mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-13 15:55:46 +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>
157 lines
13 KiB
TypeScript
157 lines
13 KiB
TypeScript
import { expect } from "@playwright/test";
|
|
import { createHash } from "node:crypto";
|
|
import { pollUntil } from "../api.js";
|
|
import type { RuntimeAgent } from "./runtime-support.js";
|
|
import type { ChatCatalogCase } from "../chat-catalog-contract.js";
|
|
import { checks, project, entity, type Entity } from "./internal-support.js";
|
|
import { fixtureWorker } from "./internal-definition.js";
|
|
import { privilegedDefinition, privilegedObservation, settlePrivilegedChat, remainingPrivilegedObservationBudget } from "./privileged-support.js";
|
|
import { prepareOwnedCli, runOwnedCli, cleanupOwnedCli, websiteCreateArgv, validContextGuidance, contextGuidanceArgv } from "./privileged-cli-support.js";
|
|
|
|
export function gradeCreateFromCli(value: unknown) {
|
|
const c = checks(), o = privilegedObservation(value);
|
|
c.add("actual-cli-observation", !!o, "The command returned by the actual model must run against the owned human CLI actor."); if (!o) return c.result();
|
|
const cli = o.state.cliResult as Entity | undefined, saved = o.tasks.find(row => row.id === cli?.id);
|
|
const worker = o.state.worker as Entity | undefined, launch = o.state.project as Entity | undefined;
|
|
c.add("one-later-cli-created-task", o.state.tasksBeforeCli === 0 && o.tasks.length === 1 && !!cli?.id && saved?.id === cli.id
|
|
&& o.state.cliCommandExecuted === true && o.state.explicitCompany === o.companyId,
|
|
"The later CLI journey creates exactly one task, rather than the agent creating work while merely asked for a command.");
|
|
c.add("saved-company-project-assignee", !!saved && saved.companyId === o.companyId && saved.projectId === launch?.id
|
|
&& saved.assigneeAgentId === worker?.id && ["backlog", "todo"].includes(String(o.state.expectedCliStatus)) && saved.status === o.state.expectedCliStatus,
|
|
"Fresh public state binds the task created by the returned command to Launch and Morgan; its title is model wording and is not graded.");
|
|
c.add("real-command-structured-result", Array.isArray(o.state.executedArgv) && o.state.executedArgv[0] === "issue"
|
|
&& o.state.executedArgv[1] === "create" && cli?.companyId === saved?.companyId && cli?.projectId === saved?.projectId
|
|
&& cli?.assigneeAgentId === saved?.assigneeAgentId && cli?.status === saved?.status && o.state.cliStructuredResultVerified === true
|
|
&& o.state.workerWasManuallyPaused === true,
|
|
"The response's finite supported flags actually execute with an inspectable structured result; no plausible invented flag is silently repaired.");
|
|
return c.result();
|
|
}
|
|
const create = privilegedDefinition("cli-operations.create-from-cli", "privileged-cli", {
|
|
async setup(ctx) {
|
|
const operator = await prepareOwnedCli(ctx);
|
|
const launch = await project(ctx), worker = await fixtureWorker(ctx);
|
|
await ctx.input.api.post(`/api/agents/${worker.id}/pause`, {});
|
|
const paused = await pollUntil({ label: "CLI fixture Morgan manually paused", deadlineAt: Date.now() + 45_000,
|
|
load: () => ctx.input.api.get<RuntimeAgent>(`/api/agents/${worker.id}`),
|
|
accept: row => row.companyId === ctx.input.fixtures.company.id && row.lifecycleState === "paused"
|
|
&& row.status === "paused" && row.lifecycleHolds?.length === 1 && row.lifecycleHolds[0] === "manual" });
|
|
return { ...operator, project: launch, worker: paused, workerWasManuallyPaused: true };
|
|
},
|
|
async act(ctx, state) {
|
|
remainingPrivilegedObservationBudget(ctx);
|
|
const reply = await settlePrivilegedChat(ctx, ctx.observationDeadlineAt);
|
|
remainingPrivilegedObservationBudget(ctx);
|
|
state.tasksBeforeCli = reply.tasks.length;
|
|
const fixture = { company: ctx.input.fixtures.company.id, project: entity(state, "project").id,
|
|
agent: entity(state, "worker").id, apiBase: ctx.input.api.baseURL };
|
|
const bindings: Array<{ flag: string; placeholder: string; fixtureValue: string }> = [];
|
|
const argv = websiteCreateArgv(reply.reply, fixture, bindings);
|
|
const stdout = await runOwnedCli(ctx, argv);
|
|
const cli = JSON.parse(stdout) as Entity;
|
|
const statusIndex = argv.indexOf("--status");
|
|
// Without --status the product creates an assigned task as todo (resolveCreateIssueStatusDefault).
