Files
PaperClipAI/tests/runner-e2e/chat-catalog-domains/privileged-cli-support.ts
T
f4337bb5fb test(evals): land the chat catalog eval harness on master (#19614)
## 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>
2026-11-10 08:02:11 -06:00

161 lines
11 KiB
TypeScript

import { execFile } from "node:child_process";
import { promisify } from "node:util";
import { mkdtemp, mkdir, realpath, rm } from "node:fs/promises";
import { fileURLToPath } from "node:url";
import path from "node:path";
import type { ChatCatalogContext } from "../chat-catalog-contract.js";
const execute = promisify(execFile);
const privateActors = new WeakMap<ChatCatalogContext, { id: string; token: string; directory: string }>();
export async function prepareOwnedCli(ctx: ChatCatalogContext) {
if (ctx.input.execution.environment.id !== "local") throw new Error("CLI companion journey requires the owned local attempt");
const root = process.env.PAPERCLIP_RUNNER_E2E_TEMP_ROOT;
if (!root) throw new Error("CLI companion requires an owned attempt directory");
const directory = await mkdtemp(path.join(await realpath(root), "operator-cli-"));
await mkdir(path.join(directory, "home"));
const key = await ctx.input.api.postSensitive<{ id: string; token: string }>("/api/board-api-keys", {
name: `catalog-operator-${ctx.input.nonce}`, requestedCompanyId: ctx.input.fixtures.company.id, expiresAt: new Date(Date.now() + 900_000).toISOString(),
});
if (!key.id || !key.token) throw new Error("Owned human CLI credential was not issued");
ctx.input.privateSecrets?.push(key.token);
privateActors.set(ctx, { ...key, directory });
return { operatorDirectory: directory, operatorKeyId: key.id };
}
export async function runOwnedCli(ctx: ChatCatalogContext, args: string[]) {
const actor = privateActors.get(ctx); if (!actor) throw new Error("Owned human CLI actor is unavailable");
// A fresh minimal environment cannot pick up another operator's context,
// stored auth, run ID, provider credentials or default company.
const env: NodeJS.ProcessEnv = { PATH: process.env.PATH, HOME: path.join(actor.directory, "home"),
PAPERCLIP_HOME: path.join(actor.directory, "home"), PAPERCLIP_CONTEXT: path.join(actor.directory, "context.json"), PAPERCLIP_API_KEY: actor.token,
PAPERCLIP_API_URL: ctx.input.api.baseURL, PAPERCLIP_TELEMETRY_DISABLED: "1", NO_COLOR: "1" };
const loader = fileURLToPath(new URL("../../../cli/node_modules/tsx/dist/loader.mjs", import.meta.url));
const entry = fileURLToPath(new URL("../../../cli/src/index.ts", import.meta.url));
try {
const result = await execute(process.execPath, ["--import", loader, entry, ...args], { cwd: actor.directory, env, timeout: 45_000, maxBuffer: 1_048_576 });
if (result.stdout.includes(actor.token) || result.stderr.includes(actor.token)) throw new Error("Private CLI credential appeared in output");
return result.stdout;
} catch { throw new Error("Owned CLI command failed; private subprocess output withheld"); }
}
export async function cleanupOwnedCli(ctx: ChatCatalogContext) {
const actor = privateActors.get(ctx); if (!actor) return;
await ctx.input.api.delete(`/api/board-api-keys/${actor.id}`);
privateActors.delete(ctx);
const root = await realpath(process.env.PAPERCLIP_RUNNER_E2E_TEMP_ROOT!);
const owned = await realpath(actor.directory);
if (!owned.startsWith(root + path.sep) || !path.basename(owned).startsWith("operator-cli-")) throw new Error("Refusing CLI cleanup outside its owned directory");
await rm(owned, { recursive: true });
}
/** Parse one finite supported command into argv; never evaluate a shell. */
export function websiteCreateArgv(reply: string, fixture: { company: string; project: string; agent: string; apiBase: string }, bindings: Array<{ flag: string; placeholder: string; fixtureValue: string }> = []) {
const blocks = [...reply.matchAll(/```(?:bash|sh|shell)?\s*\n([\s\S]*?)```/g)].map(match => match[1]!.replace(/\\\r?\n/g, " ").trim());
const commands = blocks.flatMap(block => block.split(/\r?\n/)).filter(line => /^(?:paperclipai|npx(?: -y)? paperclipai|pnpm (?:exec )?paperclipai)\s+issue\s+create\b/.test(line.trim()));
if (commands.length !== 1) throw new Error("One supported website creation command is required");
const line = commands[0]!.trim();
if (/[`;|&\n\r]|\$\(/.test(line)) throw new Error("Shell execution is not admitted in the owned CLI journey");
const tokens = line.match(/"(?:[^"\\]|\\.)*"|'[^']*'|[^\s]+/g)!;
const values = tokens.map(token => token.startsWith('"') ? JSON.parse(token) as string : token.startsWith("'") ? token.slice(1, -1) : token);
const start = values.indexOf("issue"), argv = values.slice(start);
const allowed = new Set(["--company-id", "-C", "--title", "--description", "--status", "--priority", "--assignee-agent-id", "--project-id", "--api-base"]);
const fields = new Map<string, string>();
for (let i = 2; i < argv.length; i++) {
const flag = argv[i]!; if (flag === "--json" || flag === "--standalone") continue;
if (!allowed.has(flag) || !argv[i + 1] || fields.has(flag)) throw new Error("Unsupported or ambiguous creation flags");
let value = argv[++i]!;
const target = ["--company-id", "-C"].includes(flag) ? fixture.company : flag === "--project-id" ? fixture.project : flag === "--assignee-agent-id" ? fixture.agent : undefined;
// Only unresolved identifier placeholders are bound; actual wrong IDs fail.
