Files
PaperClipAI/tests/runner-e2e/launch.ts
T
DottaandPaperclip ebed7f5237 test(evals): protect Antigravity setup and two-turn delivery (#18190)
## Thinking Path

> - Paperclip manages agents and their work.
> - A provider integration must work through ordinary agent setup and
task controls.
> - The recorded Antigravity walkthrough exposed setup, delivery, and
continuation failures.
> - A repeatable check must verify real work and retain failed attempts.
> - This pull request converts all 47 recorded steps into one fixed
Product E2E journey.
> - Maintainers can rerun the case and inspect independent delivery,
cleanup, and accounting evidence.

## Linked Issues or Issue Description

Builds on merged #18187, #18188, and #18189, and consumes the shared
repairs from merged #18230 and the equivalent manual-policy declaration
from #18222. Refs #18214 for the CI task wrapper. Refs #12503 and #14938
for related legacy adapter work.

**Agent or provider**

Google Antigravity native Runner with a saved managed Google
subscription and the exact Low model.

**Why this adapter is useful**

A user must be able to add an agent, connect its account, and receive
real task results across a follow-up. Picker visibility alone is
insufficient.

**How the agent is invoked**

A fixed Playwright driver uses ordinary product controls. The agent runs
through native ACP. Independent checks use public APIs, actual
downloads, owned process observations, and native lifetime fences.

## What Changed

- Register one explicit-only case:
`antigravity-agent-journey.runner-antigravity.local.saved-subscription-two-turn`.
- Map all 47 source steps and declare saved-account prerequisites,
finite limits, and same-host macOS ARM64 scope.
- Create an agent through the ordinary picker, connect the saved
account, select Low, run Test, and wait for readiness. Save an
instruction marker, deliver a file, and update it through an ordinary
follow-up.
- Require each run's visible completed file activity, exact downloaded
bytes, committed Done result, same tagged conversation, distinct
physical owner birth, released lease, and free native fences. Verify
reload persistence.
- Stop failed native runs that still own a retry. Replay an existing
Stop with its durable request identity and wait for the matching audited
acknowledgement. A terminal reservation without a native intent remains
unconfirmed.
- Keep private evidence in the existing result, report, dashboard, and
accounting pipeline. Retain the paused QA agent and financial history
while preserving the shared account and budget policies.
- Journal only the ordinary wizard's future ID, target, name, intended
harness and expected creator before forwarding its POST unchanged. Bind
the response and interrupted recovery to exact identity reads; a pending
404 stays unconfirmed.
- Serialize private resource snapshots through temporary-file fsync,
atomic rename and directory fsync. Abort forwarding on storage failure
and preserve legacy receipts. If later journal or diagnostic writes
fail, retain the primary error, always close the browser, and attempt
the ordinary failed report.
- Add production-component browser controls and harden task-menu
readiness, alias selection, scroll restoration, failure packaging, and
one hermetic auth-test budget.
- Wait for the existing GitHub recovery fixture to register its next
deadline before asserting the original range; preserve its
nested-transaction and zero-request checks. In the decision-expiry
fixture, verify the exact saved expiry separately from the earlier
global wake of another queue.

## Verification

- Fresh registered acceptance passed on eval source
`9060f3cd01b7c3445ecdf703d72e9cf30fe42b59` and merged serving product
`61bd4ac6919c2309f766a25e62273b074cee3261`. It completed ordinary setup
and two turns with native `.5` and `gemini-3.8-flash-low`. All 16
matchers, 14 checkpoints, and cleanup passed in **165,613 ms**. Actual
downloads were **61 bytes** and **119 bytes**. Both turns passed their
own run-bound browser activity checks and reload persistence.
- Independent read-back confirms Done, a paused retained agent, the same
connected saved default account, and no active owned run. Both accepted
results have the same full tagged conversation and distinct primary
process births. All 21 observed processes retired; both leases released
and all three native fences were free after each turn. The selected `.5`
checkpoints also showed equal agent/backend session IDs; this is an
observed case requirement, not a generic ACP invariant.
- Receipts report 339,273 input tokens, 7,224 output tokens, and 202,846
cached input tokens. Subscription prices are unavailable. Billing
coverage remains incomplete, not free.
- The normal report CLI passed. All 310 files from nine original
attempts and two separate recoveries retained their hashes. The twelve
current screenshots are byte-identical, and 194 local report links are
contained and valid. The eight earlier reports were copied without
regeneration.
- Delivery source `1bf4fb04df04710b82844b3c9355767f46b5f68f` branches
from `d3f2a60e795fc60d64c8b19d1eed4d48657d5492` and merges cleanly with
current master `1d571de5f9f886d66e520800baec8fc2f67d0948`. All 41 prior
`c8b332` authored file blobs remain identical; the 42nd path is the
decision-expiry fixture contract correction. Creation ownership, the
strict durable writer, and the ordinary successful workflow remain
unchanged. Of 43 measured file blobs, 30 remain identical. The
Antigravity oracle, artifact, billing and report gates and its
suite-specific validation block remain unchanged. The actual retained
agent matches its original creator; its original request `setupId` was
not captured and is not reconstructed.
- Later master changes include model-visible workspace guidance and
rejected-terminal-report handling. Their bounded source assessment is
separate: the first actual finish failed, then a same-turn replacement
was accepted and committed. Later read-only results match both original
native-result hashes, and the complete event catalogs have no
`session.failed` event. Master #18228 also expands detection of explicit
JSON file requests. The measured create/update-and-attach `.txt` task
already required publication under both parsers; the publication
validation gate itself is unchanged. Master #18240 changes remote
workspace cache staging and adds an OpenCode index refresh. The selected
local lane returns before those paths. Master #18213 adds optional
numeric turn timing and trace boundaries. Its selected Antigravity
model, text, IDs, result and retirement semantics remain unchanged in
the bounded source review. Local diagnostic rows and awaits can change
event ordinals and latency; the oracle checks relative semantic order.
Collector coverage and latency improvement remain outside this proof.
This does not claim that the whole current runtime is byte-identical to
serving `61bd`, or relabel the measured result as a new run.
- The adopted upstream cache controls passed 3/3 with pinned real Turbo.
Forced work disables local and remote reads and writes; ordinary work
reuses local results. The original genuine-pnpm/Turbo controls remain
historical to `9060`'s earlier write-only policy. The duplicate cache
patch is superseded by #18230. Its qualification-only pnpm addition is
removed because the adopted upstream controls do not require it.
- The E2E typecheck, map checks and 277-control replay are carried from
the unchanged eval source paths. That replay passed, including the 183
Antigravity controls, catalog/connection checks and the separate
product-journey validator. The carried creation controls include 91
config/recovery cases. The final launcher suite passed 19/19, including
five real campaign finalization fault cases; the report suite passed
7/7. These final checks did not rerun the earlier 277-control group. The
canonical production-component browser spec passed in 27.4 s under its
unchanged 60 s limit before the replay. Four paused-composer cases and
the product base's build/typecheck plus 376 focused regressions were
verified earlier.
- The shipped wizard and actual API client cover persistence before
forwarding, lost response JSON, foreign/malformed requests, wrong
response identity/creator and cleanup-error preservation. Real
filesystem controls failed 5/5 before the writer repair and passed 5/5
afterward, preserving the old journal on partial-write, temporary-file
sync and rename faults. Three persistent-sync campaign controls failed
before the finalization repair and passed afterward. All five final
campaign fault cases preserve the primary failure, cleanup failure,
usage and failed report while attempting browser closure. The strict
journal still blocks the next mutation. These are provider-free
ownership controls; the Google task proof remains the original `9060` →
`61bd` acceptance.
- The decision restart check passed in the local 20-case fixture run. A
current-source real-PostgreSQL rollback control reproduced the exact
90,000 ms earlier global wake: the old equality rejected it, while the
exact durable expiry and new future-wake bound passed. The original 60 s
open-state, exact-timestamp, continuation and quiet checks remain. The
broader local run was **FAILED: 19 passed / 1 failed**, with a 15 s
timeout and a 30 s cleanup timeout in an unchanged timestamp-pagination
case. That case passed in canonical `c8b332` CI; the local timeout cause
remains inconclusive. An extra local server typecheck failed against
stale cached Runner declarations; it did not check this excluded test
file. Both limits remain recorded for fresh canonical CI.
- A broader local GitHub recovery group passed 21/22. Its unchanged
pagination case observed an unexpected config request; the log does not
establish the cause. That result remains retained, and fresh
current-head CI is the final broader gate.

