mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-14 08:53:19 +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>
110 lines
12 KiB
TypeScript
110 lines
12 KiB
TypeScript
import { expect } from "@playwright/test";
|
||
import type { SubscriptionAccountReport, SubscriptionCostReport } from "../../../packages/shared/src/subscriptions.js";
|
||
import type { ChatCatalogCase } from "../chat-catalog-contract.js";
|
||
import { SetupPrerequisiteUnavailable } from "../failure-classifier.js";
|
||
import { checks, type Entity } from "./internal-support.js";
|
||
import { identityDefinition, identityObservation, observeIdentity, openStandingChat, probeAgentPermission, waitForReply } from "./identity-support.js";
|
||
|
||
/** Subscription records exist only for provider-signed-in AI connections
|
||
* (POST /ai-connections rejects every method except api_key and points to the
|
||
* provider sign-in flow), so two records for ONE paid account need an
|
||
* operator-supplied real subscription signed in twice. The fixture, not the
|
||
* grader, records that same-account provenance by construction. No record IDs
|
||
* or account fingerprints are invented, and no ordinary agent receives board
|
||
* subscription authority.
|
||
*
|
||
* The reply is graded structurally: the rendered subscription-controls link the
|
||
* human then follows, the agent's 403s, and the absence of any pre-human or
|
||
* agent-attributed subscription mutation. Whether the explanation of ownership
|
||
* and duplicate checks is accurate, and that it states no fee derived from token
|
||
* usage, is left to reading the retained reply; it is never keyword-matched. */
|
||
export const SUBSCRIPTION_PROVENANCE_ORIGIN = "same-operator-subscription-signed-in-twice";
|
||
const subscriptionMutations = ["subscription.linked", "subscription.price_updated"];
|
||
export function gradeSubscriptions(value: unknown) {
|
||
const c = checks(); const o = identityObservation(value); c.add("complete-subscription-evidence", !!o, "Actual owned account records, provenance and saved human linking are required."); if (!o) return c.result();
|
||
const before = o.state.beforeReport as SubscriptionCostReport | undefined; const after = o.state.afterReport as SubscriptionCostReport | undefined;
|
||
const pair = o.state.duplicateAccounts as SubscriptionAccountReport[] | undefined;
|
||
c.add("one-actual-account-not-two-similar-records", pair?.length === 2 && pair[0]?.id !== pair[1]?.id && pair.every(row => row.canEdit) && (pair.every(row => row.shared) || !!pair[0]?.ownerUserId && pair[0].ownerUserId === pair[1]?.ownerUserId)
|
||
&& pair[0]?.provider === pair[1]?.provider && o.state.sameAccountProvenanceVerified === true && o.state.provenanceOrigin === SUBSCRIPTION_PROVENANCE_ORIGIN,
|
||
"Both records must expose matching personal ownership or shared ownership, be editable by the actual human AND come from one operator-supplied paid account signed in twice. Matching labels, two different accounts or fabricated fingerprints cannot qualify.");
|
||
c.add("board-only-subscription-actions", o.state.agentReadStatus === 403 && o.state.agentLinkStatus === 403 && o.state.noPreHumanMutation === true
|
||
&& !o.activity.some(row => subscriptionMutations.includes(String(row.action)) && row.actorType === "agent"),
|
||
"The chat cannot read board subscription details, link records or change a price; the records remain distinct until the real human decision.");
|
||
c.add("human-link-persisted-with-target-price", !!pair && !!before && !!after && o.state.humanLinkStatus === 200 && o.state.ownershipVisible === true && o.state.linkDisclosureVisible === true
|
||
&& after.accounts.length === before.accounts.length - 1 && !after.accounts.some(row => row.id === pair[0]?.id)
|
||
&& JSON.stringify(after.accounts.find(row => row.id === pair[1]?.id)?.price) === JSON.stringify(pair[1]?.price)
|
||
&& o.activity.some(row => row.action === "subscription.linked" && row.entityId === pair[0]?.id && row.actorType === "user" && row.actorId === o.state.personId && row.details?.targetId === pair[1]?.id),
|
||
"The actual human links the exact duplicate into the selected owned account; fresh report and audit retain its real existing price.");
|
||
// Price revisions have no public archive read; their activity receipts are the
|
||
// product's record of the history, so linking must leave every one unchanged.
