Files
PaperClipAI/tests/runner-e2e/source-sdk-evidence.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

195 lines
17 KiB
TypeScript

import { createSdkArtifactScanner } from "./source-sdk-public-type.js";
import { constants } from "node:fs";
import { assertPrivateCellDirectories, privateCellEvidenceOwner } from "./manual-private-report-assets.js";
import { createHash } from "node:crypto";
import { chmod, lstat, mkdir, open, readFile, readdir, realpath, rm, writeFile } from "node:fs/promises";
import path from "node:path";
import { diagnosticSecrets, readPrivateDiagnosticSecrets } from "./private-diagnostics.js";
import { findSecretLeak } from "./redaction.js";
import { isStandardEvidenceFile } from "./evidence.js";
const selectedIds = new Set(["plugins.configure-plugin", "plugins.disable-plugin", "plugins.install-plugin", "plugins.use-plugin-tool", "routines.plugin-managed-routine", "secrets.least-scope-delivery"].map(id => `chat-features.runner-opencode.local.${id}`));
export const SOURCE_SDK_DIAGNOSTICS = "diagnostics/source-plugin-sdk";
const targets = ["packages/shared/dist", "packages/plugins/sdk/dist"] as const;
const hash = (bytes: string | Buffer) => createHash("sha256").update(bytes).digest("hex");
const hex = (value: unknown): value is string => typeof value === "string" && /^[a-f0-9]{64}$/.test(value);
const staging = (root: string) => path.join(root, "artifacts-private", "source-sdk-build");
interface BuiltFile { file: string; size: number; sha256: string }
// Retained bytes must use a name the private result sealer admits as a
// standard carrier: declarations (.d.ts) and source maps (.map) are retained
// byte-for-byte under a .txt suffix instead of failing the whole attempt.
export const retainedSourceSdkPath = (file: string) => {
const named = file.replace(/\/\.paperclip-build-complete$/, "/build-completion-receipt.json");
return isStandardEvidenceFile(`html-report/${SOURCE_SDK_DIAGNOSTICS}/${named}`) ? named : `${named}.txt`;
};
const retainedPath = retainedSourceSdkPath;
interface RetainedBuild {
schema: "paperclip.source-plugin-sdk/v1" | "paperclip.source-plugin-sdk/v2"; sourceRootSha256?: string; scanAudit?: Awaited<ReturnType<ReturnType<typeof createSdkArtifactScanner>["audit"]>>; executionId: string; sourceRevision: string; retainedAt: string;
buildHelperSha256: string; lockSha256: string; files: Array<BuiltFile & { retainedPath: string; retainedSha256: string }>;
certifiedTargets: Array<{ directory: string; sources: string; outputs: string }>;
}
function selected(ids: readonly string[]) { return ids.length === 1 && selectedIds.has(ids[0]!); }
function checkSecret(bytes: Buffer, secrets: readonly string[]) {
if (findSecretLeak(bytes.toString("utf8"), secrets)) throw new Error("Source SDK evidence privacy check failed; bytes withheld");
}
async function retainScanAudit(root: string, executionId: string, audit: unknown, complete: boolean, secrets: readonly string[]) {
const bytes = Buffer.from(JSON.stringify({ schema: "paperclip.source-sdk-privacy-observation/v1", complete, audit }, null, 2) + "\n");
checkSecret(bytes, secrets);
const directory = path.join(root, "artifacts-private", "cases", executionId.slice("chat-features.runner-opencode.local.".length), "snapshots");
await mkdir(directory, { recursive: true, mode: 0o700 });
await writeFile(path.join(directory, "source-sdk-privacy-observations.json"), bytes, { flag: "wx", mode: 0o600 });
}
async function boundedTree(root: string) {
const files: Array<BuiltFile & { bytes: Buffer }> = []; let total = 0;
async function visit(directory: string) {
const entries = await readdir(directory, { withFileTypes: true });
for (const entry of entries.sort((a, b) => a.name.localeCompare(b.name))) {
const filename = path.join(directory, entry.name);
if (entry.isDirectory()) await visit(filename);
else {
const info = await lstat(filename);
if (!info.isFile() || info.isSymbolicLink() || info.size > 8 * 1024 * 1024) throw new Error("Source SDK output custody invalid");
const bytes = await readFile(filename); total += bytes.length;
if (files.length >= 4096 || total > 64 * 1024 * 1024) throw new Error("Source SDK output exceeds evidence bound");
files.push({ file: path.relative(root, filename).split(path.sep).join("/"), size: bytes.length, sha256: hash(bytes), bytes });
}
}
}
const info = await lstat(root); if (!info.isDirectory() || info.isSymbolicLink()) throw new Error("Source SDK output root invalid");
await visit(root); return files;
}
function outputHash(files: Array<{ file: string; bytes: Buffer }>) {
const digest = createHash("sha256");
for (const file of files.filter(row => row.file !== ".paperclip-build-complete")) {
digest.update(JSON.stringify([file.file, file.bytes.length])); digest.update(file.bytes);
}
return digest.digest("hex");
}
export async function retainPreparedSourceSdk(input: { repositoryRoot: string; temporaryRoot: string; executionIds: readonly string[]; sourceRevision: string | null; installed: boolean }) {
if (input.installed || !selected(input.executionIds)) return;
if (!/^[a-f0-9]{40}$/.test(input.sourceRevision ?? "")) throw new Error("Source SDK retention requires exact source revision");
const root = await realpath(input.temporaryRoot), repository = await realpath(input.repositoryRoot);
// External private capture need not create the local SDK staging parent.
