Files
PaperClipAI/tests/runner-e2e/chat-catalog-domains/runtime-queue-controls.ts
T
DottaandClaude Opus 5.5 98c18784c6 fix(evals): grade full cohort 4 chat cases on saved state and rendered links (#21034)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - The chat eval catalog runs real agents through chat journeys and
grades the saved results
> - Full cohort 4 (OpenCode with DeepSeek v4.1 flash, master f1aef0cb58)
failed 14 cases for eval reasons, not product reasons
> - Some phase captures read links before the reply rendered. Some
graders matched keywords in free-form replies. One locator was stale.
One fixture answer looked like a placeholder. Some checks could not say
what failed
> - This pull request grades those cases on saved state, setup literals
and rendered links, and it fixes the observers and fixtures
> - The benefit is that these cases fail only when the product fails,
and a failure names the part that broke

## Linked Issues or Issue Description

**What happened?**
In full chat cohort 4, valid agent behavior failed these cases:
case-history, goals.explain-purpose and workspaces.preview-service
(phase captures read before the reply rendered); routines.pause-owned
("If you want that one stopped too, use Pause on its task page" matched
a cancellation regex); budgets-costs.spend-breakdown ("unknown, not
$0.00" matched a stated-zero regex); steering.edit-queued-message
("change ... and drag it above" missed a keyword list);
connection-setup.choose-provider (the resumed page says "Finish with
Notion"); questions-and-approvals.necessary-question (the answer "Acorn
2a4f72ec4ae2-1" was saved as "Acorn"). Some checks could not say what
failed: onboarding.new-company compared the whole members response as
JSON and saved no after-state; agents.choose-primary threw one combined
error.

**Expected behavior**
Each case grades the feature contract on saved state, setup literals and
rendered or clicked links. A valid reworded reply passes. A broken
product still fails. A failure names the clause and its values.

**Steps to reproduce**
Grade the retained cohort 4 evidence for the cases above with the
earlier oracles.

Refs #21009, #21002, #21007

## What Changed

- Phase captures (case-history first phase, goals.explain-purpose
phases, preview-service query phase, ambiguous-task clarification,
edit-queued-message guidance) now wait for the saved reply to render
(`waitForTurnReplyLinks`) before they observe. When a saved reply yields
no rendered links, the observation records thread and anchor diagnostics
(`linkDiagnostics`) and logs them.
- routines.pause-owned: remove the reply keyword clause. Saved state at
the reply proves the run kept running, and the final state proves it
completed normally.
- budgets-costs.spend-breakdown: the unresolved record is graded
structurally. The reply must identify it (rendered link, identifier or
run id), and a chat run must have read its saved accounting (cost
summary, run record or unranged company report, bound by input digest).
"$0.00" in the reply is not matched.
- steering.edit-queued-message: grade the rendered task link in the
reply, the refused agent edit and reorder (403), and the unchanged queue
at the reply. Keyword lists are removed.
- connection-setup.choose-provider: the sign-in locator also accepts
"Finish with Notion" (the resumed-draft label). The button has no test
id or other stable name.
- questions-and-approvals.necessary-question: the answer is now
"Quillmere". The case also accepts the exact answer in a saved
deliverable (a document revision or text attachment on the linked task,
saved by an agent of the company after the answer). The yaml expected
outcome records this decision, with a `changedFields` entry in the
testops manifest.
- onboarding.new-company: save the members after-state and compare
identity, role, status and grants, and name what changed. Guidance is
graded on a rendered link to `/onboarding` (the new-organization
wizard), the business literal from the query, and the saved company
count. The source company and person get realistic names for this case
and are restored at cleanup, because the CEO reasoned from the generated
names that the request was a graded test.
- agents.choose-primary: the settled-Chat check is split into four named
clauses with values, saved as evidence on failure.
- A question card the turn ended on counts as its reply (decision after
cohort 4) in cli-operations.owned-dev-instance, onboarding.new-company
and status-cards.other-authors-card. The card's Markdown-rendered fields
(question form description; confirmation prompt and details) count as
reply links. Link requirements still apply.
- Every changed case bumps its `oracleVersion`; the descriptor snapshot
and inventory digest are updated.

