mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-16 03:31:44 +01:00
## 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>
295 lines
9.8 KiB
TypeScript
295 lines
9.8 KiB
TypeScript
import { mkdtemp, writeFile, readFile, chmod, rm } from "node:fs/promises";
|
|
import { spawn, type ChildProcess } from "node:child_process";
|
|
import path from "node:path";
|
|
import { tmpdir } from "node:os";
|
|
import { fileURLToPath } from "node:url";
|
|
import { chromium } from "@playwright/test";
|
|
import type { ChatCatalogContext } from "../chat-catalog-contract.js";
|
|
import { signInPublicFixturePerson } from "../public-mcp-flow.js";
|
|
import {
|
|
readValidatedPublicOriginReceipt,
|
|
PUBLIC_ORIGIN_RECEIPT_ENV,
|
|
} from "../chat-public-origin.js";
|
|
import { selectedRuntimeInputs } from "./runtime-private.js";
|
|
import {
|
|
validChatgptInput,
|
|
validChatgptReceipt,
|
|
type ChatgptInput,
|
|
type ChatgptReceipt,
|
|
} from "./external-chatgpt-contract.js";
|
|
interface Custody {
|
|
directory: string;
|
|
child: ChildProcess;
|
|
exited: Promise<number | null>;
|
|
exitCode: number | null | undefined;
|
|
receipt?: ChatgptReceipt;
|
|
}
|
|
const owned = new WeakMap<ChatCatalogContext, Custody>();
|
|
export async function chatgptPublicOrigin(ctx: ChatCatalogContext) {
|
|
const receiptPath = process.env[PUBLIC_ORIGIN_RECEIPT_ENV],
|
|
attemptId = process.env.PAPERCLIP_E2E_ACCOUNTING_ATTEMPT_ID;
|
|
if (!receiptPath || !attemptId)
|
|
throw new Error(
|
|
"Setup blocker: actual named ChatGPT setup requires its selected owned HTTPS-origin receipt",
|
|
);
|
|
return readValidatedPublicOriginReceipt({
|
|
receiptPath,
|
|
attemptId,
|
|
executionId: ctx.input.execution.id,
|
|
backendOrigin: ctx.input.api.baseURL,
|
|
});
|
|
}
|
|
async function inputFor(
|
|
ctx: ChatCatalogContext,
|
|
operation: ChatgptInput["operation"],
|
|
control?: ChatgptInput["control"],
|
|
): Promise<ChatgptInput> {
|
|
const selected = selectedRuntimeInputs(
|
|
ctx,
|
|
"assistant-connections.connect-external-assistant",
|
|
),
|
|
p = selected.values,
|
|
secrets = ctx.input.privateSecrets;
|
|
if (
|
|
!secrets ||
|
|
p.operatorMode !== "interactive-private-ui" ||
|
|
!selected.resourceIds.includes(p.chatgptAuthorizationRef)
|
|
)
|
|
throw new Error(
|
|
"Setup blocker: explicitly selected available operator and authorized synthetic ChatGPT account reference are required; no ambient account/session is inferred",
|
|
);
|
|
secrets.push(p.operatorRef, p.chatgptAuthorizationRef);
|
|
await signInPublicFixturePerson(ctx.input.api, ctx.input.page.context());
|
|
const origin = await chatgptPublicOrigin(ctx),
|
|
person = await ctx.input.api.get<{ user: { id: string } }>(
|
|
"/api/auth/get-session",
|
|
),
|
|
state = await ctx.input.page.context().storageState(),
|
|
host = new URL(origin.publicOrigin).hostname;
|
|
const publicSession = {
|
|
cookies: state.cookies.filter((c) => c.domain.replace(/^\./, "") === host),
|
|
origins: state.origins.filter((o) => o.origin === origin.publicOrigin),
|
|
};
|
|
for (const c of publicSession.cookies) secrets.push(c.value);
|
|
for (const o of publicSession.origins)
|
|
for (const item of o.localStorage) secrets.push(item.value);
|
|
const channel = process.env.PAPERCLIP_PLAYWRIGHT_CHANNEL;
|
|
const input: ChatgptInput = {
|
|
operation,
|
|
executionId: ctx.input.execution.id,
|
|
attemptId: origin.attemptId,
|
|
receiptPath: process.env[PUBLIC_ORIGIN_RECEIPT_ENV]!,
|
|
backendOrigin: ctx.input.api.baseURL,
|
|
publicOrigin: origin.publicOrigin,
|
|
companyId: ctx.input.fixtures.company.id,
|
|
companyName: ctx.input.fixtures.company.name,
|
|
prefix: ctx.input.fixtures.company.issuePrefix ?? "",
|
|
humanId: person.user.id,
|
|
operatorRef: p.operatorRef,
|
|
authorizationRef: p.chatgptAuthorizationRef,
|
|
publicSession,
|
|
control,
|
|
executablePath:
|
|
process.env.PAPERCLIP_RUNNER_E2E_CHROMIUM_EXECUTABLE ??