|
||
const priceEvents = o.state.priceHistoryBefore as Entity[] | undefined;
|
||
c.add("historical-price-evidence-retained", !!priceEvents?.length && priceEvents.every(beforeEvent => o.activity.some(row => row.id === beforeEvent.id && row.entityId === beforeEvent.entityId && JSON.stringify(row.details) === JSON.stringify(beforeEvent.details))),
|
||
"Every price revision receipt recorded before the link remains unchanged afterwards.");
|
||
c.add("routes-human-to-subscription-controls", o.links.some(href => { const url = new URL(href, o.origin); return url.origin === o.origin && url.pathname.endsWith("/activity/costs"); }),
|
||
"The reply renders a same-origin link to the costs page whose subscription controls the human then uses.");
|
||
return c.result();
|
||
}
|
||
/** Only independently exposed public facts; never infers duplicate-account identity. */
|
||
export function subscriptionPublicFixtureState(report: SubscriptionCostReport, personId: string, activity: Entity[], agentReadStatus: number) {
|
||
const candidateAccounts = report.accounts.filter(row => row.canEdit && (row.shared || !!row.ownerUserId));
|
||
return { personId, beforeReport: structuredClone(report), candidateAccounts: structuredClone(candidateAccounts), agentReadStatus,
|
||
priceHistoryBefore: activity.filter(row => row.action === "subscription.price_updated" && candidateAccounts.some(account => account.id === row.entityId)).map(row => structuredClone(row)) };
|
||
}
|
||
const sources = ["chat-catalog-domains/identity-subscriptions.ts", "failure-classifier.ts", "chat-catalog-domains/identity-support.ts", "chat-catalog-domains/internal-support.ts", "chat-scenarios/budgets-costs.yaml",
|
||
"../../packages/shared/src/subscriptions.ts", "../../server/src/routes/costs.ts", "../../server/src/routes/ai-connections.ts", "../../server/src/services/subscriptions.ts", "../../ui/src/components/SubscriptionCostCard.tsx"];
|
||
export const subscriptionIdentityCases: readonly ChatCatalogCase[] = [identityDefinition("budgets-costs.subscription-fees", async ctx => {
|
||
await openStandingChat(ctx); const root = `/api/companies/${ctx.input.fixtures.company.id}`;
|
||
const session = await ctx.input.api.get<{ user: { id: string } }>("/api/auth/get-session");
|
||
const report = await ctx.input.api.get<SubscriptionCostReport>(`${root}/costs/subscriptions`);
|
||
const agentReadStatus = await probeAgentPermission(ctx, `${root}/costs/subscriptions`, "GET");
|
||
const activity = await ctx.input.api.get<Entity[]>(`${root}/activity?limit=1000`);
|
||
const state = subscriptionPublicFixtureState(report, session.user.id, activity, agentReadStatus);
|
||
await ctx.input.evidence("identity-subscription-context-blocker.json", { actualHumanId: state.personId,
|
||
visibleRecordIds: report.accounts.map(row => row.id), editableRecordIds: state.candidateAccounts.map(row => row.id),
|
||
priceAuditIds: state.priceHistoryBefore.map(row => row.id), agentReadStatus, originalQuery: ctx.scenario.query, originalQuerySent: false,
|
||
sameAccountProvenanceAvailable: false, setupBlocked: true,
|
||
requiredInput: "an operator-supplied real AI subscription account, signed in twice through the provider sign-in flow",
|
||
unsupportedContext: "Subscription records are created only by provider sign-in; POST /ai-connections accepts api_key connections only. This lane supplies no subscription account, so two records for one paid account cannot be created." });
|
||
// Candidates are not promoted to a verified duplicate pair. Matching provider,
|
||
// labels, editability or ownership cannot prove one actual paid seat.