Why each old check was wrong and why the new one still catches a broken
product:

| Case | Old check was wrong because | A broken product still fails
because |
|---|---|---|
| phase captures | runs succeeded and the chat waited before the reply
rendered, so links were empty | the same link checks run on the rendered
reply |
| routines.pause-owned | a correct offer to pause the run read as a
cancellation | the checkpoint run must be running and the run must later
succeed |
| spend-breakdown | "unknown, not $0.00" read as "$0" | the reply must
identify the record and a chat run must read its saved accounting;
totals and per-agent amounts stay graded |
| edit-queued-message | correct wording missed a keyword list | the
reply must link the task, the agent's edit and reorder must be refused,
and the queue must be unchanged at the reply |
| choose-provider | the resumed page label was not accepted | the button
must still be visible on the selected Notion surface |
| necessary-question | the answer read as a placeholder; deliverables
did not count | the exact answer must be saved in the linked work after
the answer, by an agent of the company |
| new-company | timestamps and profiles made the JSON compare fail
without saying why | membership identity, role, status and grants must
be unchanged |
| choose-primary | one combined error | the same four conditions, each
named |
| card as reply | a card answer read as an empty reply | card text must
exist and link requirements still apply |

## Verification

- Calibration tests (valid, reworded valid, broken) for each changed
grader: `integrator-routine-pause.test.ts`,
`identity-accounting.test.ts` (includes the cohort 4 reply),
`runtime-queue-controls.test.ts` (includes the cohort 4 reply),
`external-provider-choice.test.ts`, `identity.test.ts`,
`identity-onboarding-work.test.ts`,
`runtime-primary-ready-chat.test.ts`, `integrator-status-cards.test.ts`,
`turn-response-card.test.ts`, `chat-rendered-links.test.ts`.
- The corrected spend-breakdown grader passes on the retained cohort 4
evidence (its cost-summary read binds to the unresolved task) and fails
with the reads removed.
- `npx vitest run --config tests/runner-e2e/vitest.config.ts
chat-catalog chat-rendered chat-linked`: 284 files passed.
- `node --test
tests/runner-e2e/chat-full-catalog-testops-manifest.test.mjs`: 10/10
passed.
- `node scripts/typecheck-test-projects.mjs tests/runner-e2e`: passed.
- `node tests/runner-e2e/source-sdk-public-inputs.mjs <sha> --check`:
unchanged.
- Paid run (8 cells, OpenCode + DeepSeek v4.1 flash) on this branch
combined with the guidance PR: see the PR comment for results.

## Risks

- The new-company guidance check now needs a rendered link to
`/onboarding`. A reply that only names the company switcher fails. The
companion guidance PR tells agents this route.
- spend-breakdown does not grade wording, so a reply that reads the
record and still calls it "$0" is not caught. The amounts, total and
identification are still graded.
- The new-company case renames the source company and the person for the
case and restores them at cleanup.

## Model Used

- Anthropic Claude Opus 5.5 (`claude-opus-5-5`) in Claude Code, with
tool use, shell execution and code editing.

## 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
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [ ] I will address all Greptile and reviewer comments before
requesting merge