- Exact delivery-head CI completed with 55 successes, two skips, and one
optional neutral review. Required `ci / verify` and `ci / e2e`, build,
typecheck, and Runner checks passed. Fresh Greptile is 5/5 with zero
unresolved threads. Optional Paperclip is neutral 4/5; its carried
recommendations are dispositioned against the public Stop contract and
selected live identity evidence.

Seven earlier converted attempts retain their failed grades and partial
evidence. Historical Paid8 and its later read-only per-run UI
observation remain separate from the fresh acceptance above. Unknown
prices and the older partial startup accounting remain explicit. Full
lineage and raw evidence remain private in the normal report bundle.

## Risks

- This case requires an authorized saved personal default Google
subscription and finite operator candidate admission. It does not test
fresh OAuth or establish global qualification.
- The physical observer supports a same-host loopback macOS ARM64
target. Cloud, Linux, and other models remain outside this case.
- A known-cost stop cannot cap missing prices or in-flight subscription
charges. Existing operator policies remain unchanged.
- Raw account/native identities stay private. Missing retirement proof
keeps resources retained. The selected cleanup mode does not test agent
purge.

## Model Used

OpenAI Codex (GPT-6). The host did not expose the exact model revision
or context limit. The model used tools for code edits, tests, browser
automation, and review.