|
|
(channel ? undefined : chromium.executablePath()),
|
|
channel,
|
|
};
|
|
if (!validChatgptInput(input))
|
|
throw new Error(
|
|
"Setup blocker: named ChatGPT operator/public-person/control contract is invalid",
|
|
);
|
|
return input;
|
|
}
|
|
async function stop(c: Custody) {
|
|
if (c.exitCode === undefined) c.child.kill("SIGTERM");
|
|
await Promise.race([c.exited, new Promise((r) => setTimeout(r, 5000))]);
|
|
if (c.exitCode === undefined) c.child.kill("SIGKILL");
|
|
await c.exited;
|
|
}
|
|
async function stage(c: Custody) {
|
|
try {
|
|
const value = JSON.parse(
|
|
await readFile(path.join(c.directory, "stage.json"), "utf8"),
|
|
) as { stage?: string };
|
|
return [
|
|
"availability",
|
|
"product-setup",
|
|
"provider-app",
|
|
"consent",
|
|
"qualification",
|
|
"awaiting-cleanup",
|
|
"cleanup",
|
|
].includes(value.stage ?? "")
|
|
? value.stage
|
|
: "unknown";
|
|
} catch {
|
|
return "unknown";
|
|
}
|
|
}
|
|
async function start(ctx: ChatCatalogContext, input: ChatgptInput) {
|
|
if (owned.has(ctx))
|
|
throw new Error(
|
|
"A named ChatGPT private child is already owned by this case",
|
|
);
|
|
const directory = await mkdtemp(
|
|
path.join(tmpdir(), "paperclip-chatgpt-operator-"),
|
|
);
|
|
await chmod(directory, 0o700);
|
|
try {
|
|
const filename = path.join(directory, "request.json");
|
|
await writeFile(filename, JSON.stringify(input), {
|
|
mode: 0o600,
|
|
flag: "wx",
|
|
});
|
|
const child = spawn(
|
|
process.execPath,
|
|
[
|
|
"--import",
|
|
fileURLToPath(
|
|
new URL(
|
|
"../../../server/node_modules/tsx/dist/loader.mjs",
|
|
import.meta.url,
|
|
),
|
|
),
|
|
fileURLToPath(new URL("./external-chatgpt-worker.ts", import.meta.url)),
|
|
filename,
|
|
],
|
|
{
|
|
stdio: ["ignore", "ignore", "ignore"],
|
|
env: {
|
|
PATH: process.env.PATH ?? "",
|
|
HOME: directory,
|
|
TMPDIR: directory,
|
|
NODE_ENV: "test",
|
|
DO_NOT_TRACK: "1",
|
|
},
|
|
},
|
|
);
|
|
let resolve!: (code: number | null) => void;
|
|
const exited = new Promise<number | null>((r) => {
|
|
resolve = r;
|
|
});
|
|
const c: Custody = { directory, child, exited, exitCode: undefined };
|
|
owned.set(ctx, c);
|
|
child.once("error", () => {
|
|
c.exitCode = 1;
|
|
resolve(1);
|
|
});
|
|
child.once("exit", (code) => {
|
|
c.exitCode = code;
|
|
resolve(code);
|
|
});
|
|
const deadline =
|
|
Date.now() + (input.operation === "availability" ? 45000 : 900000);
|
|
while (Date.now() < deadline) {
|
|
try {
|
|
const value = JSON.parse(
|
|
await readFile(path.join(directory, "qualified.json"), "utf8"),
|
|
) as ChatgptReceipt;
|
|
if (!validChatgptReceipt(value, input))
|
|
throw new Error(
|
|
"Closed named ChatGPT receipt failed identity/isolation contract",
|
|
);
|
|
c.receipt = value;
|
|
if (input.operation === "availability") {
|
|
const code = await c.exited;
|
|
if (code !== 0)
|
|
throw new Error(
|
|
"Named ChatGPT availability child did not close cleanly",
|
|
);
|
|
owned.delete(ctx);
|
|
await rm(directory, { recursive: true, force: true });
|
|
}
|
|
return value;
|
|
} catch (error) {
|
|
if ((error as NodeJS.ErrnoException).code !== "ENOENT") throw error;
|
|
}
|
|
if (c.exitCode !== undefined)
|
|
throw new Error(
|
|
`Named ChatGPT human operation failed at ${await stage(c)}; private account/OAuth/browser diagnostics withheld`,
|
|
);
|
|
await new Promise((r) => setTimeout(r, 100));
|
|
}
|
|
throw new Error(
|
|
`Setup blocker: required named ChatGPT human operation exceeded its bounded ${await stage(c)} phase`,
|
|
);
|
|
} catch (error) {
|
|
const c = owned.get(ctx);
|
|
owned.delete(ctx);
|
|
if (c) {
|
|
// A failed private operation may still be removing its exact new app.