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 00:37:42 -06:00

158 lines
17 KiB
TypeScript

import { chatCatalogNativePhaseDeadline, chatCatalogNativePhaseIsTimely } from "../chat-catalog-contract.js";
import { expect, type Page } from "@playwright/test";
import { writeFile } from "node:fs/promises";
import type { IssueQueuedCommentQueue } from "../../../packages/shared/src/types/issue.js";
import type { ChatCatalogContext } from "../chat-catalog-contract.js";
import { pollUntil } from "../api.js";
import { submitTaskReply } from "../user-actions.js";
import { checks, hasLink, type Entity } from "./internal-support.js";
import { probeAgentPermission, cleanupIdentityProbes, waitForTurnReplyLinks } from "./identity-support.js";
import { busyRuntimeWorker, releaseRuntimeWorker } from "./runtime-busy-worker.js";
import { runtimeQueueCases } from "./runtime-queue.js";
import { observeRuntime, runtimeDefinition, runtimeObservation } from "./runtime-support.js";
const otherText = "Use a readable font with accessible labels.";
const redText = "Make the website red. Keep the existing note about accessible labels.";
const blueText = "Make the website blue. Keep the existing note about accessible labels.";
type State = Record<string, unknown>;
function endpoint(state: State) { return `/api/issues/${(state.task as Entity).id}/queued-comments`; }
async function queue(ctx: ChatCatalogContext, state: State) { return ctx.input.api.get<IssueQueuedCommentQueue>(endpoint(state)); }
function taskUrl(ctx: ChatCatalogContext, state: State) { const task = state.task as Entity; return `/${ctx.input.fixtures.company.issuePrefix}/issues/${task.identifier ?? task.id}`; }
export async function humanQueuedRuntime(ctx: ChatCatalogContext, targetText = redText): Promise<State> {
const state: State = await busyRuntimeWorker(ctx); const source = ctx.input.page.url();
await ctx.input.page.goto(taskUrl(ctx, state), { waitUntil: "domcontentloaded" });
await submitTaskReply(ctx.input.page, otherText); await submitTaskReply(ctx.input.page, targetText);
const before = await pollUntil({ label: "Two real user-owned messages saved behind Morgan's active turn", deadlineAt: Date.now() + 30_000,
load: () => queue(ctx, state), accept: q => q.entries.length === 2 && q.entries[0]?.comment.body === otherText && q.entries[1]?.comment.body === targetText });
const session = await ctx.input.api.get<{ user: { id: string } }>("/api/auth/get-session");
if (before.entries.some(row => row.comment.authorUserId !== session.user.id || row.comment.authorAgentId || !row.canEdit)) throw new Error("Queue fixture is not genuinely owned by the signed-in human");
Object.assign(state, { queueBefore: before, targetId: before.entries[1]!.comment.id, humanId: session.user.id, concurrentMarker: `Preserve the human note ${ctx.input.nonce}.` });
await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" }); return state;
}
export async function probeConsumedHumanQueue(ctx: ChatCatalogContext, humanId: string, endpoint: string, method: "PATCH" | "DELETE", data: unknown) {
const origin = ctx.input.api.baseURL, url = new URL(endpoint, origin);
if (ctx.scenario.id !== "steering.edit-queued-message" && ctx.scenario.id !== "steering.discard-queued-message") throw new Error("Consumed queue probe requires its declared human scenario");
if (!humanId || new URL(ctx.input.page.url()).origin !== origin || url.origin !== origin || url.search || url.hash
|| !/^\/api\/issues\/[^/]+\/queued-comments\/[^/]+$/.test(url.pathname)) throw new Error("Consumed queue probe escaped its human fixture");
const apiSession = await ctx.input.api.get<{ user?: { id?: string } }>("/api/auth/get-session");
const browserSession = await ctx.input.page.request.get(`${origin}/api/auth/get-session`, { maxRedirects: 0 });
if (browserSession.status() !== 200 || apiSession.user?.id !== humanId || (await browserSession.json() as { user?: { id?: string } }).user?.id !== humanId)
throw new Error("Consumed queue probe requires the original authenticated human in both contexts");
const response = await ctx.input.page.request.fetch(url.href, { method, data, headers: { Origin: origin }, maxRedirects: 0, timeout: 15_000 });