## 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
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

---------

Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-10-29 19:51:32 -05:00

1255 lines
53 KiB
TypeScript

import { readUpgradeArtifacts, verifyUpgradeArtifact, upgradeArtifactKey } from "./upgrade-artifacts.js";
import { recoverWorkerTiming } from "./worker-timing.js";
import { openEvalOperations, type EvalOperations } from "./operations.js";
import { readRecoveredFeedback } from "./feedback-recovery.js";
import { readInterruptedAttemptTrace } from "./trace-recovery.js";
import { EVAL_OBSERVABILITY_SECRET_NAMES, evalBraintrustConfig, finalizeEvalAttemptTrace } from "./braintrust.js";
import { finalizeEvalAttemptThroughRelay } from "./eval-finalization-relay.js";
import { runnerE2EPlaywrightInvocation, runnerE2EPlaywrightShutdownSignal } from "./web-server-command.js";
import { verifyInstalledDaytonaPlugin } from "./installed-daytona-plugin.js";
import { assertInstalledCliSelection, assertInstalledStartupOnly, verifyInstalledCli } from "./installed-cli.js";
import { createProcessTreeOwner, stopOwnedProcessTree } from "./process-tree-owner.js";
import { createRunnerE2ETemporaryRoot } from "./server-config.js";
import { createHash, randomBytes } from "node:crypto";
import { prepareCodexCiSandbox, requiresCodexCiSandbox } from "./codex-ci-sandbox.js";
import { spawn } from "node:child_process";
import { createWriteStream } from "node:fs";
import { createRequire } from "node:module";
import os from "node:os";
import path from "node:path";
import {
access,
chmod,
cp,
lstat,
mkdir,
readFile,
realpath,
readdir,
rm,
symlink,
writeFile,
} from "node:fs/promises";
import { isImmutableDaytonaImage, runnerMatrix } from "./catalog.js";
import { renderRunnerE2EDashboard } from "./dashboard.js";
import { packageEvidence } from "./evidence.js";
import { privateCellEvidenceOwner } from "./manual-private-report-assets.js";
import { readManualPrivateHandoff, sealManualResultEvidence, preserveOriginalFrameworkResult, sealStandardOriginalEvidence, type ManualPrivateHandoff } from "./manual-private-handoff.js";
import { requiresPrivateManualEvidence } from "./manual-private-retention.js";
import { copySharedEvidence } from "./shared-evidence.js";
import { classifyFailure } from "./failure-classifier.js";
import { mustPreserveRecoveryState, shouldKeepFailedDiagnostics } from "./cleanup-verification.js";
import { effectiveAutomaticRetryLimit, executeWithAutomaticRetry } from "./automatic-retry.js";
import { buildRunnerCampaign } from "./history.js";
import { readEvalOperationsReport } from "./operations-report.js";
import {
applyEvalGatewayEnvironment,
buildRunnerE2EProcessEnvironment,
resolvePaperclipRemoteRunnerBinaryForHarness,
resolvePaperclipRunnerBinaryForHarness,
} from "./harness-env.js";
import { assertEmbeddedDatabaseIsolation } from "./instance-isolation.js";
import {
assertSecretFree,
findSecretLeakInDirectory,
isEphemeralCodexRuntimeAuthFile,
isEphemeralPostgresScanFile,
normalizedSecrets,
secretScanFailureEvidence,
sanitizeJson,
} from "./redaction.js";
import {
buildMatrixJobs,
parseRunnerSelectors,
RunnerSelectorError,
selectRunnerExecutions,
} from "./selectors.js";
import { resolveRunnerE2ESource } from "./source.js";
import {
CREDENTIAL_NAMES,
type MatrixExecution,
type RunnerE2EResult,
} from "./types.js";
import { assertRemoteNativeEvidencePrerequisites, assertRunnerE2EPrerequisites } from "./prerequisites.js";
import { assertNativeCompletionSelection, prepareNativeCompletionPreflight, NATIVE_COMPLETION_PREFLIGHT_ENV } from "./native-completion-admission.js";
import { assertNativeInstructionSelection, prepareNativeInstructionPreflight, NATIVE_INSTRUCTION_PREFLIGHT_ENV, NATIVE_INSTRUCTION_SUITE } from "./native-instruction-consolidation.js";
import { prepareStockHarnessPreflight, STOCK_PREFLIGHT_ENV } from "./stock-harness-admission.js";
import {
reapNewDetachedDarwinSharedMemory,
snapshotDarwinSharedMemory,
} from "./shared-memory.js";
import { reserveRunnerE2EServerPort } from "./ports.js";
import {
createResultExitGuard,
enforceResultProcessIntegrity,
} from "./result-exit-guard.js";
import {
observeDescendantProcessTree,
type ObservedProcessGroup,
readProcessTable,
} from "./process-tree.js";
const repositoryRoot = path.resolve(import.meta.dirname, "../..");
const localEnvPath = path.join(repositoryRoot, ".env.runner-e2e.local");
let resultsRoot = path.join(repositoryRoot, "tests/runner-e2e/results");
const manualEvidenceSecrets = new Set<string>();
let manualPrivateHandoff: ManualPrivateHandoff | undefined;
const activeProcessGroups = new Set<number>();
const activeProcessCleanup = new Map<number, Promise<string | null>>();
const activeProcessTerminators = new Map<number, () => void>();
const completedResultExitGraceMs = 120_000;
let cancelled = false;
function cleanId(value: string) {
const result = value
.replace(/[^A-Za-z0-9_.-]+/g, "-")
.replace(/^-+|-+$/g, "");
if (!result)
throw new Error(
`Invalid empty identifier derived from ${JSON.stringify(value)}`,
);
return result;
}
function secret(bytes = 32) {
return randomBytes(bytes).toString("base64url");
}
async function makeDirectoryTreeRemovable(directory: string): Promise<void> {
await chmod(directory, 0o700);
const entries = await readdir(directory, { withFileTypes: true });
await Promise.all(
entries
.filter((entry) => entry.isDirectory() && !entry.isSymbolicLink())
.map((entry) =>
makeDirectoryTreeRemovable(path.join(directory, entry.name)),
),
);
}
function stopProcessGroup(pid: number, signal: NodeJS.Signals = "SIGTERM") {
try {
if (process.platform === "win32") process.kill(pid, signal);
else process.kill(-pid, signal);
} catch {
// The process tree already exited.
}
}
function processGroupIsAlive(pid: number) {
try {
process.kill(process.platform === "win32" ? pid : -pid, 0);
return true;
} catch {
return false;
}
}
async function terminateProcessGroup(pid: number) {
stopProcessGroup(pid, "SIGTERM");
const gracefulDeadline = Date.now() + 5_000;
while (processGroupIsAlive(pid) && Date.now() < gracefulDeadline) {
await new Promise((resolve) => setTimeout(resolve, 50));
}
if (!processGroupIsAlive(pid)) return null;
stopProcessGroup(pid, "SIGKILL");
const forcedDeadline = Date.now() + 5_000;
while (processGroupIsAlive(pid) && Date.now() < forcedDeadline) {
await new Promise((resolve) => setTimeout(resolve, 50));
}
return processGroupIsAlive(pid)
? `Paperclip/Playwright process group ${pid} survived SIGKILL`
: null;
}
function wait(milliseconds: number) {
return new Promise<void>((resolve) => setTimeout(resolve, milliseconds));
}
interface ProcessTreeDiagnostic {
summary: string;
groups: ObservedProcessGroup[];
}
async function processTreeDiagnostic(
rootPid: number,
): Promise<ProcessTreeDiagnostic> {
const table = await readProcessTable();
if (!table) {
return {
summary: `process tree ${rootPid} (member inspection unavailable)`,
groups: [],
};
}
const observed = observeDescendantProcessTree(table, rootPid);
const members = observed.members
.slice(0, 64)
.map(
({ process: candidate, depth }) =>
`pid=${candidate.pid} ppid=${candidate.parentPid} pgid=${candidate.processGroupId} depth=${depth} kind=${candidate.kind}`,
);
const descendantGroupIds = observed.groups
.filter((group) => group.processGroupId !== rootPid)
.map((group) => group.processGroupId);
return {
summary:
members.length > 0
? `process tree ${rootPid}: ${members.join("; ")}; descendant pgids=${descendantGroupIds.join(",") || "none"}`
: `process tree ${rootPid}: no members reported`,
groups: observed.groups,
};
}
for (const signal of ["SIGINT", "SIGTERM", "SIGHUP"] as const) {
process.on(signal, () => {
cancelled = true;
for (const pid of activeProcessGroups) {
const terminate = activeProcessTerminators.get(pid);
if (terminate) {
terminate();
continue;
}
activeProcessCleanup.set(pid, terminateProcessGroup(pid));
}
});
}
async function loadLocalEnvironment(target: NodeJS.ProcessEnv) {
const contents = await readFile(localEnvPath, "utf8").catch(
(error: NodeJS.ErrnoException) => {
