agentcert 0.1.1 → 0.2.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -1,10 +1,10 @@
1
1
  # AgentCert CLI
2
2
 
3
- Unified evidence, corpus, monitor, and lab CLI for AgentCert.
3
+ Unified release assurance, evidence, corpus, monitor, and lab CLI for AgentCert.
4
4
 
5
- AgentCert is regression CI for browser agents. It runs Tripwire robustness
6
- checks, converts the result into evidence bundles, writes HTML reports and
7
- badges, and accumulates a local failure corpus.
5
+ AgentCert checks what an agent may do, whether it passed pre-release evidence,
6
+ whether a high-risk runtime action may proceed, and who can verify the observed
7
+ outcome. It writes portable reports and accumulates a local failure corpus.
8
8
 
9
9
  ## 5-minute local path
10
10
 
@@ -12,6 +12,9 @@ badges, and accumulates a local failure corpus.
12
12
  npx agentcert init --subject my-browser-agent
13
13
  ```
14
14
 
15
+ This writes `agentcert.config.json` and `tripwire.yml`. Add `--github-action`
16
+ when you also want `.github/workflows/agentcert-tripwire.yml`.
17
+
15
18
  Edit `tripwire.yml` so `startUrl` and `agent.command` point at your app and
16
19
  browser agent. After Tripwire has produced `.tripwire/latest/tripwire-result.json`,
17
20
  build the AgentCert outputs:
@@ -37,10 +40,25 @@ Review/export helpers:
37
40
  npx agentcert corpus metrics --corpus .agentcert/corpus/corpus.jsonl
38
41
  npx agentcert corpus export-reviewed --corpus .agentcert/corpus/corpus.jsonl --out .agentcert/latest/reviewed-failure-dataset.jsonl
39
42
  npx agentcert corpus classifier-eval --corpus .agentcert/corpus/corpus.jsonl --out .agentcert/latest/failure-classifier-evaluation.json
43
+ npx agentcert validate .agentcert/latest/agentcert-evidence.json
44
+ npx agentcert validate .agentcert/latest/agentcert-evidence.json --check-artifacts
45
+ npx agentcert release-gate --config agentcert.config.json --strict
46
+ npx agentcert release-gate --evidence .agentcert/latest/agentcert-evidence.json --baseline .agentcert/baselines/main.json
40
47
  npx agentcert schema validate --schema evidence-bundle --file .agentcert/latest/agentcert-evidence.json
41
48
  npx agentcert schema validate --schema classifier-eval --file examples/agentcert/classifier-eval.example.json
42
49
  ```
43
50
 
51
+ The release gate writes fixed JSON, HTML, Markdown, JUnit, and badge outputs,
52
+ records SHA-256 artifact provenance, and supports optional Ed25519 signatures:
53
+
54
+ ```bash
55
+ npx agentcert evidence keygen --private-key .agentcert/keys/evidence-private.pem --public-key .agentcert/keys/evidence-public.pem
56
+ npx agentcert evidence sign .agentcert/latest/agentcert-evidence.json --private-key .agentcert/keys/evidence-private.pem
57
+ ```
58
+
59
+ Control semantics and attestation format:
60
+ [release gate checklist](https://github.com/Kakarottoooo/agentcert/blob/main/docs/release-gate-checklist.md).
61
+
44
62
  CI users can run Tripwire and AgentCert together with
45
63
  `Kakarottoooo/agentcert/actions/tripwire@v0`.