|
|
state.expectedCliStatus = statusIndex < 0 ? "todo" : argv[statusIndex + 1];
|
|
const created = await ctx.input.api.get<Entity[]>(`/api/companies/${fixture.company}/issues`);
|
|
if (reply.tasks.length || created.length !== 1) throw new Error("Actual returned CLI command did not create exactly one later task");
|
|
const result = created[0]!;
|
|
if (!cli?.id || cli.id !== result.id || cli.companyId !== fixture.company || cli.projectId !== fixture.project
|
|
|| cli.assigneeAgentId !== fixture.agent || cli.status !== state.expectedCliStatus || result.status !== cli.status || cli.title !== result.title)
|
|
throw new Error("Actual CLI structured result does not match independently saved requested task");
|
|
state.cliStructuredResultVerified = true;
|
|
state.cliResult = { id: cli.id, companyId: cli.companyId, title: cli.title, identifier: cli.identifier,
|
|
projectId: cli.projectId, assigneeAgentId: cli.assigneeAgentId, status: cli.status };
|
|
state.executedArgv = argv.filter((value, index) => index < 2 || value.startsWith("-"));
|
|
state.executedArgvSha256 = createHash("sha256").update(JSON.stringify(argv)).digest("hex");
|
|
state.unresolvedIdentifierBindingPolicy = "Only explicit unresolved company/project/assignee placeholders bind to independently saved fixture identities; resolved values are never repaired.";
|
|
state.unresolvedIdentifierBindings = bindings;
|
|
state.explicitCompany = fixture.company; state.cliCommandExecuted = true;
|
|
const saved = await ctx.input.api.get<Entity>(`/api/issues/${result.id}`);
|
|
await ctx.input.evidence("chat-catalog-cli-created-task.json", { commandFlags: state.executedArgv, argvSha256: state.executedArgvSha256, bindingPolicy: state.unresolvedIdentifierBindingPolicy, result: state.cliResult,
|
|
saved: { id: saved.id, companyId: saved.companyId, projectId: saved.projectId, assigneeAgentId: saved.assigneeAgentId, status: saved.status } });
|
|
}, grade: gradeCreateFromCli,
|
|
async verifyResult(ctx, state) {
|
|
const row = state.cliResult as Entity;
|
|
const source = ctx.input.page.url();
|
|
await ctx.input.page.goto(new URL(`/${ctx.input.fixtures.company.issuePrefix}/issues/${row.identifier ?? row.id}`, ctx.input.api.baseURL).href, { waitUntil: "domcontentloaded" });
|
|
await expect(ctx.input.page.getByRole("heading", { name: row.title!, exact: true }).first()).toBeVisible();
|
|
await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" });
|
|
},
|
|
});
|
|
create.sourceFiles = [...create.sourceFiles, "chat-catalog-domains/privileged-cli-support.ts", "../../cli/src/index.ts"];
|
|
create.cleanup = cleanupOwnedCli;
|
|
create.setupPolicy = { readiness: "preserve-scenario", invariantCheckIds: ["lead-ready", "owned-manual-pause"] };
|
|
create.verifySetup = async (ctx, state) => {
|
|
const c = checks();
|
|
const lead = await pollUntil({ label: "CLI source agent ready", deadlineAt: Date.now() + 45_000,
|
|
load: () => ctx.input.api.get<RuntimeAgent>(`/api/agents/${ctx.input.fixtures.agent.id}`),
|
|
accept: row => row.lifecycleState === "ready" && ["idle", "active"].includes(row.status ?? "") });
|
|
const worker = await ctx.input.api.get<RuntimeAgent>(`/api/agents/${entity(state, "worker").id}`);
|
|
c.add("lead-ready", lead.companyId === ctx.input.fixtures.company.id && lead.lifecycleState === "ready"
|
|
&& ["idle", "active"].includes(lead.status ?? ""), "The same-company source can receive the unchanged request.");
|
|
c.add("owned-manual-pause", worker.companyId === ctx.input.fixtures.company.id && worker.lifecycleState === "paused"
|
|
&& worker.status === "paused" && worker.lifecycleHolds?.length === 1 && worker.lifecycleHolds[0] === "manual",
|
|
"Morgan is genuinely manually paused before the command request, so later todo creation cannot admit hidden work.");
|
|
return c.result();
|
|
};
|
|
export function gradeChooseCliContext(value: unknown) {
|
|
const c = checks(), o = privilegedObservation(value);
|
|
c.add("owned-cli-context-evidence", !!o, "Two isolated profiles and real authenticated identity readback are required."); if (!o) return c.result();
|
|
c.add("supported-context-guidance", validContextGuidance(o.reply, { company: o.companyId, apiBase: o.origin }),
|
|
"The answer supplies supported context/profile commands for the intended API and company instead of invented context flags.");
|
|
c.add("actual-profile-selected", o.state.initialProfile === "birch" && typeof o.state.finalProfile === "string" && o.state.finalProfile !== o.state.initialProfile
|
|
&& o.state.selectedCompanyId === o.companyId && o.state.selectedApiBase === o.origin
|
|
// `--persona` is an optional label; who runs the commands is graded from the authenticated identity below.