if (error.code === "ENOENT") return null;
throw error;
},
);
if (contents === null) return;
for (const [index, rawLine] of contents.split(/\r?\n/).entries()) {
const line = rawLine.trim();
if (!line || line.startsWith("#")) continue;
const match = /^(?:export\s+)?([A-Za-z_][A-Za-z0-9_]*)=(.*)$/.exec(line);
if (!match)
throw new Error(
`Invalid ${path.basename(localEnvPath)} line ${index + 1}`,
);
if (target[match[1]] !== undefined) continue;
let value = match[2].trim();
if (
(value.startsWith('"') && value.endsWith('"')) ||
(value.startsWith("'") && value.endsWith("'"))
) {
value = value.slice(1, -1);
}
target[match[1]] = value;
}
}
async function prepareProviderPath(
temporaryRoot: string,
inheritedPath: string | undefined,
) {
const toolBin = path.join(temporaryRoot, "provider-bin");
await mkdir(toolBin, { recursive: true });
const runnerRequire = createRequire(
path.join(repositoryRoot, "packages/paperclip-runner/package.json"),
);
const codexAcpPackage = runnerRequire.resolve(
"@agentclientprotocol/codex-acp/package.json",
);
const codexRequire = createRequire(codexAcpPackage);
const codexPackage = codexRequire.resolve("@openai/codex/package.json");
const codexManifest = JSON.parse(await readFile(codexPackage, "utf8")) as {
bin?: string | Record<string, string>;
};
const codexBin =
typeof codexManifest.bin === "string"
? codexManifest.bin
: codexManifest.bin?.codex;
if (!codexBin)
throw new Error(
"Production Codex ACP dependency does not expose its pinned Codex executable",
);
await symlink(
path.resolve(path.dirname(codexPackage), codexBin),
path.join(toolBin, "codex"),
);
const packageBin = path.join(
repositoryRoot,
"packages/paperclip-runner/node_modules/.bin",
);
return [toolBin, packageBin, inheritedPath]
.filter(Boolean)
.join(path.delimiter);
}
async function runProcess(
invocation: { command: string; args: string[] },
env: NodeJS.ProcessEnv,
timeoutMs: number | null,
logPath: string,
completionPaths: readonly string[],
interactive: boolean,
) {
const log = createWriteStream(logPath, { flags: "a", mode: 0o600 });
const child = spawn(invocation.command, invocation.args, {
cwd: repositoryRoot,
env,
stdio: ["inherit", "pipe", "pipe"],
detached: process.platform !== "win32",
});
if (!child.pid) throw new Error("Failed to start Playwright");
activeProcessGroups.add(child.pid);
const processOwner = createProcessTreeOwner(child);
let outputTail = "";
const recordOutput = (chunk: Buffer, destination: NodeJS.WriteStream) => {
destination.write(chunk);
log.write(chunk);
outputTail += chunk.toString("utf8");
if (outputTail.length > 1024 * 1024)
outputTail = outputTail.slice(-1024 * 1024);
};
child.stdout?.on("data", (chunk: Buffer) =>
recordOutput(chunk, process.stdout),
);
child.stderr?.on("data", (chunk: Buffer) =>
recordOutput(chunk, process.stderr),
);
let childSettled = false;
let postResultStallError: string | null = null;
let boundedCleanup: Promise<string | null> | undefined;
let cleanupSettled!: () => void;
const cleanupFinished = new Promise<number>(resolve => { cleanupSettled = () => resolve(1); });
const stopChildTree = (_diagnostic?: ProcessTreeDiagnostic) => {
if (boundedCleanup) return;
boundedCleanup = stopOwnedProcessTree(child, processOwner, 45_000, 5_000, runnerE2EPlaywrightShutdownSignal)
.then(() => null, error => error instanceof Error ? error.message : String(error))
.finally(cleanupSettled);
activeProcessCleanup.set(child.pid!, boundedCleanup);
};
activeProcessTerminators.set(child.pid, stopChildTree);
const resultExitGuard = createResultExitGuard({
resultPaths: completionPaths,
interactive,
graceMs: completedResultExitGraceMs,
now: Date.now,
pathExists: (candidate) =>
access(candidate).then(
() => true,
() => false,
),
onExpired: async () => {
const diagnostic = await processTreeDiagnostic(child.pid!);
if (childSettled) return;
postResultStallError = `Playwright remained alive for ${completedResultExitGraceMs}ms after every result was written`;
recordOutput(
Buffer.from(
`\n${postResultStallError}; forcing bounded cleanup. ${diagnostic.summary}\n`,
),
process.stderr,
);
stopChildTree(diagnostic);
},
});
const completionPoll = resultExitGuard.enabled
? setInterval(() => {
void resultExitGuard.poll().catch(() => {
if (childSettled || postResultStallError) return;
postResultStallError =
"Completed-result exit guard failed during bounded inspection";
recordOutput(
Buffer.from(`\n${postResultStallError}; forcing cleanup.\n`),
process.stderr,
);
stopChildTree();
});
}, 500)
: undefined;
completionPoll?.unref();
let timedOut = false;
const timer =
timeoutMs === null
? undefined
: setTimeout(() => {
timedOut = true;
stopChildTree();
}, timeoutMs);
timer?.unref();
let spawnError: string | null = null;
const childExit = new Promise<number>((resolve, reject) => {
child.once("error", reject);
child.once("exit", (code) => {
childSettled = true;
resolve(code ?? 1);
});
}).catch((error) => {
childSettled = true;
spawnError = error instanceof Error ? error.message : String(error);
return 1;
});
const exitCode = await Promise.race([childExit, cleanupFinished]);
if (timer) clearTimeout(timer);
if (completionPoll) clearInterval(completionPoll);
// Even a successful launcher exit can leave an already-observed detached
// server behind. Retire that retained tree, never only the root group.
stopChildTree();
const processCleanupError = await boundedCleanup!;
processOwner.stopObserving();
// Even a failed direct-child kill must not hold cancellation forever or let
// inherited pipes write into an ended log. The cleanup error remains fatal.
if (processCleanupError) {
child.stdout?.destroy();
child.stderr?.destroy();
if (child.exitCode === null && child.signalCode === null) child.unref();
}
if (child.pid) {
activeProcessGroups.delete(child.pid);
activeProcessCleanup.delete(child.pid);
activeProcessTerminators.delete(child.pid);
}
await new Promise<void>((resolve) => log.end(resolve));
return {
exitCode,
timedOut,
postResultStallError,
processCleanupError,
spawnError,
outputTail,
};
}
function syntheticResult(
execution: MatrixExecution,
attempt: number,
startedAtMs: number,
error: string,
failureClass: RunnerE2EResult["failureClass"] = "transient_infrastructure",
): RunnerE2EResult {
const finishedAtMs = Date.now();
return {
schema: "paperclip.runner-e2e.result/v2",
executionId: execution.id,
suiteId: execution.suite.id,
suiteDefinitionHash: execution.suiteDefinitionHash,
source: resolveRunnerE2ESource(),
...(execution.profile.ranking
? { rankingSnapshot: execution.profile.ranking }
: {}),
attempt,
status: "failed",
failureClass,
error,
profileId: execution.profile.id,
environmentId: execution.environment.id,
caseId: execution.task.id,
provider: execution.profile.provider,
model: execution.profile.model,
runtimeMode: execution.profile.expectedRuntimeMode,
startedAt: new Date(startedAtMs).toISOString(),
finishedAt: new Date(finishedAtMs).toISOString(),
durationMs: finishedAtMs - startedAtMs,
cleanup: "not_started",
};
}
async function readResult(resultPath: string, fallback: RunnerE2EResult) {
try {
return JSON.parse(await readFile(resultPath, "utf8")) as RunnerE2EResult;
} catch {
return fallback;
}
}
async function runAttempt(input: {
executions: readonly MatrixExecution[];
attempt: number;
campaignId: string;
options: ReturnType<typeof parseRunnerSelectors>;
operations?: EvalOperations;
}): Promise<RunnerE2EResult[]> {
const { executions, attempt, campaignId, options } = input;
const accountingCampaignId = input.operations?.manifest.limits.campaignId ?? process.env.PAPERCLIP_E2E_ACCOUNTING_CAMPAIGN_ID ?? campaignId;
const accountingAttemptId = randomBytes(16).toString("hex");
const traceConfig = evalBraintrustConfig({ ...process.env, PAPERCLIP_E2E_OTLP_RELAY_URL: undefined });
const execution = executions[0];
if (!execution) throw new Error("A runner E2E cell must contain a task case");
if (
executions.some(
(candidate) =>
candidate.profile.id !== execution.profile.id ||
candidate.environment.id !== execution.environment.id,
)
) {
throw new Error(
"One isolated runner E2E harness cannot mix profiles or environments",
);
}
