Files
PaperClipAI/tests/runner-e2e/chat-catalog-domains/runtime-delegation.ts
T
f4337bb5fb test(evals): land the chat catalog eval harness on master (#19614)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - The chat-feature evals measure whether agents complete real chat
requests through the product
> - The chat catalog eval harness and its repairs live on the
`chat-eval/master-candidate` branch, not on master
> - Today's work made 139 of 242 chat evals pass in verified single-case
live runs. Each product fix landed on master as its own pull request
> - The eval-side changes (setups, graders, fault seams, evidence
capture) still exist only on the candidate
> - This pull request lands the harness on master. Master stays the
source of truth for product behavior
> - The benefit is one tree: evals run against master commits, and
product fixes reach the evals as soon as they merge

## Linked Issues or Issue Description

Refs #19441, Refs #19599, Refs #20270, Refs #20334, Refs #20342, Refs
#20344, Refs #20354, Refs #20360, Refs #20373, Refs #20382, Refs #20409

**What existing behavior does this improve?**
The chat catalog eval harness and the product it evaluates live on a
long-running branch outside master. Eval fixes do not reach master, so
master-based eval runs use old setups and graders.

**Subsystem affected**
`tests/runner-e2e` chat catalog harness, plus a small set of product
changes that exist only on the candidate.

**Proposed change**
Land the harness on master. Every product fix from this eval work
already merged to master as its own pull request. Conflicts keep
master's product behavior.

## What Changed

- Most of the diff (808 of 851 files) is under `tests/runner-e2e/`:
chat-catalog domains, scenarios, descriptors, the TestOps manifest,
private evidence tooling, and their tests.
- Graders now check saved state, setup literals and rendered or clicked
links. Many keyword and regex checks on free-form replies are removed.
Each changed grader has a bumped `oracleVersion` and calibration tests
(a valid reply, a reworded valid reply, and broken variants that must
fail).
- The shared link helper `hasLink` no longer treats a list link (for
example `/COM/issues?view=mine`) as a link to every task without a URL
key.
- Setup fixes build real state through supported product routes: routine
history, agent grants through the board grant route, budget holds with a
real cap, and a storage-layer archive fault for
`inbox-search.archive-partial-failure`.
- Browser boundary journeys (sign-in, chat-disabled, onboarding) now
write the standard pass evidence files.
- Retargeted cases, approved by the maintainer:
`agent-chat.chat-disabled` (Conference Room gate),
`agents.restore-instructions` (no saved instruction history),
`access.manage-membership` (member permission edits stay board-only).
- The private evidence trusted files (`manual-private-handoff.ts`,
`manual-private-retention.ts`, `manual-private-report-assets.ts`) are
byte-identical to master.
- Product files outside `tests/` (43 files, +1,328 −49) carry
candidate-only changes that master does not yet have. Most are tests.
The main code changes are in `request-human-input.ts`,
`heartbeat-run-summary.ts`, `native-completion-feedback.ts`,
`native-run-finalizer.ts`, `openapi.ts`, `secrets.ts` and
`IssueChatLongThreadPerf.tsx`.

## Verification

- `pnpm vitest run chat-catalog-domains chat-catalog-contract.test.ts
chat-catalog-consistency.test.ts unit-launcher.test.mjs
--testTimeout=180000` in `tests/runner-e2e`: 269 files pass, 3,550 tests
pass.
- `node --test chat-full-catalog-testops-manifest.test.mjs`: 10 pass.
- `node tests/runner-e2e/source-sdk-public-inputs.mjs <HEAD> --check`:
no change.
- `tsc` for `server` and `tests/runner-e2e`: clean.
- Live single-case verification: 139 of 242 cases passed on paid runs,
and a person read the saved evidence for each. These are single-case
runs, not one full-cohort run.

## Risks

- Large diff. Most of it is test-only, but the 43 non-test files change
product code paths. Review those files first.
- Some graders are now less strict on reply wording. A person must read
replies for meaning on those cases.
- 42 cases need disposable outside accounts (Notion, Slack, GitHub,
Daytona and others) and stay blocked. `access.invite-teammate` stays
failing until local agent isolation lands, because the runner's durable
state keeps a live invite token.
- The paid workflow compares the trusted private-evidence files with
master. They are unchanged, so this risk is low.

