Files
PaperClipAI/tests/runner-e2e/chat-catalog-domains/runtime-recovery.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

118 lines
10 KiB
TypeScript

import { chatCatalogNativePhaseDeadline, chatCatalogNativePhaseIsTimely, chatCatalogNativePhaseRemainingBudget } from "../chat-catalog-contract.js";
import { expect, type Download } from "@playwright/test";
import { readFile, realpath } from "node:fs/promises";
import path from "node:path";
import { pollUntil } from "../api.js";
import { sendChatMessage } from "../chat-flow.js";
import { readRegisteredArtifacts } from "../registered-artifact.js";
import { checks, type Entity } from "./internal-support.js";
import { runtimeDefinition, runtimeObservation, observeRuntime } from "./runtime-support.js";
import type { ChatCatalogContext } from "../chat-catalog-contract.js";
type Artifact = Awaited<ReturnType<typeof readRegisteredArtifacts>>[number];
async function settled(ctx: ChatCatalogContext, count: number) {
return pollUntil({ label: `Controller continuity phase ${count}`, deadlineAt: chatCatalogNativePhaseDeadline(ctx, Date.now() + 180_000),
load: async () => { await ctx.refreshIssue(); return { issue: ctx.issue(), runs: await ctx.allRuns() }; },
accept: ({ issue, runs }) => chatCatalogNativePhaseIsTimely(ctx) && (issue?.conversationState === "waiting" && runs.length === count && runs.every(run => run.status === "succeeded")),
reject: ({ runs }) => runs.some(run => ["failed", "cancelled", "timed_out"].includes(run.status)) ? "A continuity provider turn failed" : undefined });
}
export async function currentControllerFile(ctx: ChatCatalogContext, name: string) {
const root = process.env.PAPERCLIP_RUNNER_E2E_TEMP_ROOT; if (!root) throw new Error("Continuity inspection requires the isolated attempt root");
await ctx.refreshIssue(); const source = ctx.issue();
if (!source) throw new Error("Continuity file inspection requires its actual source");
const issue = await ctx.input.api.get<{ companyId: string; executionWorkspaceId: string | null }>(`/api/issues/${source.id}`);
const { workspace } = await ctx.input.api.get<{ workspace: { id: string; companyId: string; cwd: string } }>(`/api/issues/${source.id}/workspace`);
if (issue.companyId !== ctx.input.fixtures.company.id || workspace.companyId !== issue.companyId
|| workspace.id !== issue.executionWorkspaceId) throw new Error("Continuity workspace is not the actual same-company source binding");
const home = await realpath(workspace.cwd);
const isolated = await realpath(root);
if (!home.startsWith(`${isolated}${path.sep}`) || path.basename(name) !== name) throw new Error("Continuity file escapes its owned fixture directory");
const filename = await realpath(path.join(home, name));
if (path.dirname(filename) !== home) throw new Error("Continuity file escapes its actual source binding");
return readFile(filename, "utf8");
}
export function gradeControllerRestart(value: unknown) {
const c = checks(); const o = runtimeObservation(value);
c.add("actual-restart-continuity-evidence", !!o, "Native prior work, real controller restart, retained bytes and run-attributed final artifact are required."); if (!o) return c.result();
const before = o.state.initialArtifact as Artifact | undefined; const final = o.state.finalArtifact as Artifact | undefined;
c.add("real-prior-file-survives-restart", !!before?.contentVerified && typeof o.state.fileBeforeRestart === "string"
&& o.state.fileBeforeRestart === before.content && o.state.fileAfterRestart === o.state.fileBeforeRestart
&& o.state.restarted === true, "The actual provider-written file bytes survive the supported controller restart unchanged.");
c.add("prior-instruction-and-content-retained", !!final?.contentVerified && final.content.includes(String(o.state.headerMarker))
&& final.content.includes("hello@acorn.example") && /contact/i.test(final.content)
&& !before?.content.includes("hello@acorn.example") && o.state.finalFile === final.content,
"The final saved website retains the unpredictable original header and follows the prior persisted contact-section instruction.");
c.add("one-eligible-continuation", o.tasks.length === 0 && o.runs.length === 2 && o.runs.every(run => run.agentId === o.leadId && run.status === "succeeded")
&& o.runs.filter(run => run.id !== o.state.initialRunId).length === 1,
"One ordinary continuation follows the explicit user request without duplicate execution tasks or replayed side effects.");
c.add("source-and-history-retained", o.state.sourceIssueBefore === o.state.sourceIssueAfter && o.state.priorHistoryRetained === true,
"The same conversation and genuine prior messages survive the controller restart.");
c.add("actual-final-download-resource", !!final?.contentVerified && typeof final.attachmentId === "string" && !!final.attachmentId
&& typeof final.workProductId === "string" && !!final.workProductId,
"The final work has a genuine run-attributed downloadable attachment and registered work product; its UI is verified separately.");
return c.result();
}
export const runtimeRecoveryCases = [runtimeDefinition("recovery.controller-restart", async ctx => {