if (process.env.GITHUB_ACTIONS === "true" && !input.operations && executions.some(candidate => requiresPrivateManualEvidence(candidate.id)) && !manualPrivateHandoff) throw new Error("Sensitive manual evidence requires verified restricted retention before setup");
const upgrades = await readUpgradeArtifacts(process.env[upgradeArtifactKey]);
if (upgrades) {
// All attempts and the browser use A's installed fixture admission. The
// supervisor changes only the server artifact; the original sandbox lives on.
Object.assign(process.env, upgrades.artifacts.A.pins);
for (const artifact of [upgrades.artifacts.A, upgrades.artifacts.B]) {
await verifyUpgradeArtifact(artifact, process.env, executions);
}
}
assertInstalledCliSelection(process.env, executions);
const installedCli = await verifyInstalledCli(process.env, executions);
const installedPlugin = await verifyInstalledDaytonaPlugin(process.env, executions);
const startedAtMs = Date.now();
const sharedMemoryBaseline = snapshotDarwinSharedMemory();
const temporaryParent = await realpath(os.tmpdir());
const temporaryRoot = await createRunnerE2ETemporaryRoot(temporaryParent);
const privateDir = manualPrivateHandoff ? path.join(manualPrivateHandoff.root, "capture") : path.join(temporaryRoot, "artifacts-private");
const publishedResults: RunnerE2EResult[] = [];
const publishedResultPaths = new Map<string, string>();
let attemptSecrets: string[] = [];
let processCleanupFailed = false;
let startupReceiptPath: string | undefined;
let gateway: Awaited<ReturnType<EvalOperations["start"]>> | undefined;
const rawCleanupResults: Array<{ cleanup: string; synthetic: boolean; status: string }> = [];
try {
const paperclipHome = path.join(temporaryRoot, "paperclip-home");
const workspace = path.join(temporaryRoot, "workspace");
const instanceId = `runner-e2e-${randomBytes(8).toString("hex")}`;
const configPath = path.join(
paperclipHome,
"instances",
instanceId,
"config.json",
);
const port = await reserveRunnerE2EServerPort();
await Promise.all([
mkdir(paperclipHome, { recursive: true }),
mkdir(workspace, { recursive: true }),
mkdir(privateDir, { recursive: true }),
]);
const providerPath = await prepareProviderPath(
temporaryRoot,
process.env.PATH,
);
if (requiresCodexCiSandbox(execution)) {
await prepareCodexCiSandbox(repositoryRoot, temporaryRoot);
}
const agentJwtSecret = secret(48);
const decisionSigningSecret = secret(48);
const toolActionSigningSecret = secret(48);
const betterAuthSecret = secret(48);
gateway = await input.operations?.start(execution, accountingAttemptId, attempt);
const credentialSources = {
PAPERCLIP_E2E_GATEWAY_TOKEN: gateway?.token,
PAPERCLIP_E2E_OTLP_RELAY_TOKEN: gateway?.observability?.token,
...Object.fromEntries([...CREDENTIAL_NAMES, ...EVAL_OBSERVABILITY_SECRET_NAMES].map((name) => [name, process.env[name]])),
PAPERCLIP_AGENT_JWT_SECRET: agentJwtSecret,
PAPERCLIP_DECISION_SIGNING_SECRET: decisionSigningSecret,
PAPERCLIP_TOOL_ACTION_SIGNING_SECRET: toolActionSigningSecret,
BETTER_AUTH_SECRET: betterAuthSecret,
};
const credentials = normalizedSecrets(Object.values(credentialSources));
attemptSecrets = credentials;
if (manualPrivateHandoff) for (const credential of credentials) manualEvidenceSecrets.add(credential);
if (installedCli) await writeFile(path.join(privateDir, "installed-cli-admission.json"), `${JSON.stringify(installedCli, null, 2)}\n`, { mode: 0o600 });
if (installedPlugin) await writeFile(path.join(privateDir, "installed-daytona-plugin-admission.json"), `${JSON.stringify(installedPlugin, null, 2)}\n`, { mode: 0o600 });
const runnerBinary = installedCli ? undefined : resolvePaperclipRunnerBinaryForHarness(
executions,
repositoryRoot,
);
const childEnv: NodeJS.ProcessEnv = {
...buildRunnerE2EProcessEnvironment(process.env, executions),
PATH: providerPath,
PAPERCLIP_RUNNER_E2E_EXECUTION_IDS: JSON.stringify(
executions.map((candidate) => candidate.id),
),
PAPERCLIP_RUNNER_E2E_ATTEMPT: String(attempt),
PAPERCLIP_E2E_ACCOUNTING_CAMPAIGN_ID: accountingCampaignId,
PAPERCLIP_E2E_ACCOUNTING_ATTEMPT_ID: accountingAttemptId,
PAPERCLIP_RUNNER_E2E_INSTALLED_STARTUP_ONLY: options.installedStartupOnly ? "1" : undefined,
PAPERCLIP_RUNNER_E2E_PUBLIC_MCP: executions.some(candidate => candidate.task.flow === "public_mcp") ? "1" : "0",
PAPERCLIP_RUNNER_E2E_PORT: String(port),
PAPERCLIP_RUNNER_E2E_TEMP_ROOT: temporaryRoot,
PAPERCLIP_RUNNER_E2E_PRIVATE_DIR: privateDir,
PAPERCLIP_RUNNER_E2E_WORKSPACE: workspace,
PAPERCLIP_RUNNER_E2E_SERVER_LOG: path.join(privateDir, "server.log"),
PAPERCLIP_RUNNER_BINARY: runnerBinary,
PAPERCLIP_RUNNER_REMOTE_BINARY_PATH:
installedCli ? undefined : resolvePaperclipRemoteRunnerBinaryForHarness(executions, runnerBinary),
// Vite's optimized dependency cache embeds revision query strings. A
// private per-attempt cache prevents an earlier cell or local rebuild
// from producing `504 Outdated Optimize Dep` during browser bootstrap.
PAPERCLIP_VITE_CACHE_DIR: path.join(temporaryRoot, "vite-cache"),
PAPERCLIP_RUNNER_E2E_TEST_TIMEOUT_MS: String(
Math.max(
...executions.map(
(candidate) =>
candidate.task.attemptTimeoutMs[candidate.environment.id],
),
) + 90_000,
),
PAPERCLIP_HOME: paperclipHome,
PAPERCLIP_INSTANCE_ID: instanceId,
PAPERCLIP_CONFIG: configPath,
PAPERCLIP_AGENT_JWT_SECRET: agentJwtSecret,
PAPERCLIP_DECISION_SIGNING_SECRET: decisionSigningSecret,
PAPERCLIP_TOOL_ACTION_SIGNING_SECRET: toolActionSigningSecret,
BETTER_AUTH_SECRET: betterAuthSecret,
};
if (gateway) {
applyEvalGatewayEnvironment(childEnv, execution.profile.credential, gateway);
// The fixture/server controller needs Daytona provisioning access. The
// remote provider still receives only the scoped model and OTLP tokens.
if (execution.environment.id === "daytona") childEnv.DAYTONA_API_KEY = process.env.DAYTONA_API_KEY;
}
// The database URLs are stripped here and again at the Playwright web-server
// boundary so a developer's shell can never redirect this paid test.
delete childEnv.DATABASE_URL;
delete childEnv.DATABASE_MIGRATION_URL;
const playwrightArgs = [
"test",
"--config",
"tests/runner-e2e/playwright.config.ts",
...(options.headed ? ["--headed"] : []),
...(options.ui ? ["--ui"] : []),
...(options.debug ? ["--debug"] : []),
];
const watchdog =
options.ui || options.debug
? null
: executions.reduce(
(total, candidate) =>
total +
candidate.task.attemptTimeoutMs[candidate.environment.id] +
90_000,
0,
) +
5 * 60_000;
const processResult = await runProcess(
runnerE2EPlaywrightInvocation(repositoryRoot, playwrightArgs),
childEnv,
watchdog,
path.join(privateDir, "playwright.log"),
executions.map((candidate) =>
path.join(privateDir, "cases", candidate.task.id, "result.json"),
),
options.ui || options.debug,
);
processCleanupFailed = processResult.processCleanupError !== null;
if (options.installedStartupOnly) {
const proofDir = path.join(resultsRoot, campaignId, "installed-startup-only");
await mkdir(proofDir, { recursive: true });
// This is private setup evidence, never a successful provider case/result.
await cp(privateDir, path.join(proofDir, "private"), { recursive: true });
const probe = JSON.parse(await readFile(path.join(privateDir, "installed-startup-only.json"), "utf8"));
if (processResult.exitCode !== 0 || processResult.timedOut || processResult.spawnError || processCleanupFailed || probe.status !== "health_ui_zero_company_passed" || probe.providerCalls !== 0 || probe.qualified !== false) throw new Error("Installed startup-only actual launch chain failed");
await assertEmbeddedDatabaseIsolation(configPath, temporaryRoot);
startupReceiptPath = path.join(proofDir, "receipt.json");
await writeFile(startupReceiptPath, `${JSON.stringify({ status: "installed_launch_chain_pending_scratch_cleanup", qualified: false, providerCalls: 0, credentialsLoaded: false, installedCli, processResult, probe }, null, 2)}\n`, { mode: 0o600 });
return [];
}
const processFailure = processResult.spawnError
? `Playwright failed to start: ${processResult.spawnError}`
: processResult.timedOut
? `Harness process exceeded ${Math.round((watchdog ?? 0) / 1000)} seconds`
: `Playwright exited ${processResult.exitCode} before writing a result`;
