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

134 lines
14 KiB
TypeScript

import { expect } from "@playwright/test";
import { writeFile } from "node:fs/promises";
import type { IssueQueuedCommentQueue } from "../../../packages/shared/src/types/issue.js";
import { pollUntil } from "../api.js";
import { submitTaskReply } from "../user-actions.js";
import { readRegisteredArtifacts } from "../registered-artifact.js";
import { checks, type Entity } from "./internal-support.js";
import { probeAgentPermission, cleanupIdentityProbes } from "./identity-support.js";
import { busyRuntimeWorker, releaseRuntimeWorker } from "./runtime-busy-worker.js";
import { renderRuntimeArtifact } from "./runtime-rendered-artifact.js";
import { runtimeQueueCases } from "./runtime-queue.js";
import { runtimeDefinition, observeRuntime, runtimeObservation } from "./runtime-support.js";
export function isBlueColor(color: string) {
const parts = color.match(/^rgba?\(\s*(\d+(?:\.\d+)?)\s*[, ]\s*(\d+(?:\.\d+)?)\s*[, ]\s*(\d+(?:\.\d+)?)/i);
if (!parts) return false;
const [, red, green, blue] = parts.map(Number);
const alpha = color.match(/^rgba?\([^)]*[,/]\s*(\d*(?:\.\d+)?)\s*\)$/i);
if (/^rgba\(/i.test(color) && (!alpha || Number(alpha[1]) < 0.25)) return false;
return blue! >= 64 && blue! > red! + 24 && blue! >= green!;
}
export function hasVisibleBlueDesign(elements: Array<{ tag: string; role?: string | null; color: string; background: string }>) {
return elements.some(element => ["BODY", "MAIN", "SECTION", "HEADER", "H1", "H2", "H3", "BUTTON"].includes(element.tag) || element.role === "button"
? isBlueColor(element.background) || element.tag !== "BODY" && isBlueColor(element.color) : false);
}
const direction = "Switch to a blue design while you are working.";
export function gradeActiveSteering(value: unknown) {
const c = checks(), o = runtimeObservation(value);
c.add("actual-native-steering-context", !!o, "A genuine compatible active native run and saved human message are required."); if (!o) return c.result();
const s = o.state, before = s.queueBefore as IssueQueuedCommentQueue, after = s.queueAfterAction as IssueQueuedCommentQueue;
const entry = before?.entries.find(row => row.comment.id === s.targetId);
c.add("saved-owned-message-compatible-active-target", !!entry && entry.comment.body === direction && entry.comment.authorUserId === s.humanId && !entry.comment.authorAgentId
&& before.entries.length === 1 && before.issueId === (s.task as Entity).id && before.targetRunId === s.workerRunId && before.protocol === "paperclip_runner_v1"
&& before.steeringDisposition === "available" && s.workerWasReallyWaiting === true,
"The actual harness advertises steering for the exact saved human-owned message and running task, before the canonical request.");
c.add("explicit-human-control-and-truthful-guidance", s.humanSteerSelected === true && s.agentSteerStatus === 403
&& /steer|steering/i.test(String(s.queryReply)) && !/I(?:'ve| have)? (?:already )?(?:steered|delivered|implemented)|blue (?:design|page) (?:is|has been) (?:ready|complete)/i.test(String(s.queryReply)),
"Chat identifies the supported human Steer control; it cannot forge the action or claim queue admission already changed the active result.");
const accepted = s.steerStatus === 200;
c.add("delivery-distinct-from-queue-with-no-loss", s.steerObservedBranch === (accepted ? "same-run-acknowledged" : "provider-rejected-normal-continuation") && (accepted
? s.acknowledgedSameRun === true && after?.entries.length === 0 && s.steerAuditBound === true && s.idempotentRetryStatus === 200 && s.ordinaryConsumerCount === 0
: ["steering_rejected", "steering_unsupported"].includes(String(s.steerFailureCode)) && s.steerStatus === 409 && s.providerFailureVisible === true && s.failureRetainedMessage === true && after?.entries.some(row => row.comment.id === s.targetId && row.comment.body === direction) === true
&& s.supportedContinuationVisible === true && s.ordinaryConsumerCount === 1 && s.continuationInputBound === true),
"A durable same-run ACK is required for delivery; an actual provider rejection instead retains the exact message and visible failure until one normal continuation consumes it.");
c.add("actual-blue-output-correct-native-owner", s.actualBlueArtifact === true && s.blueRevisionBound === true && s.workerTaskDone === true && s.queueEmptyAtEnd === true
&& o.runs.length === (accepted ? 2 : 3) && o.runs.every(row => row.status === "succeeded") && o.tasks.length === 1,
"Actual delivered HTML and its task revision reflect blue under the correct consuming native run, with no duplicate work or hidden run.");
