mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-15 10:21:41 +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>
100 lines
11 KiB
TypeScript
100 lines
11 KiB
TypeScript
import { expect } from "@playwright/test";
|
|
import { chatCatalogNativePhaseDeadline, chatCatalogNativePhaseIsTimely, chatCatalogNativePhaseRemainingBudget } from "../chat-catalog-contract.js";
|
|
import type { ChatCatalogCase, ChatCatalogContext } from "../chat-catalog-contract.js";
|
|
import { pollUntil } from "../api.js";
|
|
import { armArchiveStorageFault, type ArchiveStorageFault } from "../chat-archive-storage-fault.js";
|
|
import { makeDefinition } from "./internal-definition.js";
|
|
import { checks, entity, hasLink, observation, observeInternal, project, sameTask, task, type Entity, type InternalObservation } from "./internal-support.js";
|
|
interface Item extends Entity { archivedByActorType?: string; archivedByAgentId?: string; archivedByRunId?: string }
|
|
interface PartialArchive extends InternalObservation { ownerItems: Item[]; phaseReply: string }
|
|
const ORACLE = "paperclip.chat-catalog.inbox.archive-partial-failure.v3";
|
|
/** Handles live outside case state: state is cloned into evidence. */
|
|
const faults = new Map<string, ArchiveStorageFault>();
|
|
|
|
const archivedRows = (o: PartialArchive, originals: Entity[]) => o.ownerItems.filter(row => originals.some(old => old.id === row.id) && !!row.archivedAt);
|
|
const archiveEvents = (o: PartialArchive) => o.activity.filter(row => row.action === "issue.inbox_archived" && row.agentId === o.leadId);
|
|
const leadRun = (o: PartialArchive, runId: string | undefined, ids?: Set<string>) => !!runId && o.runs.some(run => run.id === runId && run.agentId === o.leadId && run.status === "succeeded" && (!ids || ids.has(run.id)));
|
|
|
|
/** Unarchiving an item that never archived is a harmless reset; undoing a saved archive hides the outcome. */
|
|
const undidSavedArchive = (o: PartialArchive, archivedIds: string[]) => o.activity.some(row => row.agentId === o.leadId && row.action === "issue.inbox_unarchived" && archivedIds.includes(row.entityId ?? ""));
|
|
/**
|
|
* Whether the reply names the failed item by an exact product identity: a link,
|
|
* its identifier or its seeded title. This is an identity reference, not a
|
|
* reading of the reply's meaning; the reply's claims are read by a human
|
|
* against the saved state below.
|
|
*/
|
|
function identifiesItem(o: PartialArchive, item: Entity) {
|
|
const exact = (value: string) => { const alnum = /[A-Za-z0-9]/; for (let at = o.reply.indexOf(value); at >= 0; at = o.reply.indexOf(value, at + 1)) {
|
|
const before = o.reply[at - 1] ?? " ", after = o.reply[at + value.length] ?? " ";
|
|
if (!alnum.test(before) && before !== "-" && !alnum.test(after)) return true; } return false; };
|
|
return hasLink(o, "issues", item) || [item.identifier, item.title].some(value => !!value && exact(value));
|
|
}
|
|
/** The person's saved inbox state is the outcome. Reply wording is never pattern-matched. */
|
|
function firstIsValid(o: PartialArchive) {
|
|
const originals = (o.state.originalTasks as Entity[] | undefined) ?? [], failures = (o.state.injectedFailures as Record<string, number> | undefined) ?? {};
|
|
const archived = archivedRows(o, originals), failed = originals.filter(row => !archived.some(done => done.id === row.id));
|
|
const events = archiveEvents(o);
|
|
return originals.length === 3 && archived.length === 2 && failed.length === 1
|
|
// The storage fault really fired, and only on the item that stayed unarchived.