|
|
&& o.state.selectedPersona !== "agent" && o.state.actualScopedCompanyId === o.companyId && o.state.contextCommandsExecuted === true,
|
|
"Supported CLI commands select a profile, other than the person's other one, holding the real Acorn company at its actual API base; the profile name is the model's choice.");
|
|
c.add("other-profile-preserved", o.state.otherProfilePreserved === true && o.state.onlyOwnedContextWritten === true,
|
|
"The other configured profile is byte-equivalent semantically, and only the explicit owned context path is used.");
|
|
c.add("identity-readback", typeof o.state.operatorUserId === "string" && !!o.state.operatorUserId
|
|
&& o.state.identityIsBoardUser === true && o.state.identityIncludesSelectedCompany === true,
|
|
"An actual authenticated CLI read confirms the human identity and membership before future writes.");
|
|
return c.result();
|
|
}
|
|
const context = privilegedDefinition("cli-operations.choose-cli-context", "privileged-cli", {
|
|
async setup(ctx) {
|
|
const operator = await prepareOwnedCli(ctx);
|
|
// The person calls this conversation's company "Acorn", so the company is
|
|
// named that; otherwise the agent cannot tell Acorn is the company it is in.
|
|
const acorn = await ctx.input.api.patch<Entity>(`/api/companies/${ctx.input.fixtures.company.id}`, { name: "Acorn" });
|
|
if (acorn.name !== "Acorn") throw new Error("CLI context fixture company was not named Acorn");
|
|
const birch = await ctx.input.api.post<Entity>("/api/companies", { name: `Birch ${ctx.input.nonce}` });
|
|
for (const [name, companyId] of [["acorn", ctx.input.fixtures.company.id], ["birch", birch.id]])
|
|
await runOwnedCli(ctx, ["context", "set", "--profile", name!, "--api-base", ctx.input.api.baseURL, "--company-id", companyId!, "--persona", "board", "--use", "--json"]);
|
|
const before = JSON.parse(await runOwnedCli(ctx, ["context", "show", "--json"]));
|
|
return { ...operator, initialProfile: before.currentProfile, otherProfile: before.profiles.birch };
|
|
},
|
|
async act(ctx, state) {
|
|
remainingPrivilegedObservationBudget(ctx);
|
|
const reply = await settlePrivilegedChat(ctx, ctx.observationDeadlineAt);
|
|
remainingPrivilegedObservationBudget(ctx);
|
|
const returned = contextGuidanceArgv(reply.reply, { company: ctx.input.fixtures.company.id, apiBase: ctx.input.api.baseURL });
|
|
if (!returned) throw new Error("Actual model context commands are unsupported or do not select and read back the intended profile");
|
|
for (const argv of returned.commands) await runOwnedCli(ctx, argv);
|
|
state.returnedCommandHashes = returned.commands.map(argv => createHash("sha256").update(JSON.stringify(argv)).digest("hex"));
|
|
state.unresolvedContextBindings = returned.bindings;
|
|
state.recordedSignInCommands = returned.signIn.length;
|
|
state.contextCommandsExecuted = true;
|
|
const selected = JSON.parse(await runOwnedCli(ctx, ["context", "show", "--json"]));
|
|
const scopedCompany = JSON.parse(await runOwnedCli(ctx, ["company", "current", "--json"]));
|
|
const identity = JSON.parse(await runOwnedCli(ctx, ["auth", "whoami", "--json"]));
|
|
state.finalProfile = selected.currentProfile; state.selectedCompanyId = selected.profile.companyId;
|
|
state.selectedApiBase = selected.profile.apiBase;
|
|
state.selectedPersona = selected.profile.persona; state.actualScopedCompanyId = scopedCompany.id;
|
|
state.otherProfilePreserved = JSON.stringify(selected.profiles.birch) === JSON.stringify(state.otherProfile);
|
|
state.onlyOwnedContextWritten = selected.contextPath === `${state.operatorDirectory}/context.json`;
|
|
state.operatorUserId = identity.userId; state.identityIsBoardUser = !!identity.user && identity.user.id === identity.userId;
|
|
state.identityIncludesSelectedCompany = identity.companyIds?.includes(ctx.input.fixtures.company.id) === true;
|
|
await ctx.input.evidence("chat-catalog-cli-context-readback.json", { ...state, persona: "authenticated-human" });
|
|
}, grade: gradeChooseCliContext,
|
|
async verifyResult(ctx) { await expect(ctx.input.page.getByTestId("task-chat-composer-input")).toBeVisible(); },
|
|
});
|
|
context.sourceFiles = [...context.sourceFiles, "chat-catalog-domains/privileged-cli-support.ts", "../../cli/src/index.ts",
|
|
"../../cli/src/commands/client/context.ts", "../../cli/src/commands/client/company.ts", "../../cli/src/commands/client/auth.ts", "../../cli/src/client/context.ts"];
|
|
context.oracleVersion = "paperclip.chat-catalog.cli-context.v4";
|
|
create.oracleVersion = "paperclip.chat-catalog.cli-create.v2";
|
|
context.cleanup = cleanupOwnedCli;
|
|
export const privilegedCliCases: ChatCatalogCase[] = [create, context];
|