46
64
 
@@ -49,6 +67,18 @@ Playwright-based agents over the same fault suite:
49
67
 
50
68
  https://kakarottoooo.github.io/agentcert/public-demo/real-agent-robustness/
51
69
 
70
+ Minimal no-key browser-agent example in the GitHub repo:
71
+
72
+ ```text
73
+ examples/minimal-browser-agent/
74
+ ```
75
+
76
+ External integration smoke matrix in the GitHub repo:
77
+
78
+ ```text
79
+ examples/real-agents/external-integration-smokes.md
80
+ ```
81
+
52
82
  Corpus storage:
53
83
 
54
84
  ```bash
@@ -0,0 +1,92 @@
1
+ import { access } from "node:fs/promises";
2
+ import { dirname, isAbsolute, resolve } from "node:path";
3
+ export async function validateEvidenceArtifacts(input, artifactRoot) {
4
+ if (!input || typeof input !== "object" || Array.isArray(input)) {
5
+ return { checked: 0, missing: [] };
6
+ }
7
+ const paths = collectArtifactPaths(input, artifactRoot);
8
+ const uniquePaths = dedupeArtifactPathEntries(paths).filter((entry) => !looksLikeUrl(entry.path));
9
+ const missing = [];
10
+ for (const entry of uniquePaths) {
11
+ const fullPath = isAbsolute(entry.path) ? entry.path : resolve(entry.root, entry.path);
12
+ try {
13
+ await access(fullPath);
14
+ }
15
+ catch {
16
+ missing.push(entry.path);
17
+ }
18
+ }
19
+ return { checked: uniquePaths.length, missing };
20
+ }
21
+ function collectArtifactPaths(bundle, artifactRoot) {
22
+ const paths = [];
23
+ const productRoots = productArtifactRoots(bundle, artifactRoot);
24
+ collectStringValues(bundle.artifacts, paths, artifactRoot);
25
+ const results = Array.isArray(bundle.results) ? bundle.results : [];
26
+ for (const result of results) {
27
+ if (result && typeof result === "object" && !Array.isArray(result)) {
28
+ const resultRecord = result;
29
+ collectStringValues(resultRecord.artifacts, paths, artifactRoot);
30
+ }
31
+ }
32
+ const evidence = Array.isArray(bundle.evidence) ? bundle.evidence : [];
33
+ for (const item of evidence) {
34
+ if (item && typeof item === "object" && !Array.isArray(item)) {
35
+ const itemRecord = item;
36
+ const artifactPath = itemRecord.artifactPath;
37
+ if (typeof artifactPath === "string" && artifactPath.length > 0) {
38
+ const source = typeof itemRecord.source === "string" ? itemRecord.source : undefined;
39
+ paths.push({ path: artifactPath, root: source ? (productRoots.get(source) ?? artifactRoot) : artifactRoot });
40
+ }
41
+ }
42
+ }
43
+ return paths;
44
+ }
45
+ function productArtifactRoots(bundle, artifactRoot) {
46
+ const roots = new Map();
47
+ const results = Array.isArray(bundle.results) ? bundle.results : [];
48
+ for (const result of results) {
49
+ if (!result || typeof result !== "object" || Array.isArray(result))
50
+ continue;
51
+ const resultRecord = result;
52
+ const product = resultRecord.product;
53
+ const artifacts = resultRecord.artifacts;
54
+ if (typeof product !== "string" || !artifacts || typeof artifacts !== "object" || Array.isArray(artifacts))
55
+ continue;
56
+ const artifactRecord = artifacts;
57
+ const outDir = artifactRecord.outDir;
58
+ const resultPath = artifactRecord.result ?? artifactRecord.results;
59
+ if (typeof outDir === "string" && outDir.length > 0) {
60
+ roots.set(product, isAbsolute(outDir) ? outDir : resolve(artifactRoot, outDir));
61
+ }
62
+ else if (typeof resultPath === "string" && resultPath.length > 0) {
