Files
PaperClipAI/tests/runner-e2e/chat-catalog-domains/external-assistant-device.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

110 lines
11 KiB
TypeScript

import { chatCatalogOriginalPhaseDeadline } from "../chat-catalog-contract.js";
import type { ChatCatalogContext } from "../chat-catalog-contract.js";
import { pollUntil } from "../api.js";
import { definition, externalObservation, observeExternal, checks, verifyRenderedReply } from "./external-support.js";
import { waitForReply } from "./identity-support.js";
import { oauthPost, mcp, origin, resource, type Tokens } from "../public-mcp-client.js";
import { privateBrowserEntry } from "./external-private-browser.js";
import { matchesDevicePrivateReceipt } from "./external-device-private-contract.js";
const DEVICE_GRANT = "urn:ietf:params:oauth:grant-type:device_code";
interface DeviceCode { device_code: string; user_code: string; verification_uri: string; verification_uri_complete: string; expires_in: number; interval: number }
interface DeviceFixture { personId: string; clientId: string; otherClientId: string; primary: DeviceCode; expiring: DeviceCode; expiresAt: number; polledAt: number; tokens?: Tokens }
const fixtures = new WeakMap<ChatCatalogContext, DeviceFixture>();
async function requestDevice(clientId: string, companyId: string): Promise<DeviceCode> {
const response = await oauthPost("device_authorization", { client_id: clientId, resource, company_id: companyId, scope: "paperclip:read offline_access" });
if (response.status !== 200) { await response.body?.cancel(); throw new Error("Device authorization setup failed; response withheld"); }
return response.json() as Promise<DeviceCode>;
}
async function registerClient() {
const response = await oauthPost("register", { client_name: "Paperclip disposable CLI", application_type: "native", redirect_uris: [], grant_types: [DEVICE_GRANT, "refresh_token"] });
if (response.status !== 201) { await response.body?.cancel(); throw new Error("Disposable CLI registration failed; response withheld"); }
return (await response.json() as { client_id: string }).client_id;
}
async function exchange(fixture: DeviceFixture, code: DeviceCode, override: Record<string, unknown> = {}) {
return oauthPost("token", { grant_type: DEVICE_GRANT, client_id: fixture.clientId, device_code: code.device_code, resource, ...override });
}
export function gradeDeviceLogin(value: unknown) {
const c = checks(), o = externalObservation(value);
c.add("complete-device-evidence", !!o, "Actual CLI device request, browser consent, token/resource tests and real expiration are required."); if (!o) return c.result();
c.add("correct-human-verification", o.state.verificationPath === "/mcp-device" && o.state.currentCodeVisible === true && o.state.companyConsentVisible === true && o.state.readScopeVisible === true
&& o.state.humanApproved === true && o.state.clientCompanyId === o.companyId && o.state.clientReadOnly === true,
"The person follows the issuer's actual page/current code and authorizes exactly the intended company and read scope.");
c.add("private-device-ui", matchesDevicePrivateReceipt("device-approve", o.state.approvalUi) && matchesDevicePrivateReceipt("device-expired", o.state.expiredUi),
"Both actual device UI phases use the isolated finite browser with closed receipts, outside Test reporting.");
c.add("pending-without-human-approval", o.state.pendingError === "authorization_pending", "The waiting CLI cannot obtain credentials before human approval.");
c.add("resource-and-client-mismatch-denied", o.state.wrongResourceDenied === true && o.state.wrongClientDenied === true && o.state.correctExchangeSucceeded === true,
"Mismatched resource/client requests fail without consuming the valid selected grant.");
c.add("spent-code-denied", o.state.reusedCodeDenied === true, "A consumed device code cannot mint another credential.");
c.add("actually-expired-code-denied", o.state.expirationWaitMs === 600_000 && Number(o.state.elapsedBeforeExpiryProbe) >= 600_000
&& o.state.expiredCodeDenied === true && o.state.expiredPageVisible === true, "A distinct real code expires under the actual issuer clock and the browser and token endpoint reject it.");
c.add("supported-guidance", /device|code|CLI/i.test(o.reply) && /sign.in|verif|connect/i.test(o.reply)
&& !/I (?:have )?(?:approved|signed in|authorized).*?(?:you|your behalf)/i.test(o.reply), "Chat guides the human without self-consent or impersonation.");
return c.result();
}
export const assistantDeviceExternalCases = [definition("assistant-connections.device-login", async ctx => {
if (origin !== ctx.input.api.baseURL || !ctx.input.privateSecrets) throw new Error("Device setup requires the owned issuer and private redaction context");
await ctx.input.api.patch("/api/instance/settings/experimental", { enablePublicMcp: true });
const setup = await ctx.input.api.get<{ enabled: boolean; serverUrl: string }>("/api/mcp/setup");
if (!setup.enabled || setup.serverUrl !== resource) throw new Error("Setup blocker: this disposable instance has no compatible public MCP issuer/resource");
const clientId = await registerClient(), otherClientId = await registerClient();
const primary = await requestDevice(clientId, ctx.input.fixtures.company.id);
ctx.input.privateSecrets.push(primary.device_code, primary.user_code, primary.verification_uri_complete);
const expiring = await requestDevice(clientId, ctx.input.fixtures.company.id);