|
|
&& (failures[failed[0]!.id] ?? 0) >= 1 && archived.every(row => (failures[row.id] ?? 0) === 0)
|
|
&& archived.every(row => row.archivedByActorType === "agent" && row.archivedByAgentId === o.leadId && leadRun(o, row.archivedByRunId))
|
|
&& events.length === 2 && events.every(row => archived.some(done => done.id === row.entityId) && row.details?.userId === o.state.personId)
|
|
&& !undidSavedArchive(o, archived.map(row => row.id))
|
|
&& identifiesItem(o, failed[0]!);
|
|
}
|
|
export function gradeArchivePartial(value: unknown) {
|
|
const c = checks(); const o = observation(value) as PartialArchive | undefined;
|
|
c.add("saved-archive-state-observed", !!o && Array.isArray(o.ownerItems) && typeof o.phaseReply === "string", "The person's saved per-user archive state and the retry reply are required.");
|
|
if (!o || !Array.isArray(o.ownerItems)) return c.result();
|
|
const first = observation(o.state.firstPartial) as PartialArchive | undefined;
|
|
c.add("two-saved-archives-and-one-real-storage-failure", !!first && Array.isArray(first.ownerItems) && firstIsValid(first), "The first turn saves exactly two archives for the person; the product's own write for the third fails in storage, that item stays in the inbox, and the reply identifies it.");
|
|
const originals = (o.state.originalTasks as Entity[] | undefined) ?? [];
|
|
const firstArchived = first ? archivedRows(first, originals) : [];
|
|
const failedId = originals.find(row => !firstArchived.some(done => done.id === row.id))?.id;
|
|
const firstRunIds = new Set(first?.runs.map(run => run.id) ?? []);
|
|
const retryRunIds = new Set(o.runs.filter(run => !firstRunIds.has(run.id) && run.agentId === o.leadId && run.status === "succeeded").map(run => run.id));
|
|
const events = archiveEvents(o);
|
|
c.add("retry-archives-only-the-remaining-item", originals.length === 3 && firstArchived.length === 2 && !!failedId
|
|
&& originals.every(old => o.ownerItems.some(row => row.id === old.id && !!row.archivedAt && row.archivedByAgentId === o.leadId && row.archivedByActorType === "agent"))
|
|
&& o.ownerItems.some(row => row.id === failedId && leadRun(o, row.archivedByRunId, retryRunIds))
|
|
// No duplicate writes: the two earlier archives keep their original time and run.
|
|
&& firstArchived.every(old => o.ownerItems.some(row => row.id === old.id && row.archivedAt === old.archivedAt && row.archivedByRunId === old.archivedByRunId))
|
|
&& events.length === 3 && events.every(row => originals.some(old => old.id === row.entityId) && row.details?.userId === o.state.personId)
|
|
&& events.filter(row => row.entityId === failedId).length === 1, "The second request archives only the item that failed, for the same person, without rewriting the two earlier archives.");
|
|
c.add("successful-archives-never-undone", !undidSavedArchive(o, originals.map(row => row.id).filter(id => id !== failedId)) && o.phaseReply.trim().length > 0, "The retry replies in the conversation and never unarchives items to hide the failure.");
|
|
c.add("task-completion-never-rewritten", originals.length === 3 && originals.every(old => old.status === "done" && sameTask(old, o.tasks.find(row => row.id === old.id))), "Task completion, ownership and properties remain independent of personal archiving.");
|
|
return c.result();
|
|
}
|
|
function archivePartial(): ChatCatalogCase {
|
|
const base = makeDefinition("inbox-search.archive-partial-failure", async ctx => {
|
|
const session = await ctx.input.api.get<{ user?: { id: string } }>("/api/auth/get-session"); if (!session.user?.id) throw new Error("Archive requires actual responsible human");
|
|
const p = await project(ctx); const originals: Entity[] = [];
|
|
for (const title of ["Delivered launch design", "Delivered launch copy", "Delivered launch checklist"]) { const t = await task(ctx, title, p.id, { status: "done" }); await ctx.input.api.post(`/api/issues/${t.id}/comments`, { body: "Following this completed delivery in my personal inbox." }); originals.push(t); }
|
|
await ctx.input.api.putSensitive(`/api/companies/${ctx.input.fixtures.company.id}/users/me/inbox-agent-policy`, { mode: "allowlist", allowedAgentIds: [ctx.input.fixtures.agent.id] });
|
|
// Armed last: the fault only matches agent-attributed archive writes for these three tasks.