63
+ roots.set(product, dirname(isAbsolute(resultPath) ? resultPath : resolve(artifactRoot, resultPath)));
64
+ }
65
+ }
66
+ return roots;
67
+ }
68
+ function collectStringValues(input, paths, root) {
69
+ if (!input || typeof input !== "object" || Array.isArray(input)) {
70
+ return;
71
+ }
72
+ for (const value of Object.values(input)) {
73
+ if (typeof value === "string" && value.length > 0) {
74
+ paths.push({ path: value, root });
75
+ }
76
+ }
77
+ }
78
+ function dedupeArtifactPathEntries(entries) {
79
+ const seen = new Set();
80
+ const deduped = [];
81
+ for (const entry of entries) {
82
+ const key = `${entry.root}\0${entry.path}`;
83
+ if (seen.has(key))
84
+ continue;
85
+ seen.add(key);
86
+ deduped.push(entry);
87
+ }
88
+ return deduped;
89
+ }
90
+ function looksLikeUrl(value) {
91
+ return /^[a-z][a-z0-9+.-]*:\/\//i.test(value);
92
+ }
package/dist/bundle.js CHANGED
@@ -1,3 +1,4 @@
1
+ import { AGENTCERT_EVIDENCE_SCHEMA_SEMVER, AGENTCERT_EVIDENCE_SCHEMA_VERSION } from "./types.js";
1
2
  export function buildEvidenceBundle(results, subjectName, subjectType = "agent") {
2
3
  const evidence = results.flatMap((result) => result.evidence);
3
4
  const score = results.length === 0 ? 0 : Math.round(results.reduce((sum, result) => sum + result.score, 0) / results.length);
@@ -12,8 +13,8 @@ export function buildEvidenceBundle(results, subjectName, subjectType = "agent")
12
13
  }
13
14
  return {
14
15
  schemaName: "agentcert.evidence_bundle",
15
- schemaVersion: "1",
16
- schemaSemver: "1.0.0",
16
+ schemaVersion: AGENTCERT_EVIDENCE_SCHEMA_VERSION,
17
+ schemaSemver: AGENTCERT_EVIDENCE_SCHEMA_SEMVER,
17
18
  kind: "agentcert.evidence_bundle",
18
19
  runId: `agentcert_${Date.now()}`,
19
20
  generatedAt: new Date().toISOString(),
package/dist/cli.js CHANGED
@@ -1,6 +1,7 @@
1
1
  #!/usr/bin/env node
2
2
  import { mkdir, readFile, writeFile } from "node:fs/promises";
3
- import { dirname, resolve } from "node:path";
3
+ import { dirname, join, resolve } from "node:path";
4
+ import { validateEvidenceArtifacts } from "./artifact-validation.js";
4
5
  import { renderAgentCertBadge } from "./badge.js";
5
6
  import { buildEvidenceBundle } from "./bundle.js";
6
7
  import { recordsFromAgentCertResult, evaluateFailureClassifier, renderCorpusSummary, summarizeCorpus, writeReviewedFailureDataset, } from "./corpus.js";
@@ -9,10 +10,14 @@ import { applyFailureReviews, appendFailureReview, createFailureReview, findFail
9
10
  import { buildMonitorSnapshot, writeMonitorSnapshot } from "./monitor.js";
10
11
  import { normalizeMcpBenchResult, normalizeOnegentAuditPacket, normalizeTripwireResult } from "./normalizers.js";
11
12
  import { renderHtmlReport, renderMarkdownReport } from "./report.js";
13
+ import { generateEvidenceSigningKeyPair, signEvidenceFile, verifyEvidenceFile, } from "./evidence-signing.js";
14
+ import { buildReleaseGateReport, renderReleaseGateSummary, writeReleaseGateArtifacts, } from "./release-gate.js";
12
15
  import { serveAgentCertMonitor } from "./local-server.js";
13
16
  import { parseSchemaId, validateAgentCertSchema } from "./schema-validator.js";
14
- import { loadRunProfile, profileFromArtifactFlags, renderRunSummary, runAgentCertProfile, } from "./runner.js";
17
+ import { applyRunOverrides, loadRunProfile, profileFromArtifactFlags, renderRunSummary, runAgentCertProfile, } from "./runner.js";