return response.status();
}
async function edit(page: Page, id: string, body: string) {
await page.getByTestId(`task-chat-queued-message-${id}`).getByRole("button", { name: /^Queued message actions:/ }).click();
await page.getByRole("menuitem", { name: "Edit message", exact: true }).click();
await page.getByTestId("task-chat-composer-input").last().locator('[contenteditable="true"], textarea').first().fill(body);
}
export function gradeQueueEdit(value: unknown) {
const c = checks(), o = runtimeObservation(value);
c.add("real-human-queue-evidence", !!o, "The owned queue, native delivery, visible conflicts and saved full text are required."); if (!o) return c.result();
const s = o.state, before = s.queueBefore as IssueQueuedCommentQueue, after = s.queueAfter as IssueQueuedCommentQueue;
const target = before?.entries.find(row => row.comment.id === s.targetId), saved = after?.entries.find(row => row.comment.id === s.targetId);
c.add("human-owned-undispatched-target", !!target && target.comment.body === redText && target.comment.authorUserId === s.humanId && !target.comment.authorAgentId
&& target.canEdit && before.issueId === (s.task as Entity).id && before.targetRunId === s.workerRunId, "The original red request belongs to the actual signed-in human and waits behind this actual native run.");
c.add("full-edited-text-and-first-position-persist", !!saved && saved.position === 0 && saved.comment.body === `${blueText} ${s.concurrentMarker}`
&& saved.comment.authorUserId === s.humanId && !saved.comment.authorAgentId && after.entries.length === 2
&& after.entries.some(row => row.comment.body === otherText && row.position === 1), "The same comment retains its author, the complete changed text, the concurrent human note and first position; the other message is intact.");
// Graded on saved state and the rendered link, never wording ("change ... and drag it above" and "use Edit in its
// menu" are both correct guidance). The reply links the task whose queue holds the person's controls, the agent's
// own edit and reorder are refused, and at the reply the queue is exactly as the person left it: no edit applied.
const atReply = s.queueAtReply as IssueQueuedCommentQueue | undefined;
const unchangedAtReply = !!before && !!atReply && atReply.entries.length === before.entries.length && before.entries.every(old => atReply.entries.some(row =>
row.comment.id === old.comment.id && row.position === old.position && row.comment.body === old.comment.body && row.comment.authorUserId === old.comment.authorUserId && !row.comment.authorAgentId));
const replyLinks = Array.isArray(s.queryLinks) ? s.queryLinks as string[] : [];
c.add("chat-guides-human-control-without-forging", !!String(s.queryReply ?? "").trim() && !!s.task && hasLink({ ...o, links: replyLinks }, "issues", s.task as Entity)
&& unchangedAtReply && s.agentEditStatus === 403 && s.agentOrderStatus === 403,
"The canonical reply links the task whose queue holds the person's controls, the actual agent cannot edit or reorder the person's message, and no edit was applied before the person acted.");
c.add("concurrent-edit-rejects-stale-request-visibly", s.concurrentStatus === 409 && s.concurrentErrorVisible === true && s.concurrentSaveRetained === true,
"An actual held old-revision human request loses to a real second-session edit and shows its conflict, retaining the winner's text.");
c.add("dispatched-edit-is-visible-new-direction", s.dispatchedStatus === 409 && s.dispatchedWarningVisible === true && s.originalCommentStillReadable === true,
"After actual dispatch, the old edit rejects and the real composer offers a new message rather than rewriting consumed history.");
c.add("exactly-one-normal-native-consumer", s.nativeDeliveryBound === true && s.queueEmptyAfterDelivery === true && o.runs.length === 3
&& o.runs.every(row => row.status === "succeeded") && s.consumingRunCount === 1 && s.initialRevisionRetained === true,
"One actual ordinary continuation receives the edited message once, in its final order, while original work remains preserved.");
return c.result();
}