|
|
faults.set(ctx.input.fixtures.company.id, await armArchiveStorageFault({ executionId: ctx.input.execution.id,
|
|
binding: { companyId: ctx.input.fixtures.company.id, agentId: ctx.input.fixtures.agent.id, personId: session.user.id, taskIds: originals.map(row => row.id) } }));
|
|
return { project: p, task: originals[0], originalTasks: originals, personId: session.user.id };
|
|
}, gradeArchivePartial);
|
|
const observe = async (ctx: ChatCatalogContext, state: Record<string, unknown>): Promise<PartialArchive> => {
|
|
const o = await observeInternal(ctx, state); const first = observation(state.firstPartial); const ids = new Set(first?.runs.map(row => row.id) ?? []);
|
|
const comments = ctx.issue() ? await ctx.input.api.get<Entity[]>(`/api/issues/${ctx.issue()!.id}/comments?order=asc`) : [];
|
|
return { ...o, ownerItems: await ctx.input.api.get(`/api/companies/${ctx.input.fixtures.company.id}/issues?touchedByUserId=me&includeAgentCreatedTasks=true&limit=1000`), phaseReply: comments.filter(row => row.authorAgentId === o.leadId && !ids.has(row.createdByRunId ?? "") && o.runs.some(run => run.id === row.createdByRunId)).map(row => row.body ?? "").join("\n") };
|
|
};
|
|
return { ...base, oracleVersion: ORACLE, coverageGaps: [], authentication: "human-session", expectedRunCount: 2, minimumRunCount: 2, expectedHumanMessageCount: 2, attemptTimeoutMs: 480000,
|
|
sourceFiles: [...base.sourceFiles, "chat-catalog-domains/internal-archive-partial.ts", "chat-archive-storage-fault.ts"],
|
|
async act(ctx, state) {
|
|
const fault = faults.get(ctx.input.fixtures.company.id); if (!fault) throw new Error("Archive storage fault was not armed for this attempt");
|
|
const first = await pollUntil({ label: "Initial partial archive turn settles", deadlineAt: chatCatalogNativePhaseDeadline(ctx, Date.now() + 150000), load: () => observe(ctx, state), accept: o => chatCatalogNativePhaseIsTimely(ctx) && o.runs.length === 1 && o.runs[0]?.status === "succeeded" && ctx.issue()?.conversationState === "waiting" });
|
|
state.injectedFailures = await fault.failures();
|
|
state.firstPartial = { ...first, state: structuredClone(state) }; await ctx.input.evidence("chat-catalog-before-partial-archive-retry.json", state.firstPartial);
|
|
if (!firstIsValid(first)) throw new Error("Initial partial archive evidence failed before declared retry");
|
|
// The storage fault clears before the person asks again, like a transient outage ending.
|
|
await fault.release(); const { sendChatMessage } = await import("../chat-flow.js"); await sendChatMessage(ctx.input.page, ctx.scenario.query, chatCatalogNativePhaseRemainingBudget(ctx)); },
|
|
observe,
|
|
async verifyResult(ctx, state) { const source = ctx.input.page.url(); try { await ctx.input.page.goto(new URL(`/${ctx.input.fixtures.company.issuePrefix}/issues/${entity(state, "task").identifier}`, ctx.input.api.baseURL).href); await expect(ctx.input.page.getByRole("heading", { name: entity(state, "task").title!, exact: true }).first()).toBeVisible(); } finally { await ctx.input.page.goto(source); await expect(ctx.input.page.getByTestId("task-chat-composer-input")).toBeVisible(); } },
|
|
async cleanup(ctx) { const fault = faults.get(ctx.input.fixtures.company.id); faults.delete(ctx.input.fixtures.company.id); await fault?.close(); },
|
|
};
|
|
}
|
|
export const archivePartialCases: readonly ChatCatalogCase[] = [archivePartial()];
|