15
18
  import { buildRobustnessLabSnapshot, readRobustnessLabConfig, renderRobustnessLabSummary, writeRobustnessLabSnapshot } from "./lab.js";
19
+ process.on("uncaughtException", reportFatalError);
20
+ process.on("unhandledRejection", reportFatalError);
16
21
  const command = process.argv[2] ?? "help";
17
22
  if (command === "init") {
18
23
  const outPath = resolve(readFlag("--out") ?? "agentcert.config.json");
@@ -20,6 +25,7 @@ if (command === "init") {
20
25
  const githubWorkflowPath = resolve(readFlag("--github-action-out") ?? ".github/workflows/agentcert-tripwire.yml");
21
26
  const subject = readFlag("--subject") ?? "my-browser-agent";
22
27
  const force = readBoolFlag("--force");
28
+ const writeGitHubAction = (readBoolFlag("--github-action") || process.argv.includes("--github-action-out")) && !readBoolFlag("--skip-github-action");
23
29
  const config = {
24
30
  schemaVersion: "1",
25
31
  subject: {
@@ -48,6 +54,9 @@ if (command === "init") {
48
54
  },
49
55
  gate: {
50
56
  failOnVerdict: true,
57
+ strict: false,
58
+ outDir: ".agentcert/latest",
59
+ maxScoreDrop: 0,
51
60
  },
52
61
  manifest: {
53
62
  out: ".agentcert/latest/agentcert-run-manifest.json",
@@ -60,15 +69,15 @@ if (command === "init") {
60
69
  await writeStarterFile(tripwireConfigPath, starterTripwireConfig(subject), force);
61
70
  process.stdout.write(`Wrote ${tripwireConfigPath}\n`);
62
71
  }
63
- if (!readBoolFlag("--skip-github-action")) {
72
+ if (writeGitHubAction) {
64
73
  await writeStarterFile(githubWorkflowPath, starterGitHubActionWorkflow(subject), force);
65
74
  process.stdout.write(`Wrote ${githubWorkflowPath}\n`);
66
75
  }
67
76
  process.stdout.write(`
68
77
  Next:
69
78
  1. Edit tripwire.yml so startUrl and agent.command/agent.args match your app and browser agent.
70
- 2. Commit .github/workflows/agentcert-tripwire.yml to run AgentCert in CI.
71
- 3. Or run locally after Tripwire writes .tripwire/latest/tripwire-result.json:
79
+ 2. Run in CI with Kakarottoooo/agentcert/actions/tripwire@v0, or re-run init with --github-action to write a workflow template.
80
+ 3. Run locally after Tripwire writes .tripwire/latest/tripwire-result.json:
72
81
  npx agentcert run --tripwire .tripwire/latest/tripwire-result.json --subject ${JSON.stringify(subject)} --fail-on-verdict
73
82
  `);
74
83
  }
@@ -262,6 +271,106 @@ else if (command === "run") {
262
271
  process.stdout.write(renderRunSummary(outcome));
263
272
  process.exitCode = outcome.exitCode;
264
273
  }
274
+ else if (command === "release-gate") {
275
+ const overrides = readRunOverrides();
276
+ const profileName = readFlag("--profile");
277
+ const configPath = readFlag("--config");
278
+ const evidenceInput = readFlag("--evidence");
279
+ let profile;
280
+ let bundle;
281
+ let evidenceBundlePath;
282
+ let sourceArtifacts;
283
+ if (evidenceInput) {
284
+ profile = configPath ? applyRunOverrides(await loadRunProfile(profileName, configPath), overrides) : undefined;
285
+ bundle = (await readJson(evidenceInput));
286
+ evidenceBundlePath = resolve(evidenceInput);
287
+ sourceArtifacts = bundle.artifacts;
288
+ }
289
+ else {
290
+ const hasArtifactFlags = Boolean(overrides.mcpbench || overrides.tripwire || overrides.onegent);
291