## Model Used

Anthropic Claude Opus 5.5 (`claude-opus-5-5`), with tool use and code
execution, in Claude Code. Subagents used the same model.

## 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
- [x] 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>
Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-11-10 08:02:11 -06:00

243 lines
25 KiB
TypeScript

import { chatCatalogNativePhaseDeadline, chatCatalogNativePhaseIsTimely, chatCatalogNativePhaseRemainingBudget, chatCatalogOriginalPhaseDeadline } from "../chat-catalog-contract.js";
import { createHash } from "node:crypto";
import { readFile, writeFile } from "node:fs/promises";
import { expect, type Page } from "@playwright/test";
import type { ChatCatalogContext } from "../chat-catalog-contract.js";
import type { ChatQueryAttachmentReceipt } from "../chat-query-attachments.js";
import { pollUntil } from "../api.js";
import { sendChatMessage } from "../chat-flow.js";
import { readRegisteredArtifacts } from "../registered-artifact.js";
import { observeCompletionUpdate } from "../completion-update-flow.js";
import { completionDelivery } from "../completion-updates.js";
import { checks, type Entity } from "./internal-support.js";
import { openStandingChat } from "./identity-support.js";
import { prepareRuntimeWorkerGate, releaseRuntimeWorker } from "./runtime-busy-worker.js";
import { runtimeDefinition, runtimeObservation, observeRuntime, type RuntimeAgent } from "./runtime-support.js";
const keys = new WeakMap<ChatCatalogContext, { id: string; token: string; workerId: string }>();
const hash = (value: Uint8Array) => createHash("sha256").update(value).digest("hex");
/**
* User-facing promptness of "report back here": once Morgan's task is Done, the source conversation
* gets the result in one ordinary CEO turn. Measured in the 2026-11-09 cell (run 42612964001): wake
* claimed 0.3 s after completion, provider boot ~9 s, first token ~18 s, whole CEO turns 68-74 s.
* A deferred or lost wake, or extra turns before delivery, exceed this. Morgan's own build time is an
* execution allowance bounded by the case's original attempt clock, not by this requirement.
*/
export const RETURN_DELIVERY_LATENCY_MS = 90_000;
/** True when any link target (Markdown or rendered href) opens the given task in this instance. */
export function linksOpenTask(targets: readonly string[], baseURL: string, task: { id?: string; identifier?: string | null }) {
return targets.some(target => {
try { const url = new URL(target, baseURL), parts = url.pathname.split("/").map(decodeURIComponent), i = parts.indexOf("issues");
return url.origin === new URL(baseURL).origin && i >= 0 && [task.id, task.identifier].filter(Boolean).includes(parts[i + 1]); }
catch { return false; }
});
}
export function launchBriefBrowserOrigin(apiBaseURL: string, browserURL: string, target: string, companyId: string, issueId: string) {
const api = new URL(apiBaseURL), browser = new URL(browserURL), upload = new URL(target, api);
const uuid = /^[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12}$/i;
if (!uuid.test(companyId) || !uuid.test(issueId) || !["http:", "https:"].includes(api.protocol)
|| api.origin !== apiBaseURL || api.username || api.password || browser.username || browser.password
|| browser.origin !== api.origin || upload.origin !== api.origin || upload.username || upload.password || upload.search || upload.hash
|| upload.pathname !== `/api/companies/${companyId}/issues/${issueId}/attachments`) throw new Error("Human Launch brief upload escaped its owned browser and target");
return browser.origin;
}
export function completeLaunchBrief(context: string, marker: string) {
return context.includes(marker) && /Research without the busywork/i.test(context) && /\bsemantic\b.{0,30}\b(?:html|markup|structure)\b/i.test(context)
&& /\b(?:accessible|labelled|labeled)\b.{0,30}\b(?:label|input|control|form)|\blabel.{0,30}accessib/i.test(context)
&& /\b(?:primary|one|single)\b.{0,35}\b(?:sign.?up|signup|join|subscribe)\b.{0,30}\b(?:action|cta|button|link)\b/i.test(context);
}
/** What a person (or assistive technology) gets from Morgan's page, read from the browser's own rendering of the delivered bytes. */