export const runtimeQueueControlCases = [runtimeDefinition("steering.edit-queued-message", ctx => humanQueuedRuntime(ctx), gradeQueueEdit, {
sourceFiles: ["chat-catalog-domains/runtime-queue-controls.ts", "chat-catalog-domains/runtime-queue.ts", "chat-catalog-domains/runtime-busy-worker.ts", "chat-catalog-domains/runtime-support.ts", "chat-catalog-domains/identity-support.ts", "chat-catalog-domains/internal-support.ts", "chat-rendered-links.ts", "chat-scenarios/steering.yaml"],
// v3: guidance graded on the task link, refused agent writes and the unchanged queue at the reply, not keywords.
oracleVersion: "paperclip.chat-catalog.edit-queued-message.structural.v3",
authentication: "human-session", expectedRunCount: 3, attemptTimeoutMs: 720_000,
setupPolicy: runtimeQueueCases[0]!.setupPolicy, verifySetup: runtimeQueueCases[0]!.verifySetup,
async act(ctx, state) {
await pollUntil({ label: "Canonical editing guidance while the real worker remains active", deadlineAt: chatCatalogNativePhaseDeadline(ctx, Date.now() + 150_000),
load: async () => { await ctx.refreshIssue(); return { source: ctx.issue(), runs: await ctx.allRuns() }; },
accept: v => chatCatalogNativePhaseIsTimely(ctx) && (v.source?.conversationState === "waiting" && v.runs.length === 2 && v.runs.some(row => row.id === state.workerRunId && row.status === "running")
&& v.runs.some(row => row.agentId === ctx.input.fixtures.agent.id && row.status === "succeeded")) });
// The finished run and waiting chat precede the saved reply's render; read its links only after it renders.
await waitForTurnReplyLinks(ctx);
const answered = await observeRuntime(ctx, state); state.queryReply = answered.reply; state.queryLinks = answered.links; const source = ctx.input.page.url();
const initial = await queue(ctx, state), targetId = String(state.targetId), mutation = `${endpoint(state)}/${targetId}`;
state.queueAtReply = initial;
state.agentEditStatus = await probeAgentPermission(ctx, mutation, "PATCH", { queueId: initial.queueId, revision: initial.revision, body: blueText });
state.agentOrderStatus = await probeAgentPermission(ctx, `${endpoint(state)}/order`, "PUT", { queueId: initial.queueId, revision: initial.revision, orderedCommentIds: initial.entries.map(row => row.comment.id).reverse() });
await ctx.input.page.goto(taskUrl(ctx, state), { waitUntil: "domcontentloaded" });
await edit(ctx.input.page, targetId, blueText);
// Hold a real old-revision PATCH before admission. A second human tab then
// saves the competing revision through the ordinary supported control.
let release!: () => void, arrived!: () => void;
const held = new Promise<void>(resolve => { release = resolve; }), intercepted = new Promise<void>(resolve => { arrived = resolve; });
let first = true; const pattern = `**${mutation}`;
await ctx.input.page.route(pattern, async route => {
if (route.request().method() === "PATCH" && first) { first = false; arrived(); await held; }
await route.continue();
});
const concurrent = await ctx.input.page.context().newPage();
try {
const saving = ctx.input.page.getByRole("button", { name: "Save queued message", exact: true }).click();
await Promise.all([intercepted, saving]);
await concurrent.goto(taskUrl(ctx, state), { waitUntil: "domcontentloaded" });
await edit(concurrent, targetId, `${redText} ${state.concurrentMarker}`);
await concurrent.getByRole("button", { name: "Save queued message", exact: true }).click();
await pollUntil({ label: "Concurrent human edit persisted", deadlineAt: Date.now() + 15_000, load: () => queue(ctx, state),
accept: q => q.entries.some(row => row.comment.id === targetId && row.comment.body === `${redText} ${state.concurrentMarker}`) });
const rejected = ctx.input.page.waitForResponse(response => response.request().method() === "PATCH" && new URL(response.url()).pathname === mutation);
release(); state.concurrentStatus = (await rejected).status(); await saving;
await expect(ctx.input.page.getByText(/queue.*changed|changed.*queue|revision|refresh|try again/i).first()).toBeVisible(); state.concurrentErrorVisible = true;
state.concurrentSaveRetained = (await queue(ctx, state)).entries.some(row => row.comment.id === targetId && row.comment.body === `${redText} ${state.concurrentMarker}`);
