mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-15 01:52:55 +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>
90 lines
9.7 KiB
TypeScript
90 lines
9.7 KiB
TypeScript
import { createHash } from "node:crypto";
|
|
import { mkdtemp, writeFile, rm } from "node:fs/promises";
|
|
import path from "node:path";
|
|
import { pathToFileURL } from "node:url";
|
|
import { expect } from "@playwright/test";
|
|
import type { IssueAttachment } from "../../../packages/shared/src/types/issue.js";
|
|
import type { ChatCatalogCase, ChatCatalogContext } from "../chat-catalog-contract.js";
|
|
import { pollUntil } from "../api.js";
|
|
import { checks, entity, hasLink, project, type Entity } from "./internal-support.js";
|
|
import { identityDefinition, identityObservation, identityWorker, observeIdentity, openStandingChat } from "./identity-support.js";
|
|
|
|
interface Website { attachment: IssueAttachment; body: string; sha256: string }
|
|
const folders = new WeakMap<ChatCatalogContext, string>();
|
|
async function actualWebsite(ctx: ChatCatalogContext, state: Record<string, unknown>) {
|
|
const o = await observeIdentity(ctx, state); const files: Website[] = [];
|
|
for (const task of o.tasks) {
|
|
const attachments = await ctx.input.api.get<IssueAttachment[]>(`/api/issues/${task.id}/attachments`);
|
|
for (const attachment of attachments.filter(row => row.contentType === "text/html" || /\.html?$/i.test(row.originalFilename ?? "")).slice(0, 3)) {
|
|
if (attachment.byteSize > 1_048_576) continue;
|
|
const bytes = await ctx.input.api.getAttachmentContent(attachment.id);
|
|
files.push({ attachment, body: bytes.toString("utf8"), sha256: createHash("sha256").update(bytes).digest("hex") });
|
|
}
|
|
}
|
|
state.websites = files; return o;
|
|
}
|
|
export function gradeFirstWork(value: unknown) {
|
|
const c = checks(); const o = identityObservation(value); c.add("complete-first-website-work", !!o, "Actual first request, ordinary linked execution, authorized artifact bytes and browser rendering are required."); if (!o) return c.result();
|
|
// The expectation is linked work under the available execution policy, not
|
|
// a particular assignee: the CEO may delegate to Morgan or own the task and
|
|
// run it itself. Either way the task's assignee must deliver the file.
|
|
const work = o.tasks[0]; const worker = entity(o.state, "worker"); const executor = [worker.id, o.leadId].includes(work?.assigneeAgentId ?? "") ? work!.assigneeAgentId! : undefined; const websites = o.state.websites as Website[] | undefined;
|
|
const website = websites?.find(file => file.attachment.id === o.state.reviewedAttachmentId);
|
|
c.add("actual-first-request-context", o.state.beforeSourceHumanComments === 0 && o.state.beforeSourceRuns === 0 && o.humanQueries.length === 1
|
|
&& o.humanQueries[0]?.trim() === "Make me a simple website introducing our business.", "The canonical request is the first real human instruction in this ready company's conversation.");
|
|
c.add("separate-requested-ordinary-work", o.tasks.length === 1 && work?.companyId === o.companyId && !!executor && work.id !== o.state.sourceId && work.parentId == null
|
|
&& work.status === "done" && hasLink(o, "issues", work) && o.state.sourceStillConversation === true
|
|
&& o.activity.some(row => row.action === "issue.created" && row.entityId === work.id && row.agentId === o.leadId), "The first chat creates one linked ordinary completed task; its standing conversation stays separate. Its delivered HTML (graded below), not its model-written title, proves it is the business website.");
|
|
c.add("actual-run-attributed-website", !!work && !!website && website.attachment.issueId === work.id && website.attachment.companyId === o.companyId
|
|
&& website.sha256 === website.attachment.sha256 && !!executor && website.attachment.createdByAgentId === executor
|
|
&& o.runs.some(run => run.id === website.attachment.originatingRunId && run.agentId === executor && run.status === "succeeded"
|
|
&& (run.contextSnapshot?.issueId === work.id || (run as unknown as { nativeIssueId?: string }).nativeIssueId === work.id))
|
|
&& /<html\b/i.test(website.body) && /<body\b/i.test(website.body) && /<h[1-6]\b/i.test(website.body)
|
|
&& /Acorn Notes/i.test(website.body) && /research|science/i.test(website.body), "Authorized exact HTML bytes match the immutable actual upload hash and originating worker run, introducing the real business rather than an empty onboarding task.");
|
|
c.add("actual-review-and-usable-rendering", o.state.visibleArtifactName === true && o.state.renderedHeading === true && o.state.renderedBusinessName === true && o.state.renderedBusinessDescription === true
|
|
&& o.state.renderedBytesMatch === true && o.state.renderedWithoutScriptExecution === true,
|
|
"The real task link exposes its delivered file, and the exact downloaded HTML renders a visible heading, the business name and its description in an isolated browser with scripts disabled. The name may sit in a brand link rather than the hero heading.");
|
|
return c.result();
|
|
}
|
|
