settings.enableAgentChat,
enableClassicTaskInterface: settings.enableClassicTaskInterface,
});
}
}