|
|
// Allow only its finite existing cleanup phase before termination.
|
|
if ((await stage(c)) === "cleanup")
|
|
await Promise.race([
|
|
c.exited,
|
|
new Promise((r) => setTimeout(r, 125000)),
|
|
]);
|
|
await stop(c);
|
|
}
|
|
await rm(directory, { recursive: true, force: true });
|
|
throw error;
|
|
}
|
|
}
|
|
export async function privateChatgptAvailability(ctx: ChatCatalogContext) {
|
|
return start(ctx, await inputFor(ctx, "availability"));
|
|
}
|
|
export async function privateChatgptSetup(
|
|
ctx: ChatCatalogContext,
|
|
control: NonNullable<ChatgptInput["control"]>,
|
|
) {
|
|
return start(ctx, await inputFor(ctx, "setup", control));
|
|
}
|
|
/** Only this fixed cleanup gate is sent to the already-owned child; no generic
|
|
* execute payload, provider token, browser state or arbitrary URL is accepted. */
|
|
export async function cleanupPrivateChatgpt(ctx: ChatCatalogContext) {
|
|
const c = owned.get(ctx);
|
|
owned.delete(ctx);
|
|
if (!c) return;
|
|
try {
|
|
if (!c.receipt?.grantId)
|
|
throw new Error(
|
|
"Actual named ChatGPT setup never reached its closed qualification receipt",
|
|
);
|
|
await writeFile(
|
|
path.join(c.directory, "cleanup.json"),
|
|
JSON.stringify({
|
|
operation: "cleanup",
|
|
attemptId: c.receipt.attemptId,
|
|
grantId: c.receipt.grantId,
|
|
}),
|
|
{ mode: 0o600, flag: "wx" },
|
|
);
|
|
const timeout = new Promise<"timeout">((r) => {
|
|
const t = setTimeout(() => r("timeout"), 135000);
|
|
t.unref();
|
|
});
|
|
const code = await Promise.race([c.exited, timeout]);
|
|
if (code !== 0)
|
|
throw new Error(
|
|
"Named ChatGPT cleanup did not finish; selected operator must inspect the exact newly created test app",
|
|
);
|
|
const receipt = JSON.parse(
|
|
await readFile(path.join(c.directory, "cleanup-complete.json"), "utf8"),
|
|
) as {
|
|
attemptId: string;
|
|
grantId: string;
|
|
providerAppRemoved: boolean;
|
|
privateBrowserClosed: boolean;
|
|
};
|
|
if (
|
|
Object.keys(receipt).sort().join(",") !==
|
|
"attemptId,grantId,privateBrowserClosed,providerAppRemoved" ||
|
|
receipt.attemptId !== c.receipt.attemptId ||
|
|
receipt.grantId !== c.receipt.grantId ||
|
|
receipt.providerAppRemoved !== true ||
|
|
receipt.privateBrowserClosed !== true
|
|
)
|
|
throw new Error(
|
|
"Named ChatGPT cleanup lacks its exact closed actual UI deletion receipt",
|
|
);
|
|
} finally {
|
|
await stop(c);
|
|
await rm(c.directory, { recursive: true, force: true });
|
|
}
|
|
}
|