return c.result();
}
export const runtimeActiveSteeringCases = [runtimeDefinition("steering.steer-active", async ctx => {
const state: Record<string, unknown> = await busyRuntimeWorker(ctx), task = state.task as Entity, source = ctx.input.page.url();
await ctx.input.page.goto(`/${ctx.input.fixtures.company.issuePrefix}/issues/${task.identifier ?? task.id}`, { waitUntil: "domcontentloaded" });
await submitTaskReply(ctx.input.page, direction);
const before = await pollUntil({ label: "Actual native same-turn steering advertised for saved human blue message", deadlineAt: Date.now() + 30_000,
load: () => ctx.input.api.get<IssueQueuedCommentQueue>(`/api/issues/${task.id}/queued-comments`),
accept: q => q.entries.length === 1 && q.entries[0]?.comment.body === direction && q.targetRunId === state.workerRunId && q.protocol === "paperclip_runner_v1" && q.steeringDisposition === "available" });
const session = await ctx.input.api.get<{ user: { id: string } }>("/api/auth/get-session");
if (before.entries[0]?.comment.authorUserId !== session.user.id || before.entries[0].comment.authorAgentId) throw new Error("Steering fixture lacks actual human message ownership");
await expect(ctx.input.page.getByTestId(`task-chat-queued-steer-${before.entries[0].comment.id}`)).toBeEnabled();
await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" });
return { ...state, humanId: session.user.id, queueBefore: before, targetId: before.entries[0].comment.id };
}, gradeActiveSteering, {
authentication: "human-session", expectedRunCount: 3, minimumRunCount: 2, attemptTimeoutMs: 720_000,
setupPolicy: runtimeQueueCases[0]!.setupPolicy, verifySetup: runtimeQueueCases[0]!.verifySetup,
sourceFiles: ["chat-catalog-domains/runtime-active-steering.ts", "chat-catalog-domains/runtime-rendered-artifact.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", "registered-artifact.ts", "chat-scenarios/steering.yaml"],
async act(ctx, state) {
await pollUntil({ label: "Canonical steering guidance while the original native owner remains active", deadlineAt: Date.now() + 150_000,
load: async () => { await ctx.refreshIssue(); return { source: ctx.issue(), runs: await ctx.allRuns() }; },
accept: v => 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") });
state.queryReply = (await observeRuntime(ctx, state)).reply;
const source = ctx.input.page.url(), task = state.task as Entity, endpoint = `/api/issues/${task.id}/queued-comments`, targetId = String(state.targetId);
const before = await ctx.input.api.get<IssueQueuedCommentQueue>(endpoint);
if (before.entries.length !== 1 || before.entries[0]?.comment.id !== targetId || before.targetRunId !== state.workerRunId) throw new Error("Canonical guidance lost or duplicated the saved human direction");
const mutation = { queueId: before.queueId, revision: before.revision, targetRunId: before.targetRunId };
state.agentSteerStatus = await probeAgentPermission(ctx, `${endpoint}/${targetId}/steer`, "POST", mutation);
await ctx.input.page.goto(`/${ctx.input.fixtures.company.issuePrefix}/issues/${task.identifier ?? task.id}`, { waitUntil: "domcontentloaded" });
const response = ctx.input.page.waitForResponse(r => r.request().method() === "POST" && new URL(r.url()).pathname === `${endpoint}/${targetId}/steer`);
await ctx.input.page.getByTestId(`task-chat-queued-steer-${targetId}`).click(); state.humanSteerSelected = true;
const result = await response; state.steerStatus = result.status();
const after = await ctx.input.api.get<IssueQueuedCommentQueue>(endpoint); state.queueAfterAction = after;
if (result.status() === 200) {
state.steerObservedBranch = "same-run-acknowledged";
const run = await ctx.input.api.get<{ resultJson?: { queuedSteeringAcknowledgements?: Record<string, { status?: string; queueId?: string; turnId?: string }> } }>(`/api/heartbeat-runs/${state.workerRunId}`);
const ack = run.resultJson?.queuedSteeringAcknowledgements?.[targetId];
state.acknowledgedSameRun = ack?.status === "acknowledged" && ack.queueId === before.queueId && typeof ack.turnId === "string" && ack.turnId.length > 0;
const retry = await ctx.input.page.request.post(`${endpoint}/${targetId}/steer`, { data: mutation }); state.idempotentRetryStatus = retry.status(); await retry.dispose();
} else {
if (result.status() !== 409) throw new Error("Steering failed outside its supported visible rejection contract");
state.steerObservedBranch = "provider-rejected-normal-continuation";
const body = await result.json() as { code?: string; details?: { code?: string } };
state.steerFailureCode = body.code ?? body.details?.code;