} finally { release(); await ctx.input.page.unroute(pattern); await concurrent.close(); }
await ctx.input.page.reload({ waitUntil: "domcontentloaded" });
await edit(ctx.input.page, targetId, `${blueText} ${state.concurrentMarker}`);
await ctx.input.page.getByRole("button", { name: "Save queued message", exact: true }).click();
const handle = ctx.input.page.getByRole("button", { name: `Reorder queued message: ${blueText} ${state.concurrentMarker}`, exact: true });
await handle.focus(); await handle.press("Space"); await handle.press("ArrowUp"); await handle.press("Space");
state.queueAfter = await pollUntil({ label: "Actual full blue text and first queue position persisted", deadlineAt: Date.now() + 15_000,
load: () => queue(ctx, state), accept: q => q.entries[0]?.comment.id === targetId && q.entries[0]?.comment.body === `${blueText} ${state.concurrentMarker}` });
await ctx.input.page.reload({ waitUntil: "domcontentloaded" });
await edit(ctx.input.page, targetId, `${blueText} ${state.concurrentMarker}`);
const pending = state.queueAfter as IssueQueuedCommentQueue;
await writeFile((state.gate as { gate: string }).gate, "The human brief is ready. Save the current website brief as a task document, then finish this turn normally. Leave saved follow-up directions intact for the normal continuation.", "utf8");
const delivered = await pollUntil({ label: "Edited messages reach one ordinary native continuation", deadlineAt: Date.now() + 240_000,
load: async () => ({ runs: await ctx.allRuns(), queue: await queue(ctx, state), task: await ctx.input.api.get<Entity>(`/api/issues/${(state.task as Entity).id}`) }),
accept: v => v.runs.length === 3 && v.runs.every(row => row.status === "succeeded") && v.queue.entries.length === 0 && v.task.status === "done",
reject: v => v.runs.length > 3 || v.runs.some(row => ["failed", "cancelled", "timed_out"].includes(row.status)) ? "The declared edited-message delivery failed or duplicated work" : undefined });
await expect(ctx.input.page.getByText("Queued message changed", { exact: true })).toBeVisible();
await expect(ctx.input.page.getByRole("button", { name: "Queue as new message", exact: true })).toBeVisible(); state.dispatchedWarningVisible = true;
state.dispatchedStatus = await probeConsumedHumanQueue(ctx, String(state.humanId), mutation, "PATCH", { queueId: pending.queueId, revision: pending.revision, body: "Late stale edit must not replace history" });
const comments = await ctx.input.api.get<Entity[]>(`/api/issues/${(state.task as Entity).id}/comments?order=asc`);
state.originalCommentStillReadable = comments.some(row => row.id === targetId && row.body === `${blueText} ${state.concurrentMarker}` && row.authorUserId === state.humanId);
const consumers = delivered.runs.filter(row => row.agentId === (state.worker as Entity).id && row.id !== state.workerRunId);
const run = consumers[0] && await ctx.input.api.get<{ contextSnapshot?: { wakeCommentIds?: string[] }; runnerProfileJson?: { nativeExecutionInput?: { task?: { prompt?: string } } } }>(`/api/heartbeat-runs/${consumers[0].id}`);
const ids = run?.contextSnapshot?.wakeCommentIds ?? [], prompt = run?.runnerProfileJson?.nativeExecutionInput?.task?.prompt ?? "";
state.consumingRunCount = consumers.length; state.nativeDeliveryBound = ids.length === 2 && ids[0] === targetId
&& ids[1] === initial.entries.find(row => row.comment.id !== targetId)?.comment.id && prompt.includes(`${blueText} ${state.concurrentMarker}`) && prompt.includes(otherText);
state.queueEmptyAfterDelivery = delivered.queue.entries.length === 0;
await ctx.input.capture("actual-queued-edit-stale-dispatch", "Consumed human direction remains history and stale editing offers a new message", "runtime-queued-edit.png");
await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" });
},
async observe(ctx, state) {
const o = await observeRuntime(ctx, state);
state.initialRevisionRetained = o.revisions.some(row => row.createdByRunId === state.workerRunId && row.createdByAgentId === (state.worker as Entity).id);
return o;
},
async cleanup(ctx, state) { try { await cleanupIdentityProbes(ctx); } finally { await releaseRuntimeWorker(state); } },
})];