+ const loadedProfile = hasArtifactFlags && !profileName && !configPath
292
+ ? profileFromArtifactFlags(overrides)
293
+ : await loadRunProfile(profileName, configPath);
294
+ profile = applyRunOverrides(loadedProfile, overrides);
295
+ const outcome = await runAgentCertProfile(profile, {
296
+ runCommands: !readBoolFlag("--skip-commands"),
297
+ });
298
+ bundle = outcome.bundle;
299
+ evidenceBundlePath = resolve(outcome.manifest.outputs.evidenceBundle ?? join(profile.outputDir, "agentcert-evidence.json"));
300
+ sourceArtifacts = profile.artifacts;
301
+ if (!outcome.manifest.outputs.evidenceBundle) {
302
+ await mkdir(dirname(evidenceBundlePath), { recursive: true });
303
+ await writeFile(evidenceBundlePath, `${JSON.stringify(outcome.bundle, null, 2)}\n`);
304
+ }
305
+ }
306
+ const outDir = resolve(readFlag("--out") ?? profile?.run?.gate?.outDir ?? profile?.outputDir ?? dirname(evidenceBundlePath));
307
+ await mkdir(outDir, { recursive: true });
308
+ const baselinePath = readFlag("--baseline") ?? profile?.run?.gate?.baseline;
309
+ const baseline = baselinePath ? (await readJson(baselinePath)) : undefined;
310
+ const report = await buildReleaseGateReport(bundle, {
311
+ strict: readBoolFlag("--strict") || profile?.run?.gate?.strict,
312
+ requireBaseline: readBoolFlag("--require-baseline") || profile?.run?.gate?.requireBaseline,
313
+ maxScoreDrop: parseOptionalNonNegativeNumber(readFlag("--max-score-drop"), "--max-score-drop") ?? profile?.run?.gate?.maxScoreDrop,
314
+ baseline,
315
+ attestations: profile?.run?.gate?.controls,
316
+ sourceArtifacts,
317
+ evidenceBundlePath,
318
+ });
319
+ const paths = await writeReleaseGateArtifacts(outDir, report);
320
+ const saveBaselinePath = readFlag("--save-baseline");
321
+ if (saveBaselinePath) {
322
+ if (report.verdict.passed) {
323
+ await mkdir(dirname(resolve(saveBaselinePath)), { recursive: true });
324
+ await writeFile(resolve(saveBaselinePath), `${JSON.stringify(bundle, null, 2)}\n`);
325
+ process.stdout.write(`Saved passing baseline ${resolve(saveBaselinePath)}\n`);
326
+ }
327
+ else {
328
+ process.stdout.write(`Baseline not saved because the release gate failed: ${resolve(saveBaselinePath)}\n`);
329
+ }
330
+ }
331
+ const privateKey = readFlag("--sign-private-key") ?? profile?.run?.gate?.signing?.privateKey;
332
+ if (privateKey) {
333
+ await signEvidenceFile(evidenceBundlePath, privateKey);
334
+ await signEvidenceFile(paths.json, privateKey);
335
+ process.stdout.write(`Signed ${evidenceBundlePath}\nSigned ${paths.json}\n`);
336
+ }
337
+ process.stdout.write(renderReleaseGateSummary(report));
338
+ process.stdout.write(`Release gate JSON: ${paths.json}\nRelease gate HTML: ${paths.html}\nRelease gate JUnit: ${paths.junit}\n`);
339
+ process.exitCode = report.verdict.passed ? 0 : 1;
340
+ }
341
+ else if (command === "evidence") {
342
+ const action = process.argv[3] ?? "help";
343
+ if (action === "keygen") {
344
+ const privateKeyPath = readFlag("--private-key") ?? ".agentcert/keys/evidence-private.pem";
345
+ const publicKeyPath = readFlag("--public-key") ?? ".agentcert/keys/evidence-public.pem";
346
+ await generateEvidenceSigningKeyPair(privateKeyPath, publicKeyPath);
347
+ process.stdout.write(`Wrote private key ${resolve(privateKeyPath)}\nWrote public key ${resolve(publicKeyPath)}\n`);
348
+ }
349
+ else if (action === "sign") {