if (!["steering_rejected", "steering_unsupported"].includes(String(state.steerFailureCode))) throw new Error("A non-provider queue conflict does not qualify as a provider steering rejection");
await expect(ctx.input.page.getByText(/steer|temporarily unavailable|timed out/i).filter({ hasNot: ctx.input.page.locator("button") }).first()).toBeVisible();
state.providerFailureVisible = true; state.failureRetainedMessage = after.entries.some(row => row.comment.id === targetId && row.comment.body === direction);
await expect(ctx.input.page.getByTestId(`task-chat-queued-message-${targetId}`)).toBeVisible();
await expect(ctx.input.page.getByTestId("task-chat-composer-input").last()).toBeVisible(); state.supportedContinuationVisible = true;
}
await ctx.input.evidence("runtime-steering-observed-branch.json", { branch: state.steerObservedBranch, failureCode: state.steerFailureCode ?? null, status: state.steerStatus, issueId: task.id, runId: state.workerRunId, commentId: targetId, otherBranchLiveQualified: false });
await ctx.input.capture("actual-steering-delivery", "Selected human steering action and actual native delivery or retained rejection", "runtime-steer-active.png");
await writeFile((state.gate as { gate: string }).gate, "The human brief is ready. Implement a small semantic website.html using the actual design direction you received, with a visible h1 and a labelled signup action. Register the real website.html file as a downloadable artifact work product. Save the actual chosen color and design in a non-plan website-delivery task document, then complete normally. Preserve saved human requirements for normal admission; do not create tasks or publish externally.", "utf8");
const finished = await pollUntil({ label: "Actual selected steering or preserved normal continuation produces its website", deadlineAt: Date.now() + 240_000,
load: async () => ({ runs: await ctx.allRuns(), queue: await ctx.input.api.get<IssueQueuedCommentQueue>(endpoint), task: await ctx.input.api.get<Entity>(`/api/issues/${task.id}`) }),
accept: v => v.task.status === "done" && v.queue.entries.length === 0 && v.runs.length === (state.steerStatus === 200 ? 2 : 3) && v.runs.every(row => row.status === "succeeded"),
reject: v => v.runs.length > 3 || v.runs.some(row => ["failed", "cancelled", "timed_out"].includes(row.status)) ? "Steered work failed or admitted an undeclared execution" : undefined });
const successors = finished.runs.filter(row => row.agentId === (state.worker as Entity).id && row.id !== state.workerRunId);
state.ordinaryConsumerCount = successors.length; state.outputRunId = state.steerStatus === 200 ? state.workerRunId : successors[0]?.id;
if (successors.length === 1) {
const full = await ctx.input.api.get<{ contextSnapshot?: { wakeCommentIds?: string[] }; runnerProfileJson?: { nativeExecutionInput?: { task?: { prompt?: string } } } }>(`/api/heartbeat-runs/${successors[0]!.id}`);
state.continuationInputBound = full.contextSnapshot?.wakeCommentIds?.filter(id => id === targetId).length === 1 && full.runnerProfileJson?.nativeExecutionInput?.task?.prompt?.includes(direction) === true;
}
state.workerTaskDone = finished.task.status === "done"; state.queueEmptyAtEnd = finished.queue.entries.length === 0;
await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" });
},
async observe(ctx, state) {
const o = await observeRuntime(ctx, state), task = state.task as Entity;
state.steerAuditBound = o.activity.some(row => row.action === "issue.queued_comment_steered" && row.entityId === task.id && row.actorId === state.humanId
&& row.details?.commentId === state.targetId && row.details?.targetRunId === state.workerRunId && row.details?.duplicate === false);
if (state.outputRunId) {
const [artifact] = await readRegisteredArtifacts(ctx.input.api, task.id, String(state.outputRunId), ["website.html"]);
if (artifact?.contentVerified) {
state.actualBlueArtifact = hasVisibleBlueDesign(await renderRuntimeArtifact(ctx, artifact.content));
}
state.blueRevisionBound = o.revisions.some(row => row.createdByRunId === state.outputRunId && row.createdByAgentId === (state.worker as Entity).id && /blue/i.test(row.body ?? "")
&& o.documents.some(doc => doc.id === row.documentId && doc.latestRevisionId === row.id));
}
return o;
},
async verifyResult(ctx, state) { const source = ctx.input.page.url(), task = 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 expect(ctx.input.page.getByTestId("task-chat-queued-messages")).toHaveCount(0); await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" }); },
async cleanup(ctx, state) { try { await cleanupIdentityProbes(ctx); } finally { await releaseRuntimeWorker(state); } },
})];