export type LaunchPageFacts = { headline: string | null; landmarks: number; unlabelledControls: number; namedSignupActions: number; signupForms: number };
const SIGNUP_ACTION_NAME = /sign.?up|join|subscribe|early access|waitlist|wait list|register|get started/i;
/** Render the delivered page with every network request blocked and read it like a person would. */
export async function launchPageFacts(page: Page, html: string): Promise<LaunchPageFacts> {
const view = await page.context().newPage();
try {
await view.route("**/*", route => route.abort());
await view.setContent(html, { waitUntil: "domcontentloaded" });
const h1 = view.locator("h1").first();
const headline = await h1.count() ? (await h1.textContent()) ?? "" : null;
const controls = view.locator("input:not([type=hidden]):not([type=submit]):not([type=button]):not([type=reset]):not([type=image]), select, textarea");
const unlabelledControls = (await controls.evaluateAll(nodes => nodes.map(node => {
const element = node as HTMLInputElement, labels = element.labels ? Array.from(element.labels) : [];
return labels.some(label => (label.textContent ?? "").trim()) || !!(element.getAttribute("aria-label") ?? "").trim()
|| (element.getAttribute("aria-labelledby") ?? "").split(/\s+/).some(id => !!id && !!(document.getElementById(id)?.textContent ?? "").trim());
}))).filter(labelled => !labelled).length;
return {
headline, unlabelledControls,
landmarks: await view.locator("main, section, header").count(),
namedSignupActions: await view.getByRole("button", { name: SIGNUP_ACTION_NAME }).or(view.getByRole("link", { name: SIGNUP_ACTION_NAME })).count(),
signupForms: await view.locator("form:has(input[type=email]):has(button:not([type=button]):not([type=reset]), input[type=submit])").count(),
};
} finally { await view.close(); }
}
/** The brief: exact headline, semantic structure, every control labelled, a signup action (named as one, or an email form that submits). */
export function launchPageReflectsBrief(facts: LaunchPageFacts | undefined) {
return !!facts && (facts.headline ?? "").replace(/\s+/g, " ").trim().toLowerCase() === "research without the busywork"
&& facts.landmarks > 0 && facts.unlabelledControls === 0 && (facts.namedSignupActions > 0 || facts.signupForms > 0);
}
export function gradeDelegation(value: unknown) {
const c = checks(), o = runtimeObservation(value);
c.add("actual-launch-brief-and-native-handoff-evidence", !!o, "A real human-attached Launch brief, exactly one CEO-created Morgan child, native result and source completion are required."); if (!o) return c.result();
const s = o.state, parent = s.parent as Entity, task = s.task as Entity, receipt = s.briefAttachment as ChatQueryAttachmentReceipt;
const availableBrief = ["copied-attachment", "direct-attachment-reference"].includes(String(s.briefContextKind)) ? s.workerBriefReadStatus === 200 && s.workerBriefHash === s.briefHash
: s.briefContextKind === "readable-parent-reference" ? s.workerParentReadStatus === 200 && s.parentLinkInspectable === true && s.workerBriefReadStatus === 200 && s.workerBriefHash === s.briefHash
: s.briefContextKind === "faithful-durable-copy" && s.durableBriefComplete === true && s.actualWebsiteArtifact === true;
c.add("real-parent-owner-and-human-brief", s.parentOwnedByCeo === true && !!receipt && receipt.issueId === parent.id && receipt.companyId === o.companyId
&& receipt.createdByAgentId == null && receipt.createdByUserId === s.humanId && receipt.issueCommentId === s.briefCommentId && receipt.sha256 === s.briefHash
&& s.briefDownloadedHash === s.briefHash && s.parentContextRetained === true,
"The real Launch task remains owned by the CEO and its human comment retains the exact uploaded design brief.");
c.add("one-authorized-child-with-durable-context", !!task && (task.parentId === parent.id || s.linkedToLaunch === true) && task.companyId === o.companyId && task.assigneeAgentId === (s.worker as Entity).id
&& s.createdFromSource === true && o.tasks.length === 2 && s.childHasBrief === true && availableBrief && s.workerTaskReadStatus === 200
&& o.activity.some(row => row.action === "issue.created" && row.entityId === task.id && row.agentId === o.leadId),