export const firstWorkIdentityCases: readonly ChatCatalogCase[] = [identityDefinition("onboarding.first-work", async ctx => {
|
|
const source = await openStandingChat(ctx); const worker = await identityWorker(ctx);
|
|
const p = await project(ctx, "Launch", { description: "Acorn Notes is a research notebook for small science teams. It helps teams capture observations, organize evidence and share research notes. Our business website should introduce these benefits and offer a contact link. No external publishing is requested." });
|
|
const comments = await ctx.input.api.get<Entity[]>(`/api/issues/${source.id}/comments`);
|
|
return { sourceId: source.id, worker, project: p, beforeSourceHumanComments: comments.filter(comment => comment.authorUserId).length, beforeSourceRuns: (await ctx.allRuns()).length };
|
|
}, gradeFirstWork, {
|
|
oracleVersion: "paperclip.chat-catalog.identity.first-work.v3", authentication: "human-session", expectedRunCount: 3, minimumRunCount: 2, attemptTimeoutMs: 600_000,
|
|
sourceFiles: ["chat-catalog-domains/identity-first-work.ts", "chat-catalog-domains/identity-support.ts", "chat-catalog-domains/internal-support.ts", "chat-scenarios/onboarding.yaml"],
|
|
async act(ctx, state) {
|
|
const o = await pollUntil({ label: "first real completed website and delivered HTML", deadlineAt: Date.now() + 360_000,
|
|
load: () => actualWebsite(ctx, state), accept: current => current.tasks.length === 1 && current.tasks[0]?.status === "done" && (state.websites as Website[]).some(file => /Acorn Notes/i.test(file.body))
|
|
&& current.runs.every(run => run.status === "succeeded") && ctx.issue()?.conversationState === "waiting" });
|
|
const work = o.tasks[0]!; const file = (state.websites as Website[]).find(file => /Acorn Notes/i.test(file.body))!; state.reviewedAttachmentId = file.attachment.id;
|
|
const href = o.links.find(link => hasLink({ ...o, links: [link] }, "issues", work)); if (!href) throw new Error("First website has no actual task review link");
|
|
const source = ctx.input.page.url(); await ctx.input.page.locator(`[id^="comment-"] a[href=${JSON.stringify(href)}]`).first().click();
|
|
if (!file.attachment.originalFilename) throw new Error("Delivered website has no visible file name");
|
|
await expect(ctx.input.page.getByText(file.attachment.originalFilename, { exact: true }).first()).toBeVisible(); state.visibleArtifactName = true;
|
|
const privateDir = process.env.PAPERCLIP_RUNNER_E2E_PRIVATE_DIR; if (!privateDir) throw new Error("Attempt-private output directory is required");
|
|
const folder = await mkdtemp(path.join(privateDir, "identity-website-")); folders.set(ctx, folder);
|
|
const filename = path.join(folder, "website.html"); const bytes = await ctx.input.api.getAttachmentContent(file.attachment.id);
|
|
await writeFile(filename, bytes, { mode: 0o600 }); state.renderedBytesMatch = createHash("sha256").update(bytes).digest("hex") === file.sha256;
|
|
const browser = ctx.input.page.context().browser(); if (!browser) throw new Error("Actual website rendering requires a browser");
|
|
const isolated = await browser.newContext({ javaScriptEnabled: false });
|
|
try {
|
|
// No rewrite, fabricated server, or script execution: inspect the exact
|
|
// authorized delivered artifact bytes through the browser's file loader.
|
|
await isolated.route(/^https?:\/\//, route => route.abort()); const page = await isolated.newPage(); await page.goto(pathToFileURL(filename).href);
|
|
// The business name is the setup's literal; where the page places it
|
|
// (hero heading, brand link, title banner) is the worker's design choice.
|
|
await expect(page.getByRole("heading").filter({ visible: true }).first()).toBeVisible(); state.renderedHeading = true;
|
|
await expect(page.locator("body").getByText(/Acorn Notes/i).filter({ visible: true }).first()).toBeVisible(); state.renderedBusinessName = true;
|
|
await expect(page.locator("body").getByText(/research|science/i).filter({ visible: true }).first()).toBeVisible(); state.renderedBusinessDescription = true; state.renderedWithoutScriptExecution = true;
|
|
} finally { await isolated.close(); }
|
|
await ctx.input.evidence("identity-real-first-website.json", { taskId: work.id, attachmentId: file.attachment.id, originatingRunId: file.attachment.originatingRunId,
|
|
sha256: file.sha256, visibleArtifactName: state.visibleArtifactName, renderedHeading: state.renderedHeading, renderedBusinessName: state.renderedBusinessName, renderedBusinessDescription: state.renderedBusinessDescription,
|
|
artifactBytesRewritten: false, scriptsExecuted: false, externallyPublished: false });
|
|
await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" });
|
|
},
|
|
async observe(ctx, state) { const o = await actualWebsite(ctx, state); state.sourceStillConversation = ctx.issue()?.status === "in_review" && ctx.issue()?.assigneeAgentId === ctx.input.fixtures.agent.id && ctx.issue()?.conversationState === "waiting"; return o; },
|
|
async cleanup(ctx) { const folder = folders.get(ctx); if (folder) await rm(folder, { recursive: true, force: true }); folders.delete(ctx); },
|
|
})];
|