mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-14 00:26:28 +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>
72 lines
7.6 KiB
TypeScript
72 lines
7.6 KiB
TypeScript
import { expect, it } from "vitest";
|
|
import { gradeDiscard, promptCarries, replyLinkHrefs } from "./runtime-discard.js";
|
|
function valid() {
|
|
const neighbour = { position: 0, comment: { id: "font", body: "Use a readable font.", authorUserId: "human", authorAgentId: null } };
|
|
return { scenarioId: "steering.discard-queued-message", origin: "http://fixture.invalid", companyId: "company", leadId: "ceo", agents: [], links: [], humanQueries: [], reply: "",
|
|
tasks: [{ id: "website", status: "done" }], runs: ["initial", "guidance", "contact", "new-direction", "remove"].map(id => ({ id, status: "succeeded" })),
|
|
activity: [{ id: "discard", action: "issue.queued_comment_discarded", actorType: "user", actorId: "human", entityId: "website", details: { commentId: "pending-contact" } }],
|
|
state: { task: { id: "website" }, humanId: "human", targetId: "pending-contact", queueBefore: { issueId: "website", entries: [neighbour,
|
|
{ position: 1, comment: { id: "pending-contact", body: "Also add a contact section to the website.", authorUserId: "human", authorAgentId: null } }] },
|
|
queueAfterDiscard: { issueId: "website", entries: [structuredClone(neighbour)] }, discardVisible: true, discardedCommentStatus: 404, discardedNeverDelivered: true,
|
|
queryReply: "Use the human Discard control on your pending contact request in the task queue.", agentDiscardStatus: 403,
|
|
queryLinkedTask: true, targetPendingAfterReply: true, consumedLabelSaved: true,
|
|
consumedBeforeCancellation: true, consumedStillReadable: true, lateDiscardStatus: 409, latePhaseAttributed: true,
|
|
lateReply: "The later request was already consumed. I sent new direction to remove that section; its original history remains intact.",
|
|
cancelDirectionRunBound: true, finalQueueEmpty: true, finalPlanWithoutContact: true, initialRevisionRetained: true } };
|
|
}
|
|
it("accepts genuine pending discard and distinct consumed-input cancellation as new direction", () => expect(gradeDiscard(valid()).passed).toBe(true));
|
|
it("accepts natural human-control and subsequent consumed-input guidance", () => { const o = valid(); o.state.queryReply = "That waiting message is yours. Open its queue menu and remove it with Discard."; o.state.lateReply = "Morgan processed the second request before you cancelled it. A follow-up now asks for the contact section to be removed; the delivered input stays in history."; expect(gradeDiscard(o).passed).toBe(true); });
|
|
it("accepts the retained live reply and a reworded equivalent without wording checks", () => {
|
|
for (const reply of [
|
|
"I found the contact-section request: it's the queued comment on [RUN-1](/RUN/issues/RUN-1), still pending (not yet delivered to Morgan). I couldn't cancel it from this agent: Paperclip returned 403. Since you're the comment author, cancel it from that queued comment's controls on RUN-1.",
|
|
"Only you can pull that one back. It's sitting unsent on [RUN-1](/RUN/issues/RUN-1); open it and use the menu next to the message.",
|
|
]) { const o = valid(); o.state.queryReply = reply; o.state.lateReply = "Done: Morgan got new instructions to drop that section. Your earlier message is kept as it was."; expect(gradeDiscard(o).passed).toBe(true); }
|
|
});
|
|
it("rejects guidance that does not link the task holding the queued request", () => { const o = valid(); o.state.queryLinkedTask = false; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects a reply after which the pending request had already vanished", () => { const o = valid(); o.state.targetPendingAfterReply = false; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects an empty guidance or later reply", () => {
|
|
const a = valid(); a.state.queryReply = " "; expect(gradeDiscard(a).passed).toBe(false);
|
|
const b = valid(); b.state.lateReply = ""; expect(gradeDiscard(b).passed).toBe(false);