"The canonical request creates one real Morgan child linked to Launch, with accessible durable brief bytes. The completion request is the product's created-from-source return path, proven by the correlated delivery below; the description's wording is left to semantic review.");
const delivery = s.delivery as ReturnType<typeof completionDelivery> | undefined;
c.add("actual-morgan-result-incorporated-back-here", s.workerWasReallyWaiting === true && task.status === "done" && s.actualWebsiteArtifact === true
&& !!delivery && delivery.checks.every(row => row.passed),
"Morgan produces the actual brief-based page and the same source receives a reply run given Morgan's Done result, linking or quoting it. Whether the wording credits Morgan is semantic review; parent execution is checked structurally below.");
c.add("prompt-return-delivery", typeof s.returnDeliveryLatencyMs === "number" && s.returnDeliveryLatencyMs >= 0 && s.returnDeliveryLatencyMs <= RETURN_DELIVERY_LATENCY_MS,
`Once Morgan's task is Done, the result reaches this conversation within ${RETURN_DELIVERY_LATENCY_MS / 1000} s (one ordinary turn, no deferred wake or extra turns).`);
c.add("separate-native-ownership-and-no-duplicate-work", o.runs.length === 4 && o.runs.every(row => row.status === "succeeded")
&& o.runs.filter(row => row.agentId === (s.worker as Entity).id).length === 1
&& !o.runs.some(row => row.contextSnapshot?.issueId === parent.id || (row as typeof row & { nativeIssueId?: string }).nativeIssueId === parent.id),
"The declared setup acknowledgement, canonical CEO turn, single Morgan child execution and one correlated completion are the entire native ledger; Launch has no masquerading execution.");
return c.result();
}
async function renderedDescriptionLinks(ctx: ChatCatalogContext, task: Entity): Promise<string[]> {
const page = ctx.input.page, source = page.url();
try {
await page.goto(`/${ctx.input.fixtures.company.issuePrefix}/issues/${task.identifier ?? task.id}`, { waitUntil: "domcontentloaded" });
const bubble = page.getByTestId("task-chat-description-bubble").first();
await expect(bubble).toBeVisible();
let links: string[] = [];
// Issue references link once the company's prefixes load; give the render a moment, then read it.
await expect.poll(async () => (links = await bubble.locator("a[href]").evaluateAll(nodes => nodes.map(node => node.getAttribute("href")!))).length, { timeout: 15_000 })
.toBeGreaterThan(0).catch(() => {});
return links;
} finally { await page.goto(source, { waitUntil: "domcontentloaded" }); }
}
export const runtimeDelegationCases = [runtimeDefinition("task-coordination.delegate-with-context", async ctx => {
const prepared = await prepareRuntimeWorkerGate(ctx), root = `/api/companies/${ctx.input.fixtures.company.id}`, marker = `Launch brief ${ctx.input.nonce}`;
const parent = await ctx.input.api.post<Entity>(`${root}/issues`, { title: "Launch", status: "backlog", description: "The CEO owns the launch. The attached human design brief defines the page; implementation should be delegated to Morgan when requested." });
const briefComment = await ctx.input.api.post<Entity>(`/api/issues/${parent.id}/comments`, { body: "Approved design brief for later authorized implementation; no work is assigned yet." });
const brief = `# Launch design\nHeadline: Research without the busywork\nUse semantic HTML, accessible labels and one primary signup action.\nReference: ${marker}\n`;
const uploadTarget = `${root}/issues/${parent.id}/attachments`;
const origin = launchBriefBrowserOrigin(ctx.input.api.baseURL, ctx.input.page.url(), uploadTarget, ctx.input.fixtures.company.id, parent.id);
const uploaded = await ctx.input.page.request.post(uploadTarget, { headers: { Origin: origin }, multipart: { issueCommentId: briefComment.id,
file: { name: "design-brief.md", mimeType: "text/markdown", buffer: Buffer.from(brief) } } });
if (uploaded.status() !== 201) { await uploaded.dispose(); throw new Error("Real human Launch brief upload did not succeed"); }
const attachment = await uploaded.json() as ChatQueryAttachmentReceipt; await uploaded.dispose();
const session = await ctx.input.api.get<{ user: { id: string } }>("/api/auth/get-session");