const processFailureClass =
processResult.spawnError || processResult.timedOut
? "transient_infrastructure"
: classifyFailure(
new Error(
processResult.outputTail
? `${processFailure}\n${processResult.outputTail}`
: processFailure,
),
);
const results = await Promise.all(
executions.map(async (candidate) => {
const resultPath = path.join(
privateDir,
"cases",
candidate.task.id,
"result.json",
);
const fallback = syntheticResult(
candidate,
attempt,
startedAtMs,
processFailure,
processFailureClass,
);
const result = await readResult(resultPath, fallback);
if (!result.workerTiming) result.workerTiming = await recoverWorkerTiming(path.join(path.dirname(resultPath), "snapshots", "worker-timing.json"),
{ executionId: candidate.id, attempt, startedAt: result.startedAt, finishedAt: result.finishedAt });
if (result === fallback) {
result.feedback = await readRecoveredFeedback(path.dirname(resultPath), {
campaignId: accountingCampaignId, attemptId: accountingAttemptId, executionId: candidate.id,
});
if (traceConfig) result.tracing = await readInterruptedAttemptTrace(path.dirname(resultPath), {
projectId: traceConfig.projectId, executionId: candidate.id,
});
}
// Keep cleanup authority before evidence copying/publication can fail.
rawCleanupResults.push({ cleanup: result.cleanup, synthetic: result === fallback, status: result.status });
return enforceResultProcessIntegrity(result, processResult);
}),
);
let isolationError: unknown;
try {
await assertEmbeddedDatabaseIsolation(configPath, temporaryRoot);
} catch (error) {
isolationError = error;
}
let persistedStateError: unknown;
let persistedStateFinding: { file: string; stateArea: string; reason: string; matchCategory?: string } | undefined;
// Onboarding evaluates behavior; credential persistence belongs to a separate layer.
// Keep artifact redaction/publication checks independent of this filesystem scan.
const persistenceCheckedExecutions = executions.filter(
(candidate) => candidate.suite.id !== "first-task",
);
if (persistenceCheckedExecutions.length > 0) {
try {
const expectedEphemeralCredentials = new Set<string>();
for (const [label, directory] of [
["Paperclip home", paperclipHome],
["workspace", workspace],
] as const) {
while (true) {
// The managed Codex home may legitimately contain upstream source-code
// fixtures with fake `sk-*` strings. Reject exact campaign credentials.
const leak = await findSecretLeakInDirectory(directory, credentials, {
includeShapes: false,
sources: credentialSources,
...(label === "Paperclip home" ? { privateCodexRuntimeIdentityRoot: paperclipHome } : {}),
ignoreFile: (file) => expectedEphemeralCredentials.has(file),
allowDisappearedFile: (file) =>
label === "Paperclip home" &&
isEphemeralPostgresScanFile(paperclipHome, file),
});
if (!leak) break;
const isManagedCodexRuntimeAuth =
label === "Paperclip home" &&
isEphemeralCodexRuntimeAuthFile(paperclipHome, leak.file);
if (isManagedCodexRuntimeAuth) {
const metadata = await lstat(leak.file);
if (metadata.isFile() && (metadata.mode & 0o777) === 0o600) {
// Codex CLI API-key mode requires this one runtime auth file. It
// lives only in the disposable cell root, is never published,
// must be owner-only, and is removed with the root below.
expectedEphemeralCredentials.add(leak.file);
continue;
}
}
persistedStateFinding = { ...leak, stateArea: label, file: path.relative(temporaryRoot, leak.file) };
throw new Error(`Secret leak in persisted ${label} state at ${persistedStateFinding.file}: ${leak.reason}${leak.matchCategory ? ` (${leak.matchCategory})` : ""}`);
}
}
} catch (error) {
persistedStateError = error;
}
}
for (const [index, candidate] of executions.entries()) {
let result = {
...results[index],
accountingIdentity: { campaignId: accountingCampaignId, attemptId: accountingAttemptId },
};
if (
isolationError &&
result.failureClass !== "cleanup_failure" &&
result.failureClass !== "secret_leak"
) {
const isolationMessage =
isolationError instanceof Error
? isolationError.message
: String(isolationError);
const configWasUnavailableDuringTransientBootstrap =
result.failureClass === "transient_infrastructure" &&
typeof isolationError === "object" &&
isolationError !== null &&
"code" in isolationError &&
isolationError.code === "ENOENT";
result = {
...result,
status: "failed",
failureClass: configWasUnavailableDuringTransientBootstrap
? result.failureClass
: "permanent_infrastructure",
error: configWasUnavailableDuringTransientBootstrap
? `${result.error}; isolated config could not be inspected after bootstrap failure: ${isolationMessage}`
: isolationMessage,
};
}
if (persistedStateError && persistenceCheckedExecutions.includes(candidate)) {
const persistedStateMessage =
persistedStateError instanceof Error
? persistedStateError.message
: String(persistedStateError);
const persistedStateClass = classifyFailure(persistedStateError);
if (
persistedStateClass === "secret_leak" ||
(result.failureClass !== "secret_leak" &&
result.failureClass !== "cleanup_failure")
) {
result = {
...result,
status: "failed",
failureClass:
persistedStateClass === "secret_leak"
? "secret_leak"
: "permanent_infrastructure",
error: persistedStateMessage,
};
} else {
result = {
...result,
error: `${result.error}; persisted-state scan also failed: ${persistedStateMessage}`,
};
}
}
const casePrivateDir = path.join(privateDir, "cases", candidate.task.id);
const resultPath = path.join(casePrivateDir, "result.json");
const uploadDir = path.join(
resultsRoot,
campaignId,
candidate.suite.id,
candidate.profile.id,
candidate.environment.id,
candidate.task.id,
`attempt-${attempt}`,
);
await mkdir(casePrivateDir, { recursive: true });
await copySharedEvidence(privateDir, casePrivateDir, temporaryRoot, credentials);
if (persistedStateError && persistenceCheckedExecutions.includes(candidate)) {
const snapshots = path.join(casePrivateDir, "snapshots");
await mkdir(snapshots, { recursive: true });
const original = results[index];
await writeFile(path.join(snapshots, "persisted-state-scan.json"), `${JSON.stringify(secretScanFailureEvidence({
scanError: persistedStateError instanceof Error ? persistedStateError.message : String(persistedStateError),
finding: persistedStateFinding,
originalResult: { status: original.status, failureClass: original.failureClass, error: original.error },
}, credentials), null, 2)}\n`, { mode: 0o600 });
}
if (manualPrivateHandoff) await preserveOriginalFrameworkResult(casePrivateDir);
await writeFile(
resultPath,
`${JSON.stringify(sanitizeJson(result, credentials), null, 2)}\n`,
"utf8",
);
let evidence = await packageEvidence({
privateDir: casePrivateDir,
uploadDir,
secrets: credentials,
expectPassScreenshot: result.status === "passed",
...(manualPrivateHandoff ? { privateCell: { owner: await privateCellEvidenceOwner(manualPrivateHandoff, candidate.id), layout: "original" as const } } : {}),
});
const originalEvidence = evidence, originalFeaturePassed = result.status === "passed";
if (evidence.leaks.length > 0 || evidence.missing.length > 0) {
result = {
...result,
status: "failed",
failureClass:
evidence.leaks.length > 0
? "secret_leak"
: "permanent_infrastructure",
error:
evidence.leaks.length > 0
? `Secret leak rejected from evidence: ${evidence.leaks.map((leak) => leak.file).join(", ")}`
: `Required evidence missing: ${evidence.missing.join(", ")}`,
};
await writeFile(
resultPath,
`${JSON.stringify(sanitizeJson(result, credentials), null, 2)}\n`,
"utf8",
);
evidence = await packageEvidence({
privateDir: casePrivateDir,
uploadDir,
secrets: credentials,
expectPassScreenshot: false,
...(manualPrivateHandoff ? { privateCell: { owner: await privateCellEvidenceOwner(manualPrivateHandoff, candidate.id), layout: "original" as const } } : {}),
});
}
if (manualPrivateHandoff) await sealStandardOriginalEvidence({ context: manualPrivateHandoff,
executionId: candidate.id, source: casePrivateDir, evidence: originalEvidence, secrets: credentials, expectPassScreenshot: originalFeaturePassed });
console.log(
`${result.status === "passed" ? "PASS" : "FAIL"} ${candidate.id} attempt ${attempt} -> ${uploadDir}`,
);
publishedResults.push(result);