if (ctx.input.execution.environment.id !== "local") throw new Error("Owned-file controller continuity fixture requires the actual local runtime");
const headerMarker = `Retained Acorn header ${ctx.input.nonce}`; const filename = `website-${ctx.input.nonce}.html`;
await sendChatMessage(ctx.input.page,
`Start a small website draft in ${filename} in your current working directory, with a single header containing exactly ${JSON.stringify(headerMarker)}. Register that real file as a downloadable artifact work product titled ${filename}. This is a draft: do not add the contact section yet. Remember this instruction for when I ask to continue: keep the existing header and file, then add a contact section containing hello@acorn.example and register the final file as a new artifact from that continuation. Do this within our current conversation; no project, delegated task or external action is needed.`, chatCatalogNativePhaseRemainingBudget(ctx));
const first = await settled(ctx, 1); const source = ctx.issue()!;
const [initialArtifact] = await readRegisteredArtifacts(ctx.input.api, source.id, first.runs[0]!.id, [filename]);
const fileBeforeRestart = await currentControllerFile(ctx, filename);
if (fileBeforeRestart !== initialArtifact!.content) throw new Error("Initial downloadable artifact does not match the actual saved draft file");
const priorHistory = await ctx.input.api.get<Entity[]>(`/api/issues/${source.id}/comments?order=asc`);
const url = ctx.input.page.url();
await ctx.input.restart();
await ctx.input.page.goto(url, { waitUntil: "commit", timeout: 60_000 });
await expect(ctx.input.page.getByTestId("task-chat-composer-input")).toBeVisible({ timeout: 60_000 });
await ctx.refreshIssue();
const afterHistory = await ctx.input.api.get<Entity[]>(`/api/issues/${source.id}/comments?order=asc`);
return { headerMarker, filename, initialArtifact, initialRunId: first.runs[0]!.id, fileBeforeRestart,
fileAfterRestart: await currentControllerFile(ctx, filename), restarted: true, sourceIssueBefore: source.id, sourceIssueAfter: ctx.issue()?.id,
priorHistoryRetained: priorHistory.every(prior => afterHistory.some(row => row.id === prior.id && row.body === prior.body && row.createdByRunId === prior.createdByRunId)) };
}, gradeControllerRestart, {
setupHumanMessageCount: 1, expectedHumanMessageCount: 2, expectedRunCount: 2, attemptTimeoutMs: 600_000,
async observe(ctx, state) {
const runs = await ctx.allRuns(); const source = ctx.issue();
const final = runs.find(run => run.id !== state.initialRunId && run.status === "succeeded");
if (source && final) {
const [artifact] = await readRegisteredArtifacts(ctx.input.api, source.id, final.id, [String(state.filename)]);
state.finalArtifact = artifact; state.finalFile = await currentControllerFile(ctx, String(state.filename));
}
return observeRuntime(ctx, state);
},
async verifyResult(ctx, state, value) {
const o = runtimeObservation(value); const artifact = state.finalArtifact as Artifact;
if (!o || !artifact) throw new Error("No actual final continuation artifact");
const href = o.links.find(link => { const url = new URL(link, o.origin); return url.origin === new URL(o.origin).origin
&& url.pathname === `/api/attachments/${artifact.attachmentId}/content`; });
// HTML attachments use the production download policy. Verify the actual
// human result control and download, not merely an API registration row.
// Attachment links open the in-app preview panel; its visible download control must then
// deliver the same file. A direct download also counts.
const page = ctx.input.page; const name = String(state.filename);
let download: Download | null = null;
if (href) {
const direct = page.waitForEvent("download", { timeout: 30_000 }).then(row => ({ download: row })).catch(() => null);
await page.locator(`[id^="comment-"] a[href=${JSON.stringify(href)}]`).last().click();
const control = page.locator("header").filter({ has: page.getByRole("heading", { name, exact: true }) })
.getByRole("link", { name: `Download ${name}`, exact: true }).last();
const opened = await Promise.race([direct, control.waitFor({ state: "visible", timeout: 30_000 }).then(() => ({ download: null }))]);
download = opened?.download ?? null;
if (!download) {
await expect(control).toHaveAttribute("href", new RegExp(`/api/attachments/${artifact.attachmentId}/content\\?download=1$`));
[download] = await Promise.all([page.waitForEvent("download"), control.click()]);
}
} else {
const downloadPromise = page.waitForEvent("download");
await page.goto(new URL(`/api/attachments/${artifact.attachmentId}/content?download=1`, o.origin).href).catch(error => {
if (!/Download is starting|net::ERR_ABORTED/i.test(String(error))) throw error;
});
download = await downloadPromise;
}
if (download.suggestedFilename() !== state.filename) throw new Error("The visible result downloaded a different file");
await ctx.input.evidence("runtime-controller-final-download.json", { attachmentId: artifact.attachmentId, filename: download.suggestedFilename(), sha256: artifact.sha256 });
},
})];