if (attachment.createdByUserId !== session.user.id || attachment.issueCommentId !== briefComment.id || attachment.sha256 !== hash(Buffer.from(brief))) throw new Error("Launch brief lacks actual human comment/byte provenance");
await ctx.input.api.patch(`/api/issues/${parent.id}`, { assigneeAgentId: ctx.input.fixtures.agent.id });
if ((await ctx.allRuns()).length) throw new Error("Backlog Launch prerequisite unexpectedly admitted work");
await ctx.input.api.saveAgentInstructions(prepared.worker.id, `When an authorized Website task arrives, first run node ${JSON.stringify(prepared.gate.scriptPath)} once and wait for the real brief-release output. Do not replace the script or invent the brief. Then read the actual attached design brief referenced by the task. Implement it as website.html and register that real file as a downloadable artifact work product titled website.html. Save a non-plan delivery document with key website-delivery containing the design's reference and the actual result summary. Complete the same task without creating another task or publishing externally.`);
if (!ctx.input.privateSecrets) throw new Error("Actual worker brief read requires the private probe sink");
const key = await ctx.input.api.postSensitive<{ id: string; token: string }>(`/api/agents/${prepared.worker.id}/keys`, { name: "delegated-launch-brief-read" });
ctx.input.privateSecrets.push(key.token); keys.set(ctx, { ...key, workerId: prepared.worker.id });
await openStandingChat(ctx);
const setupMessage = `This conversation tracks Launch ${parent.identifier ?? parent.id}, which you own. It has the actual approved design brief attached to human comment ${briefComment.id}; Morgan is available for implementation. Remember this work context and acknowledge it here, but do not start, modify, delegate or complete work until my next request.`;
await sendChatMessage(ctx.input.page, setupMessage, chatCatalogNativePhaseRemainingBudget(ctx));
await pollUntil({ label: "Genuine source context acknowledgement before delegation", deadlineAt: chatCatalogNativePhaseDeadline(ctx, Date.now() + 150_000),
load: async () => { await ctx.refreshIssue(); return { source: ctx.issue(), runs: await ctx.allRuns() }; },
accept: v => chatCatalogNativePhaseIsTimely(ctx) && (v.source?.conversationState === "waiting" && v.runs.length === 1 && v.runs[0]?.agentId === ctx.input.fixtures.agent.id && v.runs[0].status === "succeeded") });
return { worker: prepared.worker, gate: prepared.gate, parent: await ctx.input.api.get<Entity>(`/api/issues/${parent.id}`), marker,
humanId: session.user.id, briefCommentId: briefComment.id, briefAttachment: attachment, briefHash: hash(Buffer.from(brief)),
briefDownloadedHash: hash(await ctx.input.api.getAttachmentContent(attachment.id)), sourceId: ctx.issue()!.id };
}, gradeDelegation, {
oracleVersion: "paperclip.chat-catalog.runtime-delegation.v4",
authentication: "human-session", setupHumanMessageCount: 1, expectedHumanMessageCount: 2, expectedRunCount: 4, attemptTimeoutMs: 900_000,
sourceFiles: ["chat-catalog-domains/runtime-delegation.ts", "chat-catalog-domains/runtime-busy-worker.ts", "chat-catalog-domains/runtime-support.ts", "chat-catalog-domains/identity-support.ts", "completion-update-flow.ts", "completion-updates.ts", "registered-artifact.ts", "chat-scenarios/task-coordination.yaml"],
async act(ctx, state) {
const phase = await pollUntil({ label: "Canonical source creates exactly one actual Morgan child with Launch context", deadlineAt: chatCatalogNativePhaseDeadline(ctx, Date.now() + 180_000),
load: async () => { await ctx.refreshIssue(); return { source: ctx.issue(), runs: await ctx.allRuns(), waiting: await readFile((state.gate as { ready: string }).ready, "utf8").catch(() => ""),
tasks: await ctx.input.api.get<Entity[]>(`/api/companies/${ctx.input.fixtures.company.id}/issues?includeAgentCreatedTasks=true&limit=1000`) }; },
accept: v => chatCatalogNativePhaseIsTimely(ctx) && (v.source?.conversationState === "waiting" && v.runs.length === 3 && v.waiting === "waiting"
&& v.runs.some(row => row.agentId === (state.worker as Entity).id && row.status === "running")