// Own its initialization before retention instead of relying on later server-log setup.
const privateRoot = path.join(root, "artifacts-private");
try { await mkdir(privateRoot, { mode: 0o700 }); }
catch (error) { if ((error as NodeJS.ErrnoException).code !== "EEXIST") throw error; }
if (await realpath(privateRoot) !== privateRoot) throw new Error("Source SDK staging custody invalid");
const secrets = diagnosticSecrets(readPrivateDiagnosticSecrets(path.join(root, "selected-private-inputs", "diagnostic-redaction.json"), root)).values;
const scanner = createSdkArtifactScanner(repository, secrets);
const buffers: Array<BuiltFile & { bytes: Buffer }> = [], certifiedTargets: RetainedBuild["certifiedTargets"] = [];
for (const directory of targets) {
const files = await boundedTree(path.join(repository, directory));
const marker = files.find(row => row.file === ".paperclip-build-complete"), entry = files.find(row => row.file === "index.js");
if (!marker || !entry) throw new Error("Source SDK build completion evidence missing");
const certificate = JSON.parse(marker.bytes.toString("utf8"));
if (!hex(certificate.sources) || !hex(certificate.outputs) || outputHash(files) !== certificate.outputs) throw new Error("Source SDK build output no longer matches its completed compiler receipt");
certifiedTargets.push({ directory, sources: certificate.sources, outputs: certificate.outputs });
for (const file of files) {
if (!await scanner.scan(`${directory}/${file.file}`, file.bytes)) {
await retainScanAudit(root, input.executionIds[0]!, await scanner.audit(), false, secrets);
throw new Error("Source SDK evidence privacy check failed; bytes withheld");
}
buffers.push({ ...file, file: `${directory}/${file.file}` });
}
}
if (buffers.length >= 4096 || buffers.reduce((total, file) => total + file.size, 0) > 63 * 1024 * 1024) throw new Error("Source SDK output exceeds combined evidence bound");
const scanAudit = await scanner.audit();
await retainScanAudit(root, input.executionIds[0]!, scanAudit, true, secrets);
const syntax = scanAudit.observations.some(row => row.classification === "public_schema_type");
const receipt: RetainedBuild = { schema: syntax ? "paperclip.source-plugin-sdk/v2" : "paperclip.source-plugin-sdk/v1", ...(syntax ? { sourceRootSha256: hash(repository), scanAudit } : {}), executionId: input.executionIds[0]!, sourceRevision: input.sourceRevision!, retainedAt: new Date().toISOString(),
buildHelperSha256: hash(await readFile(path.join(repository, "scripts/ensure-plugin-build-deps.mjs"))), lockSha256: hash(await readFile(path.join(repository, "pnpm-lock.yaml"))),
certifiedTargets, files: buffers.map(({ bytes: _, ...row }) => ({ ...row, retainedPath: retainedPath(row.file), retainedSha256: row.sha256 })) };
if (new Set(receipt.files.map(file => file.retainedPath)).size !== receipt.files.length) throw new Error("Source SDK retained filename collision");
const receiptBytes = Buffer.from(JSON.stringify(receipt, null, 2) + "\n"); checkSecret(receiptBytes, secrets);
const destination = staging(root); await mkdir(destination, { mode: 0o700 });
try {
for (const file of buffers) {
const filename = path.join(destination, file.file); await mkdir(path.dirname(filename), { recursive: true, mode: 0o700 });
await writeFile(filename, file.bytes, { flag: "wx", mode: 0o600 });
}