publishedResultPaths.set(
candidate.id,
path.join(uploadDir, "result.json"),
);
}
return [...publishedResults];
} finally {
let gatewayCleanupError: unknown;
try { await gateway?.close(); } catch (error) { gatewayCleanupError = error; }
if (!processCleanupFailed) reapNewDetachedDarwinSharedMemory(sharedMemoryBaseline);
let cleanupError: unknown;
const resourceAdmissionStarted = await access(path.join(privateDir, "resource-admission-started"))
.then(() => true).catch((error: NodeJS.ErrnoException) => error.code !== "ENOENT");
if (mustPreserveRecoveryState({ processCleanupFailed, resourceAdmissionStarted, results: rawCleanupResults })) {
// Remote allocation cleanup is journaled in this instance's database.
// Deleting it after the controller exits would strand uncertain creates.
await chmod(temporaryRoot, 0o700);
cleanupError = new Error(`Preserving private recovery state after unconfirmed cleanup: ${temporaryRoot}`);
} else if (shouldKeepFailedDiagnostics({
enabled: process.env.PAPERCLIP_RUNNER_E2E_KEEP_FAILED_PRIVATE === "1",
expectedResults: executions.length,
results: publishedResults,
})) {
// Explicit diagnosis only. Confirmed resource cleanup stays confirmed;
// keep private traces for investigation without publishing provider data.
await chmod(temporaryRoot, 0o700);
console.warn(`Retained private failed-case diagnostics: ${temporaryRoot}`);
} else if (
temporaryRoot.startsWith(`${temporaryParent}${path.sep}paperclip-runner-e2e-`)
) {
for (let cleanupAttempt = 1; cleanupAttempt <= 3; cleanupAttempt += 1) {
try {
await rm(temporaryRoot, { recursive: true, force: true });
cleanupError = undefined;
break;
} catch (error) {
cleanupError = error;
if (cleanupAttempt < 3) {
await makeDirectoryTreeRemovable(temporaryRoot).catch(() => {});
await new Promise((resolve) =>
setTimeout(resolve, cleanupAttempt * 250),
);
}
}
}
} else {
cleanupError = new Error(
`Refusing to remove unexpected temporary path ${temporaryRoot}`,
);
}
if (gatewayCleanupError) cleanupError = new AggregateError([cleanupError, gatewayCleanupError].filter(Boolean), "Eval gateway cleanup failed");
if (startupReceiptPath) {
const receipt = JSON.parse(await readFile(startupReceiptPath, "utf8"));
receipt.status = cleanupError ? "installed_launch_chain_cleanup_failed" : "installed_launch_chain_passed";
receipt.temporaryRootRemoved = !cleanupError;
await writeFile(startupReceiptPath, `${JSON.stringify(receipt, null, 2)}\n`, { mode: 0o600 });
}
if (cleanupError) {
const message = `Temporary runner E2E state cleanup failed at ${temporaryRoot}: ${cleanupError instanceof Error ? cleanupError.message : String(cleanupError)}`;
for (const publishedResult of publishedResults) {
const publishedResultPath = publishedResultPaths.get(
publishedResult.executionId,
);
if (!publishedResultPath) continue;
Object.assign(publishedResult, {
status: "failed",
failureClass: "cleanup_failure",
error: publishedResult.error ? `${publishedResult.error}; ${message}` : message,
cleanup: "failed",
} satisfies Partial<RunnerE2EResult>);
const safeResult = `${JSON.stringify(
sanitizeJson(publishedResult, attemptSecrets),
null,
2,
)}\n`;
assertSecretFree(safeResult, attemptSecrets, publishedResultPath);
await writeFile(publishedResultPath, safeResult, "utf8");
}
// Cleanup failure is a failed cell, but must not suppress later cells in
// the same campaign. The result above carries the terminal failure.
console.error(message);
if (options.installedStartupOnly) throw new Error(message);
}
for (const result of publishedResults) {
const resultPath = publishedResultPaths.get(result.executionId);
const candidate = executions.find(candidate => candidate.id === result.executionId);
if (!traceConfig || !result.tracing || !resultPath || !candidate) continue;
const finalizationInput = {
receipt: result.tracing,
identity: { campaignId: accountingCampaignId, attemptId: accountingAttemptId, executionId: candidate.id,
provider: candidate.profile.provider, model: candidate.profile.model, environment: candidate.environment.id },
provenance: { suiteId: candidate.suite.id, suiteDefinitionHash: candidate.suiteDefinitionHash, profileId: candidate.profile.id,
caseId: candidate.task.id, sourceRevision: resolveRunnerE2ESource().sha },
result,
};
result.tracing.finalization = input.operations
? await finalizeEvalAttemptThroughRelay(traceConfig, finalizationInput,
path.join(`${input.operations.manifest.ledgerPath}.otlp`, `${createHash("sha256").update(accountingAttemptId).digest("hex")}-finalization.jsonl`))
: await finalizeEvalAttemptTrace(traceConfig, finalizationInput);
const safeResult = `${JSON.stringify(sanitizeJson(result, attemptSecrets), null, 2)}\n`;
assertSecretFree(safeResult, attemptSecrets, resultPath);
await writeFile(resultPath, safeResult, "utf8");
}
}
}
function printList(executions: readonly MatrixExecution[]) {
console.log("ID\tSUITE\tGENERATION\tPROVIDER\tMODEL\tCREDENTIALS");
for (const execution of executions) {
console.log(
[
execution.id,
execution.suite.id,
execution.profile.generation,
execution.profile.provider,
execution.profile.model,
execution.requiredCredentials.join(","),
].join("\t"),
);
}
}
async function runExecutionWithRetry(input: {
execution: MatrixExecution;
campaignId: string;
options: ReturnType<typeof parseRunnerSelectors>;
onAttempt: (result: RunnerE2EResult) => void;
operations?: EvalOperations;
}): Promise<RunnerE2EResult> {
const { execution, campaignId, options } = input;
return executeWithAutomaticRetry({
task: execution.task, options,
qualificationCandidate: execution.profile.qualificationCandidate !== undefined,
cancelled: () => cancelled,
onRetry: result => console.warn(
`Retrying ${execution.id} in a fresh isolated harness after ${result.failureClass!.replaceAll("_", " ")}`,
),
runAttempt: async attempt => {
const [result] = await runAttempt({ executions: [execution], attempt, campaignId, options, operations: input.operations });
if (!result) throw new Error(`No result produced for ${execution.id} attempt ${attempt}`);
input.onAttempt(result);
return result;
},
});
}
async function runWithConcurrency<T, R>(
values: readonly T[],
concurrency: number,
worker: (value: T) => Promise<R>,
): Promise<R[]> {
const results = new Array<R>(values.length);
let cursor = 0;
const workers = Array.from(
{ length: Math.min(concurrency, values.length) },
async () => {
while (true) {
const index = cursor;
cursor += 1;
if (index >= values.length) return;
if (cancelled) throw new Error("Runner E2E campaign cancelled");
results[index] = await worker(values[index]!);
}
},
);
const settled = await Promise.allSettled(workers);
const failures = settled.filter((result): result is PromiseRejectedResult => result.status === "rejected");
if (failures.length) throw new AggregateError(failures.map(result => result.reason), "Eval campaign workers failed");
return results;
}
async function main() {
const args = process.argv.slice(2);
if (args.includes("--target-config") || args.includes("product-journeys")) {
const { launchProductJourneys } = await import("./product-journeys/launch.js");
return launchProductJourneys(args, repositoryRoot);
}
let options: ReturnType<typeof parseRunnerSelectors>;
try {
options = parseRunnerSelectors(process.argv.slice(2));
} catch (error) {
if (error instanceof RunnerSelectorError)
throw new Error(`${error.message}\nUse --list to inspect valid cells.`);
throw error;
}
const executions = selectRunnerExecutions(options, runnerMatrix);
manualPrivateHandoff = await readManualPrivateHandoff(process.env, executions.map(execution => execution.id));
if (manualPrivateHandoff) resultsRoot = path.join(manualPrivateHandoff.root, "results");
if (options.installedStartupOnly) {
assertInstalledStartupOnly(process.env, executions, options);
assertRunnerE2EPrerequisites(executions);
const campaignId = cleanId(process.env.PAPERCLIP_E2E_CAMPAIGN_ID ?? `installed-startup-${Date.now()}`);
await runAttempt({ executions, attempt: 1, campaignId, options });
console.log(`PASS installed launch chain (no provider qualification) -> ${path.join(resultsRoot, campaignId, "installed-startup-only")}`);
return;