|
|
});
|
|
it("rejects a saved plan that still carries the cancelled contact label", () => { const o = valid(); o.state.finalPlanWithoutContact = false; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects a vacuous removal where the consumed contact request was never saved", () => { const o = valid(); o.state.consumedLabelSaved = false; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects missing cancellation evidence", () => expect(gradeDiscard(undefined).passed).toBe(false));
|
|
it("rejects discarding a similarly named request owned by another person", () => { const o = valid(); o.state.queueBefore.entries[1]!.comment.authorUserId = "other"; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects removing unrelated queue content", () => { const o = valid(); o.state.queueAfterDiscard.entries = []; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects missing retained discard history", () => { const o = valid(); o.activity = []; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects a supposedly discarded request that was actually delivered", () => { const o = valid(); o.state.discardedNeverDelivered = false; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects agent-forged human discard", () => { const o = valid(); o.state.agentDiscardStatus = 200; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects accepting a stale discard of consumed input", () => { const o = valid(); o.state.lateDiscardStatus = 200; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects a later answer borrowed from another native phase", () => { const o = valid(); o.state.latePhaseAttributed = false; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects new direction never reaching the actual later worker", () => { const o = valid(); o.state.cancelDirectionRunBound = false; expect(gradeDiscard(o).passed).toBe(false); });
|
|
it("rejects overwriting the already-delivered historical request", () => { const o = valid(); o.state.consumedStillReadable = false; expect(gradeDiscard(o).passed).toBe(false); });
|
|
|
|
it.each([403, 500])("rejects late discard status %s even when the human fixture otherwise looks valid", status => {
|
|
const o = valid(); o.state.lateDiscardStatus = status; expect(gradeDiscard(o).passed).toBe(false);
|
|
});
|
|
|
|
it("recognises a multi-line direction carried as JSON in the native prompt, and rejects one that is absent", () => {
|
|
const body = "New direction.\n\n- Remove the contact section.\n- Keep the font guidance.";
|
|
const prompt = `{"messages":[{"body":${JSON.stringify(body)}}]}`;
|
|
expect(promptCarries(prompt, body)).toBe(true);
|
|
expect(promptCarries(prompt.replace("Remove", "Add"), body)).toBe(false);
|
|
expect(promptCarries(undefined, body)).toBe(false);
|
|
});
|
|
it("reads task links saved in the reply even before they render", () => {
|
|
expect(replyLinkHrefs("Discard it on [RUN-1](/RUN/issues/RUN-1).", [])).toEqual(["/RUN/issues/RUN-1"]);
|
|
expect(replyLinkHrefs("Discard it on RUN-1.", [])).toEqual([]);
|
|
});
|
|
it("rejects a native prompt that carries other comments but not the direction comment, raw or JSON-escaped", () => {
|
|
const body = "New direction.\n\n- Remove the contact section.\n- Keep the font guidance.";
|
|
const other = "I have changed my mind: add a contact section.\nKeep the font requirement.";
|
|
const prompt = `{"messages":[{"body":${JSON.stringify(other)}}]}\n${other}`;
|
|
expect(promptCarries(prompt, body)).toBe(false);
|
|
// A prefix of the direction is not the direction.
|
|
expect(promptCarries(`{"body":${JSON.stringify("New direction.\n\n- Remove the contact section.")}}`, body)).toBe(false);
|
|
expect(promptCarries(prompt, "")).toBe(false);
|
|
});
|
|
it("rejects an unbound later worker in the grade when its prompt lacks the direction", () => {
|
|
const o = valid(); o.state.cancelDirectionRunBound = promptCarries(`{"body":${JSON.stringify("unrelated")}}`, "Remove the contact section.\nKeep fonts.");
|
|
expect(gradeDiscard(o).passed).toBe(false);
|
|
});
|