if (target && /^(?:<[A-Z_ -]+>|\$\{?[A-Z_]+\}?)$/.test(value)) { bindings.push({ flag, placeholder: value, fixtureValue: target }); value = target; argv[i] = target; }
fields.set(flag, value);
}
if ((fields.get("--company-id") ?? fields.get("-C")) !== fixture.company || fields.get("--project-id") !== fixture.project
|| fields.get("--assignee-agent-id") !== fixture.agent || !fields.get("--title")
|| fields.has("--status") && !["backlog", "todo"].includes(fields.get("--status")!) || fields.has("--api-base") && fields.get("--api-base") !== fixture.apiBase)
throw new Error("Creation command does not target exactly the intended company, project and Morgan");
if (fields.has("--company-id") && fields.has("-C")) throw new Error("Duplicate company context flags are ambiguous");
return argv;
}
/** The person's own Paperclip address is not knowable to the agent; a placeholder for it binds to the fixture origin. */
const addressPlaceholder = /^(?:<[^<>]+>|\$\{?[A-Z_]+\}?)$/;
const profileName = /^[A-Za-z0-9][A-Za-z0-9_.-]*$/;
export function contextGuidanceArgv(reply: string, fixture: { company: string; apiBase: string }) {
const lines = [...reply.matchAll(/```(?:bash|sh|shell|zsh|console)?\s*\n([\s\S]*?)```/g)]
.flatMap(match => match[1]!.replace(/\\\r?\n/g, " ").split(/\r?\n/))
.map(line => line.trim().replace(/^\$\s+/, "")).filter(line => line && !line.startsWith("#"));
let selected = false, readback = false;
const commands: string[][] = [], bindings: Array<{ commandIndex: number; flag: string; placeholder: string; fixtureValue: string }> = [];
const signIn: string[][] = [];
for (const line of lines) {
if (/[`;|&\n\r]|\$\(/.test(line)) return;
const prefix = /^(?:paperclipai|npx(?: -y)? paperclipai(?:@latest)?|pnpm (?:exec )?paperclipai)\s+/;
if (!prefix.test(line)) return;
let tokens: string[];
try { tokens = line.replace(prefix, "").match(/<[^<>]+>|"(?:[^"\\]|\\.)*"|'[^']*'|[^\s]+/g)!.map(token => token.startsWith('"') ? JSON.parse(token) as string : token.startsWith("'") ? token.slice(1, -1) : token); }
catch { return; }
const original = tokens;
const family = original[0];
const command = original[1], args = original.slice(2);
// A sign-in step for a terminal that may not be signed in yet. The owned
// human actor is already signed in and browser approval cannot run here,
// so a well-formed login is recorded and not executed; other flags fail.
if (family === "auth" && command === "login") {
const flags = new Map<string, string>();
while (args.length) {
const flag = args.shift()!, value = args.shift();
if (!["--api-base", "--company-id"].includes(flag) || !value || flags.has(flag)) return;
flags.set(flag, value);
}
const base = flags.get("--api-base"), company = flags.get("--company-id");
if (base !== undefined && base !== fixture.apiBase && !addressPlaceholder.test(base) || company !== undefined && company !== fixture.company) return;
signIn.push(original); continue;
}
if (family === "company" || family === "auth") {
// Reads only: listing companies, the selected company and the signed-in identity.
if (family === "company" && command !== "current" && command !== "list" || family === "auth" && command !== "whoami"
|| args.length > 1 || args.length === 1 && args[0] !== "--json") return;
if (family === "company" && command === "current" && selected) readback = true;
commands.push(original); continue;
}
if (family !== "context") return;
if (command === "use") { const name = args.shift(); if (!name || !profileName.test(name)) return; selected = true; }
else if (command !== "show" && command !== "set" && command !== "list") return;
let profile: string | undefined, company: string | undefined, apiBase: string | undefined, persona: string | undefined, use = false;
const allowedFlags = command === "set" ? ["--profile", "--company-id", "--api-base", "--persona", "--use", "--json"]
: command === "show" ? ["--profile", "--json"] : command === "list" ? ["--json"] : [];
const flags = new Set<string>();
while (args.length) {
const flag = args.shift();
if (!flag || !allowedFlags.includes(flag) || flags.has(flag)) return;
flags.add(flag);
if (flag === "--json") continue;
if (flag === "--use") { use = true; continue; }
if (!args.length) return;
const value = args.shift()!;
if (flag === "--profile") profile = value;
if (flag === "--company-id") company = value;
if (flag === "--api-base") apiBase = value;
if (flag === "--persona") persona = value;
}
if (profile !== undefined && !profileName.test(profile)) return;
if (command === "set") {
// A set without --profile rewrites the person's current (other) profile.
// The company is known in the conversation, so it must be the real id;
// only the person's own Paperclip address may stay a placeholder.
if (profile === undefined || company !== undefined && company !== fixture.company
|| apiBase !== undefined && apiBase !== fixture.apiBase && !addressPlaceholder.test(apiBase)
|| persona !== undefined && persona !== "board") return;
if (apiBase !== undefined && apiBase !== fixture.apiBase) {
const index = original.indexOf(apiBase);
bindings.push({ commandIndex: commands.length, flag: "--api-base", placeholder: apiBase, fixtureValue: fixture.apiBase });
original[index] = fixture.apiBase;
}
selected ||= use;
}
if (command === "show") readback ||= selected;
commands.push(original);
}
return selected && readback ? { commands, bindings, signIn } : undefined;
}
export function validContextGuidance(reply: string, fixture: { company: string; apiBase: string }) {
return !!contextGuidanceArgv(reply, fixture);
}