}
if (options.list) {
printList(executions);
return;
}
if (options.matrixJson) {
const jobs = buildMatrixJobs(executions);
console.log(
JSON.stringify({
include: jobs,
needsDaytona: jobs.some((job) => job.needsDaytona),
executionIds: executions.map((execution) => execution.id),
}),
);
return;
}
if (process.env.PAPERCLIP_E2E_OPERATIONS_MANIFEST && executions.some(execution => execution.task.flow === "provider_connection")) throw new Error("Provider connection campaign has no budgeted routing qualification");
if (executions.some(execution => execution.task.flow === "antigravity_agent_journey")) {
if (options.connectionConfig || options.githubReviewConfig || options.provisionGitHubRepository) throw new Error("Use --antigravity-journey-config for the Antigravity journey.");
const { runAntigravityJourneyCampaign } = await import("./antigravity-journey-launch.js");
await runAntigravityJourneyCampaign({ executions, catalog: runnerMatrix, configFile: options.antigravityJourneyConfig, repositoryRoot,
preflight: options.antigravityJourneyPreflight, recoverFile: options.antigravityJourneyRecover });
return;
}
if (options.antigravityJourneyConfig || options.antigravityJourneyPreflight || options.antigravityJourneyRecover) throw new Error("Antigravity journey options require its explicit registered cell.");
if (executions.some(execution => execution.task.flow === "github_review_journey")) {
if (options.connectionConfig) throw new Error("Use --github-review-config for the GitHub journey.");
const { runGitHubReviewCampaign } = await import("./github-review-launch.js");
await runGitHubReviewCampaign({ executions, catalog: runnerMatrix, configFile: options.githubReviewConfig, repositoryRoot, provisionRepository: options.provisionGitHubRepository });
return;
}
if (options.githubReviewConfig || options.provisionGitHubRepository) throw new Error("GitHub journey options require an explicit GitHub review journey selection.");
if (executions.some(execution => execution.task.flow === "provider_connection")) {
const { runConnectionCampaign } = await import("./connection-launch.js");
await runConnectionCampaign({ executions, catalog: runnerMatrix, configFile: options.connectionConfig, repositoryRoot });
return;
}
if (options.connectionConfig) throw new Error("--connection-config is only supported by the provider-connections suite");
// Keep admission before local-env loading and credential checks. Pending
// profiles remain discoverable, but cannot reach a provider.
assertRunnerE2EPrerequisites(executions);
assertNativeCompletionSelection(executions);
assertNativeInstructionSelection(executions);
const campaignId = cleanId(
process.env.PAPERCLIP_E2E_CAMPAIGN_ID ??
`local-${new Date().toISOString().replace(/[:.]/g, "-")}`,
);
const summaryDir = path.join(resultsRoot, campaignId);
await mkdir(summaryDir, { recursive: true });
if (executions.some(execution => execution.suite.id === "native-completion")) {
process.env[NATIVE_COMPLETION_PREFLIGHT_ENV] = prepareNativeCompletionPreflight(summaryDir);
}
if (executions.some(execution => execution.suite.id === NATIVE_INSTRUCTION_SUITE)) {
process.env[NATIVE_INSTRUCTION_PREFLIGHT_ENV] = prepareNativeInstructionPreflight(summaryDir);
}
if (executions.some(execution => execution.suite.id === "stock-harness")) {
process.env[STOCK_PREFLIGHT_ENV] = prepareStockHarnessPreflight(summaryDir);
}
await loadLocalEnvironment(process.env);
assertRemoteNativeEvidencePrerequisites(executions, process.env);
const missingCredentials = process.env.PAPERCLIP_E2E_OPERATIONS_MANIFEST ? [] : [
...new Set(
executions.flatMap((execution) => execution.requiredCredentials),
),
].filter((name) => !process.env[name]?.trim());
if (missingCredentials.length > 0) {
throw new Error(
`Missing runner E2E credentials: ${missingCredentials.join(", ")}`,
);
}
if (
executions.some((execution) => execution.environment.id === "daytona") &&
!isImmutableDaytonaImage(process.env.PAPERCLIP_E2E_DAYTONA_IMAGE)
) {
throw new Error(
"PAPERCLIP_E2E_DAYTONA_IMAGE must be an immutable image@sha256 digest for Daytona cells",
);
}
await writeFile(
path.join(summaryDir, "invocation-policy.json"),
`${JSON.stringify(
{
version: 1,
maxAutomaticRetries: options.maxAutomaticRetries,
retryClasses: ["transient_infrastructure", "provider_variance"],
executions: executions.map(execution => ({
executionId: execution.id,
automaticRetryPolicy: execution.task.automaticRetryPolicy ?? "default",
maxAutomaticRetries: effectiveAutomaticRetryLimit(execution.task, options.maxAutomaticRetries),
})),
},
null,
2,
)}\n`,
"utf8",
);
const requestedParallelism =
options.headed || options.ui || options.debug ? 1 : options.maxParallel;
console.log(
`Running ${executions.length} isolated execution(s) with max parallelism ${requestedParallelism}`,
);
const attempts: RunnerE2EResult[] = [];
const operations = process.env.PAPERCLIP_E2E_OPERATIONS_MANIFEST
? await openEvalOperations(process.env.PAPERCLIP_E2E_OPERATIONS_MANIFEST, executions, options.maxAutomaticRetries, process.env) : undefined;
let finalResults: RunnerE2EResult[] = [];
let campaignFailed = false;
try {
if (operations) await writeFile(path.join(summaryDir, "operations-preflight.json"), `${JSON.stringify(operations.manifest, null, 2)}\n`, { mode: 0o600 });
finalResults = await runWithConcurrency(executions, operations ? Math.min(requestedParallelism, operations.manifest.limits.maxConcurrent) : requestedParallelism,
execution => runExecutionWithRetry({ execution, campaignId, options, operations, onAttempt: result => attempts.push(result) }));
} catch {
campaignFailed = true;
finalResults = [...new Map(attempts.map(result => [result.executionId, result])).values()];
} finally {
try { await operations?.close(); } catch { campaignFailed = true; }
}
const generatedAt = new Date().toISOString();
const operationsReport = operations ? await readEvalOperationsReport(operations.manifest.ledgerPath) : undefined;
if (operationsReport?.status === "unavailable") campaignFailed = true;
const campaign = buildRunnerCampaign({
campaignId,
generatedAt,
expected: executions.map((execution) => execution.id),
results: finalResults,
attempts,
...(operationsReport ? { operations: operationsReport } : {}),
});
const campaignSecrets = normalizedSecrets(
[...CREDENTIAL_NAMES, ...EVAL_OBSERVABILITY_SECRET_NAMES].map((name) => process.env[name]),
);
const safeCampaign = sanitizeJson(
campaign,
campaignSecrets,
) as typeof campaign;
const campaignText = `${JSON.stringify(safeCampaign, null, 2)}\n`;
assertSecretFree(campaignText, campaignSecrets, "campaign.json");
await writeFile(path.join(summaryDir, "campaign.json"), campaignText, "utf8");
const dashboard = renderRunnerE2EDashboard({
title: `Runner E2E · ${campaignId}`,
generatedAt,
expected: executions.map((execution) => execution.id),
catalog: runnerMatrix,
campaign: safeCampaign,
entries: safeCampaign.results.map((result) => ({
result,
valid: result.status === "passed" && result.cleanup === "passed",
errors:
result.status === "passed" && result.cleanup === "passed"
? []
: [
result.error ??
result.failureClass ??
`cleanup=${result.cleanup}`,
],
evidenceBaseHref: [
result.suiteId ?? "core-compatibility",
result.profileId,
result.environmentId,
result.caseId,
`attempt-${result.attempt}`,
].join("/"),
})),
});
assertSecretFree(dashboard, campaignSecrets, "dashboard.html");
await writeFile(path.join(summaryDir, "dashboard.html"), dashboard, "utf8");
console.log(
`Campaign ${campaignId}: ${safeCampaign.passed}/${safeCampaign.selected} passed`,
);
if (campaignFailed) console.error("Campaign finalization incomplete; inspect retained attempt artifacts and the operations ledger.");
if (campaignFailed || safeCampaign.failed > 0) process.exitCode = 1;
}
await main().catch((error) => {
console.error(error instanceof Error ? error.message : String(error));
process.exitCode = 1;
}).finally(async () => {
if (manualPrivateHandoff) await sealManualResultEvidence(manualPrivateHandoff, [...manualEvidenceSecrets]);
}).catch(() => { process.stderr.write("Private evidence finalization failed; no public fallback\n"); process.exitCode = 1; });