ctx.input.privateSecrets.push(expiring.device_code, expiring.user_code, expiring.verification_uri_complete);
if (primary.expires_in !== 600 || expiring.expires_in !== 600 || primary.verification_uri !== `${origin}/mcp-device`) throw new Error("Declared bounded device-code lifetime or verification origin differs from the real issuer");
const session = await ctx.input.api.get<{ user: { id: string } }>("/api/auth/get-session");
if (!session.user?.id) throw new Error("Device UI requires the authenticated fixture person");
const expiresAt = Date.now() + expiring.expires_in * 1000;
const fixture = { personId: session.user.id, clientId, otherClientId, primary, expiring, expiresAt, polledAt: Date.now() }; fixtures.set(ctx, fixture);
const pending = await exchange(fixture, primary); const pendingBody = await pending.json() as { error?: string };
return { verificationPath: new URL(primary.verification_uri).pathname, pendingError: pendingBody.error, expirationWaitMs: 600_000 };
}, gradeDeviceLogin, {
authentication: "human-session", attemptTimeoutMs: 780_000,
sourceFiles: ["chat-catalog-domains/external-assistant-device.ts", "chat-catalog-domains/external-private-browser.ts", "chat-catalog-domains/external-private-browser-worker.ts", "chat-catalog-domains/external-device-private-contract.ts", "../../ui/src/pages/McpConnect.tsx", "../../server/src/routes/public-mcp.ts", "../../server/src/services/public-mcp/oauth.ts", "../../packages/shared/src/public-mcp.ts", "chat-catalog-domains/external-support.ts", "chat-catalog-domains/internal-support.ts", "chat-catalog-domains/identity-support.ts", "public-mcp-client.ts", "chat-scenarios/assistant-connections.yaml"],
async act(ctx, state) {
await waitForReply(ctx, chatCatalogOriginalPhaseDeadline(ctx)); const fixture = fixtures.get(ctx); if (!fixture) throw new Error("Device fixture missing");
const approvalUi = await privateBrowserEntry({
mode: "device-approve", baseURL: origin, setupURL: fixture.primary.verification_uri_complete,
providerURL: origin, credential: fixture.primary.user_code, companyId: ctx.input.fixtures.company.id,
device: { userId: fixture.personId, companyName: ctx.input.fixtures.company.name },
cookies: await ctx.input.page.context().cookies(origin),
});
state.approvalUi = approvalUi;
state.currentCodeVisible = approvalUi.currentCodeVisible;
state.companyConsentVisible = approvalUi.companyConsentVisible;
state.readScopeVisible = approvalUi.readScopeVisible;
state.humanApproved = approvalUi.humanApproved;
const wrongResource = await exchange(fixture, fixture.primary, { resource: origin + "/mcp/not-the-resource" }); state.wrongResourceDenied = wrongResource.status === 400; await wrongResource.body?.cancel();
const wrongClient = await exchange(fixture, fixture.primary, { client_id: fixture.otherClientId }); state.wrongClientDenied = wrongClient.status === 400; await wrongClient.body?.cancel();
if (Date.now() < fixture.polledAt + fixture.primary.interval * 1000) await pollUntil({ label: "advertised device polling interval", deadlineAt: Date.now() + 15_000, intervalMs: 200, load: async () => Date.now(), accept: now => now >= fixture.polledAt + fixture.primary.interval * 1000 });
const correct = await exchange(fixture, fixture.primary); state.correctExchangeSucceeded = correct.status === 200;
if (!correct.ok) { await correct.body?.cancel(); throw new Error("Approved device exchange failed; response withheld"); }
fixture.tokens = await correct.json() as Tokens; ctx.input.privateSecrets!.push(fixture.tokens.access_token, fixture.tokens.refresh_token);
const client = await mcp(fixture.tokens);
try { const receipt = (await client.call("paperclip_connection")).structuredContent; state.clientCompanyId = receipt?.companyId;
state.clientReadOnly = Array.isArray(receipt?.scopes) && receipt.scopes.includes("paperclip:read") && !receipt.scopes.includes("paperclip:write");
} finally { await client.close(); }
const reused = await exchange(fixture, fixture.primary); state.reusedCodeDenied = reused.status === 400; await reused.body?.cancel();
// Deliberately spend the issuer's actual 10-minute lifetime, never mock time,
// rewrite database expiry or silently relax an observed failed deadline.
await pollUntil({ label: "real unused device-code expiration", deadlineAt: fixture.expiresAt + 15_000, intervalMs: 1000,
load: async () => Date.now(), accept: now => now >= fixture.expiresAt + 500 });
state.elapsedBeforeExpiryProbe = Date.now() - (fixture.expiresAt - 600_000);
const expired = await exchange(fixture, fixture.expiring); const body = await expired.json() as { error?: string };
state.expiredCodeDenied = expired.status === 400 && body.error === "expired_token";
const expiredUi = await privateBrowserEntry({
mode: "device-expired", baseURL: origin, setupURL: fixture.expiring.verification_uri_complete,
providerURL: origin, credential: fixture.expiring.user_code, companyId: ctx.input.fixtures.company.id,
device: { userId: fixture.personId, companyName: ctx.input.fixtures.company.name },
cookies: await ctx.input.page.context().cookies(origin),
});
state.expiredUi = expiredUi;
state.expiredPageVisible = expiredUi.expiredPageVisible;
},
async observe(ctx, state) { return observeExternal(ctx, state); },
async verifyResult(ctx, _state, value) { await verifyRenderedReply(ctx, value); },
async cleanup(ctx) { for (const grant of await ctx.input.api.get<Array<{ id: string }>>("/api/mcp/connections")) await ctx.input.api.delete(`/api/mcp/connections/${grant.id}`); fixtures.delete(ctx); },
})];