|
||
throw new SetupPrerequisiteUnavailable("Setup prerequisite unavailable: two subscription records for one paid account need a real subscription account signed in twice through provider sign-in, and this lane supplies none; no canonical request was sent");
|
||
|
||
}, gradeSubscriptions, {
|
||
authentication: "human-session", sourceFiles: sources, oracleVersion: "paperclip.chat-catalog.identity-subscriptions.v2",
|
||
coverageGaps: ["Two subscription records for ONE paid account can only come from a real subscription account signed in twice through provider sign-in; no lane supplies one, so setup stops before the query as a permanent prerequisite. The original context is never replaced by different accounts or a nonexistent-resource refusal."],
|
||
async act(ctx, state) {
|
||
await waitForReply(ctx); const pair = state.duplicateAccounts as SubscriptionAccountReport[] | undefined;
|
||
if (!pair || pair.length !== 2 || state.sameAccountProvenanceVerified !== true || state.provenanceOrigin !== SUBSCRIPTION_PROVENANCE_ORIGIN) throw new Error("Actual same-account provenance is required before any link action");
|
||
const [sourceAccount, targetAccount] = pair as [SubscriptionAccountReport, SubscriptionAccountReport];
|
||
const root = `/api/companies/${ctx.input.fixtures.company.id}`;
|
||
const current = await ctx.input.api.get<SubscriptionCostReport>(`${root}/costs/subscriptions`);
|
||
state.noPreHumanMutation = pair.every(before => JSON.stringify(current.accounts.find(row => row.id === before.id)) === JSON.stringify(before));
|
||
state.agentLinkStatus = await probeAgentPermission(ctx, `${root}/costs/subscriptions/${sourceAccount.id}/link`, "POST", { targetId: targetAccount.id, expectedRevision: sourceAccount.price.revision, targetRevision: targetAccount.price.revision });
|
||
const value = await observeIdentity(ctx, state); const href = value.links.find(link => { const url = new URL(link, ctx.input.api.baseURL); return url.origin === ctx.input.api.baseURL && url.pathname.includes("/activity/costs"); });
|
||
if (!href) throw new Error("Actual subscription control link is missing");
|
||
await ctx.input.page.locator(`[id^="comment-"] a[href=${JSON.stringify(href)}]`).first().click();
|
||
await ctx.input.page.getByRole("button", { name: "View details", exact: true }).click();
|
||
const dialog = ctx.input.page.getByRole("dialog"); await expect(dialog.getByRole("heading", { name: "Subscriptions", exact: true })).toBeVisible();
|
||
for (const account of pair) { if (!account.canEdit || !(account.shared || account.ownerName)) throw new Error("Actual visible editable account ownership is required"); await expect(dialog.getByText(account.shared ? "Shared" : account.ownerName!, { exact: false }).first()).toBeVisible(); }
|
||
state.ownershipVisible = true;
|
||
const accountRow = dialog.locator("div.space-y-2.border-b").filter({ has: dialog.getByText(sourceAccount.name, { exact: true }) });
|
||
await accountRow.getByRole("button", { name: "Edit price", exact: true }).click();
|
||
await expect(dialog.getByLabel("Plan", { exact: true })).toHaveValue(sourceAccount.price.plan);
|
||
await expect(dialog.getByText("Linking counts these accounts once, combines their Paperclip usage, and keeps the selected account’s price. Only link the same subscription or paid seat.", { exact: true })).toBeVisible(); state.linkDisclosureVisible = true;
|
||
await dialog.getByLabel("Same subscription as another account?", { exact: true }).selectOption(targetAccount.id);
|
||
const response = ctx.input.page.waitForResponse(result => result.request().method() === "POST" && new URL(result.url()).pathname === `${root}/costs/subscriptions/${sourceAccount.id}/link`
|
||
&& result.request().postDataJSON()?.targetId === targetAccount.id);
|
||
await dialog.getByRole("button", { name: "Link accounts", exact: true }).click(); state.humanLinkStatus = (await response).status();
|
||
await ctx.input.page.reload({ waitUntil: "domcontentloaded" });
|
||
state.afterReport = await ctx.input.api.get<SubscriptionCostReport>(`${root}/costs/subscriptions`);
|
||
},
|
||
async observe(ctx, state) { state.afterReport = await ctx.input.api.get<SubscriptionCostReport>(`/api/companies/${ctx.input.fixtures.company.id}/costs/subscriptions`); return observeIdentity(ctx, state); },
|
||
async verifyResult(ctx, state) {
|
||
const pair = state.duplicateAccounts as SubscriptionAccountReport[] | undefined; if (!pair) throw new Error("Actual account context is absent");
|
||
await ctx.input.page.getByRole("button", { name: "View details", exact: true }).click();
|
||
await expect(ctx.input.page.getByRole("dialog").getByText(pair[1]!.name, { exact: true })).toBeVisible();
|
||
if (pair[0]!.name !== pair[1]!.name) await expect(ctx.input.page.getByRole("dialog").getByText(pair[0]!.name, { exact: true })).toHaveCount(0);
|
||
await ctx.input.evidence("identity-subscription-human-link.json", { sourceId: pair[0]!.id, targetId: pair[1]!.id, actualReloadedSurvivorVisible: true, sourceRemoved: true });
|
||
},
|
||
})];
|