&& v.runs.filter(row => row.agentId === ctx.input.fixtures.agent.id).every(row => row.status === "succeeded")
&& v.tasks.filter(row => row.id !== state.sourceId && row.id !== (state.parent as Entity).id).length === 1) });
const task = phase.tasks.find(row => row.id !== state.sourceId && row.id !== (state.parent as Entity).id)!; state.task = task; state.workerWasReallyWaiting = true;
const fromSource = await ctx.input.api.get<Entity[]>(`/api/companies/${ctx.input.fixtures.company.id}/issues?createdFromIssueId=${state.sourceId}&includeAgentCreatedTasks=true&limit=1000`);
state.createdFromSource = fromSource.length === 1 && fromSource[0]?.id === task.id;
const attachments = await ctx.input.api.get<ChatQueryAttachmentReceipt[]>(`/api/issues/${task.id}/attachments`), original = state.briefAttachment as ChatQueryAttachmentReceipt;
const documents = await ctx.input.api.get<Entity[]>(`/api/issues/${task.id}/documents`), bodies: Entity[] = [];
for (const doc of documents) bodies.push(await ctx.input.api.get<Entity>(`/api/issues/${task.id}/documents/${encodeURIComponent(doc.key!)}`));
const context = `${task.description ?? ""}\n${bodies.map(doc => doc.body).join("\n")}`;
// The task page renders bare identifiers (COM-1) as task links, so read what a person sees there too.
const renderedLinks = await renderedDescriptionLinks(ctx, task);
state.renderedDescriptionLinks = renderedLinks;
state.linkedToLaunch = linksOpenTask([...[...context.matchAll(/\[[^\]]*\]\(([^)\s]+)\)/g)].map(match => match[1]!), ...renderedLinks], ctx.input.api.baseURL, state.parent as Entity);
const copied = attachments.find(row => row.sha256 === state.briefHash);
let path = copied?.contentPath ?? (context.includes(original.contentPath) ? original.contentPath : undefined);
state.parentLinkInspectable = task.parentId === (state.parent as Entity).id || state.linkedToLaunch === true;
const key = keys.get(ctx); if (!key || key.workerId !== (state.worker as Entity).id) throw new Error("Actual Morgan brief probe identity is missing");
async function workerRead(path: string) {
const target = new URL(path, ctx.input.api.baseURL);
if (target.origin !== ctx.input.api.baseURL || !/^\/api\/(?:issues|attachments)\//.test(target.pathname)) throw new Error("Worker brief read escaped fixture resources");
return fetch(target, { redirect: "error", signal: AbortSignal.timeout(15_000), headers: { Authorization: `Bearer ${key!.token}` } });
}
const ownRead = await workerRead(`/api/issues/${task.id}`); state.workerTaskReadStatus = ownRead.status;
let workerContext = "";
if (ownRead.ok) { const own = await ownRead.json() as Entity;
if (own.id !== task.id || own.companyId !== ctx.input.fixtures.company.id || own.assigneeAgentId !== key.workerId) throw new Error("Worker task read returned a foreign owner or scope");
workerContext = own.description ?? "";
for (const doc of documents) { const response = await workerRead(`/api/issues/${task.id}/documents/${encodeURIComponent(doc.key!)}`);
if (response.ok) workerContext += `\n${(await response.json() as Entity).body ?? ""}`; else await response.body?.cancel(); }
} else await ownRead.body?.cancel();
state.durableBriefComplete = completeLaunchBrief(workerContext, String(state.marker));
if (path) state.briefContextKind = copied ? "copied-attachment" : "direct-attachment-reference";
else if (state.parentLinkInspectable) {
const parent = await workerRead(`/api/issues/${(state.parent as Entity).id}`); state.workerParentReadStatus = parent.status;
if (parent.ok) { const row = await parent.json() as Entity;
if (row.id !== (state.parent as Entity).id || row.companyId !== ctx.input.fixtures.company.id) throw new Error("Worker parent reference escaped its actual Launch scope");
const receipts = await workerRead(`/api/issues/${row.id}/attachments`);
if (receipts.ok && (await receipts.json() as ChatQueryAttachmentReceipt[]).some(item => item.id === original.id && item.sha256 === original.sha256 && item.issueCommentId === state.briefCommentId)) {
path = original.contentPath; state.briefContextKind = "readable-parent-reference";
} else if (!receipts.ok) await receipts.body?.cancel();
} else await parent.body?.cancel();
}