await writeFile(path.join(destination, "receipt.json"), receiptBytes, { flag: "wx", mode: 0o600 });
} catch (error) { await rm(destination, { recursive: true, force: true }); throw error; }
}
export async function verifyRetainedSourceSdk(root: string, executionId: string, sourceRevision: string | null, knownSecrets: readonly string[], repositoryRoot = path.resolve(import.meta.dirname, "../..")) {
if (!selected([executionId])) throw new Error("Source SDK retained selection mismatch");
if (await realpath(staging(root)) !== staging(root)) throw new Error("Source SDK retained custody invalid");
const secrets = diagnosticSecrets(knownSecrets).values, files = await boundedTree(staging(root));
const receiptFile = files.find(row => row.file === "receipt.json"); if (!receiptFile) throw new Error("Source SDK retained receipt missing");
const receipt = JSON.parse(receiptFile.bytes.toString("utf8")) as RetainedBuild;
if (!["paperclip.source-plugin-sdk/v1", "paperclip.source-plugin-sdk/v2"].includes(receipt.schema) || receipt.executionId !== executionId || receipt.sourceRevision !== sourceRevision
|| !hex(receipt.buildHelperSha256) || !hex(receipt.lockSha256) || !Array.isArray(receipt.files)) throw new Error("Source SDK retained identity mismatch");
const actual = files.filter(row => row.file !== "receipt.json");
if (actual.length !== receipt.files.length || actual.some(row => !targets.some(target => row.file.startsWith(`${target}/`))
|| !receipt.files.some(expected => expected.file === row.file && expected.retainedPath === retainedPath(row.file) && expected.size === row.size && expected.sha256 === row.sha256 && expected.retainedSha256 === row.sha256))) throw new Error("Source SDK retained bytes changed");
if (receipt.schema === "paperclip.source-plugin-sdk/v2") {
if (!receipt.scanAudit || receipt.sourceRootSha256 !== hash(await realpath(repositoryRoot))) throw new Error("Source SDK public source custody changed");
const scanner = createSdkArtifactScanner(await realpath(repositoryRoot), secrets);
for (const file of actual) if (!await scanner.scan(file.file, file.bytes)) throw new Error("Source SDK evidence privacy check failed; bytes withheld");
if (JSON.stringify(await scanner.audit()) !== JSON.stringify(receipt.scanAudit)) throw new Error("Source SDK public observations changed");
checkSecret(receiptFile.bytes, secrets);
} else for (const file of files) checkSecret(file.bytes, secrets);
return { receipt, files };
}
export async function copyRetainedSourceSdk(root: string, executionId: string, sourceRevision: string | null, uploadDir: string, knownSecrets: readonly string[], repositoryRoot = path.resolve(import.meta.dirname, "../..")) {
if (!selected([executionId])) return;
const { files } = await verifyRetainedSourceSdk(root, executionId, sourceRevision, knownSecrets, repositoryRoot);
const destination = path.join(uploadDir, SOURCE_SDK_DIAGNOSTICS); await mkdir(destination, { recursive: true, mode: 0o700 });
for (const file of files) {
// Verification proved every secret-shaped span in these bytes is a public
// schema type, but the private result sealer scans packaged text for key
// shapes without that proof (e.g. `OPENROUTER_API_KEY: z.ZodOptional` in
// validators/agent.d.ts) and rejects the whole attempt. Keep such files out
// of the upload; the retained receipt still records their size and hash.