if (!path && state.durableBriefComplete) state.briefContextKind = "faithful-durable-copy";
state.childHasBrief = !!path || state.durableBriefComplete === true;
if (path) {
const target = new URL(path, ctx.input.api.baseURL); if (target.origin !== ctx.input.api.baseURL || !target.pathname.startsWith("/api/attachments/")) throw new Error("Delegated brief escaped the fixture API");
const response = await fetch(target, { redirect: "error", signal: AbortSignal.timeout(15_000), headers: { Authorization: `Bearer ${key.token}` } });
state.workerBriefReadStatus = response.status;
if (response.ok) { const bytes = new Uint8Array(await response.arrayBuffer()); if (bytes.byteLength > 65_536) throw new Error("Delegated fixture brief size exceeded"); state.workerBriefHash = hash(bytes); }
else await response.body?.cancel();
}
await writeFile((state.gate as { gate: string }).gate, "The human has authorized this delegated page implementation. Read the actual task-bound approved design brief, implement its headline and signup design as website.html, retain its exact reference in the delivery document, and complete this same work task. Do not publish or create further tasks.", "utf8");
// Morgan builds a whole page here, so its execution runs on the case's original attempt clock (as
// runtime-handoffs does); promptness of the return to this conversation is graded separately.
const delivered = await observeCompletionUpdate({ page: ctx.input.page, api: ctx.input.api, sourceId: String(state.sourceId), workerId: task.id,
marker: String(state.marker), fixtureRequest: ctx.scenario.query, originalObservationDeadlineAt: chatCatalogOriginalPhaseDeadline(ctx),
allRuns: () => ctx.allRuns(), evidence: ctx.input.evidence, capture: ctx.input.capture });
const delivery = completionDelivery(delivered); state.delivery = delivery;
state.returnDeliveryLatencyMs = Date.parse(String(delivery.response?.createdAt)) - Date.parse(String(delivered.worker.completedAt));
},
async observe(ctx, state) {
const o = await observeRuntime(ctx, state), parent = await ctx.input.api.get<Entity>(`/api/issues/${(state.parent as Entity).id}`);
state.parentOwnedByCeo = parent.assigneeAgentId === ctx.input.fixtures.agent.id;
const original = state.briefAttachment as ChatQueryAttachmentReceipt, attachment = (await ctx.input.api.get<ChatQueryAttachmentReceipt[]>(`/api/issues/${parent.id}/attachments`)).find(row => row.id === original.id);
state.parentContextRetained = !!attachment && attachment.sha256 === original.sha256 && attachment.issueCommentId === state.briefCommentId;
const task = state.task as Entity | undefined;
if (task) {
state.task = await ctx.input.api.get<Entity>(`/api/issues/${task.id}`);
const workerRun = o.runs.find(row => row.agentId === (state.worker as Entity).id);
if (workerRun) { const [artifact] = await readRegisteredArtifacts(ctx.input.api, task.id, workerRun.id, ["website.html"]);
const facts = artifact?.contentVerified === true ? await launchPageFacts(ctx.input.page, artifact.content) : undefined;
state.websiteFacts = facts;
state.actualWebsiteArtifact = artifact?.contentVerified === true && launchPageReflectsBrief(facts)
&& o.revisions.some(row => row.createdByRunId === workerRun.id && row.createdByAgentId === (state.worker as Entity).id && (row.body ?? "").includes(String(state.marker))
&& o.documents.some(doc => doc.id === row.documentId && doc.latestRevisionId === row.id)); }
}
return o;
},
async verifyResult(ctx, state) {
const source = ctx.input.page.url();
for (const task of [state.parent as Entity, state.task as Entity]) { await ctx.input.page.goto(`/${ctx.input.fixtures.company.issuePrefix}/issues/${task.identifier ?? task.id}`, { waitUntil: "domcontentloaded" });
await ctx.input.page.reload({ waitUntil: "domcontentloaded" }); await expect(ctx.input.page.getByRole("heading", { name: task.title!, exact: true }).first()).toBeVisible(); }
await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" });
},
async cleanup(ctx, state) { try { const key = keys.get(ctx); keys.delete(ctx); if (key) await ctx.input.api.delete(`/api/agents/${key.workerId}/keys/${key.id}`, { allowNotFound: true }); }
finally { await releaseRuntimeWorker(state); } },
})];