if (file.file !== "receipt.json" && findSecretLeak(file.bytes, [])) continue;
const filename = path.join(destination, retainedPath(file.file)); await mkdir(path.dirname(filename), { recursive: true, mode: 0o700 });
await writeFile(filename, file.bytes, { flag: "wx", mode: 0o600 }); await chmod(filename, 0o600);
if (hash(await readFile(filename)) !== file.sha256) throw new Error("Source SDK published bytes changed");
}
}
export interface SourceSdkAuditCapture { executionId: string; context: Parameters<typeof privateCellEvidenceOwner>[0] }
function closedSdkAudit(value: unknown): boolean {
const object = (v: any, keys: string[]) => !!v && typeof v === "object" && !Array.isArray(v) && Object.keys(v).sort().join() === [...keys].sort().join();
const v = value as any, a = v?.audit;
if (!object(v, ["schema", "complete", "audit"]) || v.schema !== "paperclip.source-sdk-privacy-observation/v1" || typeof v.complete !== "boolean"
|| !object(a, ["schema", "suppressedFindings", "bounded", "publicSource", "observations"]) || a.schema !== "paperclip.source-sdk-scan/v1" || a.suppressedFindings !== 0
|| typeof a.bounded !== "boolean" || !Array.isArray(a.observations) || a.observations.length > 4096) return false;
if (a.publicSource !== null) {
const hashes = ["manifestSha256", "lockSha256", "compilerPackageSha256", "parserPackageSha256", "parserEntrySha256", "zodPackageSha256", "zodEntrySha256", "zodTypeGraphSha256"];
if (!object(a.publicSource, ["inputs", ...hashes]) || !Number.isSafeInteger(a.publicSource.inputs) || a.publicSource.inputs < 1 || a.publicSource.inputs > 20000 || !hashes.every(k => hex(a.publicSource[k]))) return false;
}
return a.observations.every((r: any) => object(r, ["pathSha256", "fileSha256", "detector", "start", "end", "classification"])
&& hex(r.pathSha256) && hex(r.fileSha256) && Number.isInteger(r.detector) && r.detector >= -1 && r.detector <= 3
&& Number.isSafeInteger(r.start) && Number.isSafeInteger(r.end) && r.start >= 0 && r.end >= r.start && r.end <= 8 * 1024 * 1024
&& (r.detector === -1 ? r.classification === "private_exact" : r.detector === 3 ? ["unresolved_context", "public_schema_type"].includes(r.classification) : r.classification === "direct_token"));
}
/** Copy only the already-produced typed audit, never rejected SDK output. The
* private producer owner binds external capture even when server setup failed. */
export async function copyPreparedSourceSdkAudit(temporaryRoot: string, casePrivateDir: string, secrets: readonly string[], input: SourceSdkAuditCapture) {
if (!selected([input.executionId])) return;
const owner = await privateCellEvidenceOwner(input.context, input.executionId);
const scenario = input.executionId.slice("chat-features.runner-opencode.local.".length);
if (path.resolve(casePrivateDir) !== path.join(owner.root, "capture", "cases", scenario)) throw new Error("SDK audit capture does not match its exact private cell");
await assertPrivateCellDirectories(owner, casePrivateDir);
const root = await realpath(temporaryRoot), rootInfo = await lstat(temporaryRoot);
if (root !== path.resolve(temporaryRoot) || !rootInfo.isDirectory() || rootInfo.isSymbolicLink() || rootInfo.mode & 0o077
|| typeof process.getuid === "function" && rootInfo.uid !== process.getuid()) throw new Error("SDK audit source custody invalid");
const source = path.join(root, "artifacts-private", "cases", scenario, "snapshots", "source-sdk-privacy-observations.json");
const info = await lstat(source).catch((error: NodeJS.ErrnoException) => { if (error.code === "ENOENT") return; throw error; });
if (!info) return;
if (!info.isFile() || info.isSymbolicLink() || info.size > 1024 * 1024 || await realpath(source) !== source) throw new Error("SDK audit source custody invalid");
const handle = await open(source, constants.O_RDONLY | constants.O_NOFOLLOW);
let bytes: Buffer;
try {
const before = await handle.stat(); bytes = await handle.readFile(); const after = await handle.stat();
if (!before.isFile() || before.ino !== info.ino || before.dev !== info.dev || before.size !== info.size || bytes.length !== info.size
|| before.ino !== after.ino || before.dev !== after.dev || before.size !== after.size || before.mtimeMs !== after.mtimeMs || before.ctimeMs !== after.ctimeMs || await realpath(source) !== source) throw new Error("SDK audit changed during capture");
} finally { await handle.close(); }
checkSecret(bytes, secrets);
if (!closedSdkAudit(JSON.parse(bytes.toString("utf8")))) throw new Error("SDK audit has an unsupported observation shape");
const destination = path.join(casePrivateDir, "snapshots"); await mkdir(destination, { recursive: true, mode: 0o700 });
await assertPrivateCellDirectories(owner, destination);
await writeFile(path.join(destination, "source-sdk-privacy-observations.json"), bytes, { flag: "wx", mode: 0o600 });
}