350
+ const file = readFlag("--file") ?? readPositional(4);
351
+ if (!file)
352
+ throw new Error("Missing evidence file. Usage: agentcert evidence sign <file> --private-key <path>.");
353
+ const privateKeyPath = requiredFlag("--private-key");
354
+ const signaturePath = readFlag("--out") ?? `${file}.sig.json`;
355
+ const signature = await signEvidenceFile(file, privateKeyPath, signaturePath);
356
+ process.stdout.write(`Signed ${resolve(file)}\nSignature: ${resolve(signaturePath)}\nKey id: ${signature.keyId}\n`);
357
+ }
358
+ else if (action === "verify") {
359
+ const file = readFlag("--file") ?? readPositional(4);
360
+ if (!file)
361
+ throw new Error("Missing evidence file. Usage: agentcert evidence verify <file> --signature <path> --public-key <path>.");
362
+ const result = await verifyEvidenceFile(file, requiredFlag("--signature"), requiredFlag("--public-key"));
363
+ process.stdout.write(`${result.valid ? "Valid" : "Invalid"} evidence signature: ${resolve(file)}\nKey id: ${result.keyId}\nSHA-256: ${result.artifactSha256}\n`);
364
+ process.exitCode = result.valid ? 0 : 1;
365
+ }
366
+ else {
367
+ process.stdout.write(`Usage:
368
+ agentcert evidence keygen --private-key .agentcert/keys/evidence-private.pem --public-key .agentcert/keys/evidence-public.pem
369
+ agentcert evidence sign .agentcert/latest/agentcert-evidence.json --private-key .agentcert/keys/evidence-private.pem
370
+ agentcert evidence verify .agentcert/latest/agentcert-evidence.json --signature .agentcert/latest/agentcert-evidence.json.sig.json --public-key .agentcert/keys/evidence-public.pem
371
+ `);
372
+ }
373
+ }
265
374
  else if (command === "lab") {
266
375
  const action = process.argv[3] ?? "help";
267
376
  if (action === "build") {
@@ -314,13 +423,46 @@ else if (command === "schema") {
314
423
  agentcert schema validate --schema monitor-snapshot --file .agentcert/latest/monitor.json
315
424
  agentcert schema validate --schema corpus-record --file examples/agentcert/corpus-record.example.json
316
425
  agentcert schema validate --schema classifier-eval --file examples/agentcert/classifier-eval.example.json
426
+ agentcert schema validate --schema release-gate --file .agentcert/latest/agentcert-release-gate.json
427
+ agentcert schema validate --schema evidence-signature --file .agentcert/latest/agentcert-evidence.json.sig.json
317
428
  `);
318
429
  }
319
430
  }
431
+ else if (command === "validate") {
432
+ const schema = parseSchemaId(readFlag("--schema") ?? "evidence-bundle");
433
+ const file = readFlag("--file") ?? readFirstPositionalAfterCommand();
434
+ if (!file) {
435
+ throw new Error("Missing evidence file. Usage: agentcert validate <file> [--schema evidence-bundle].");
436
+ }
437
+ const input = await readJson(file);
438
+ const result = validateAgentCertSchema(schema, input);
439
+ if (result.valid) {
440
+ process.stdout.write(`Valid ${schema}: ${resolve(file)}\n`);
441
+ }
442
+ else {
443
+ process.stdout.write(`Invalid ${schema}: ${resolve(file)}\n`);
444
+ for (const error of result.errors) {
445
+ process.stdout.write(`- ${error}\n`);
446
+ }
447
+ process.exitCode = 1;
448
+ }
449
+ if (result.valid && schema === "evidence-bundle" && readBoolFlag("--check-artifacts")) {
450
+ const artifactResult = await validateEvidenceArtifacts(input, resolve(readFlag("--artifact-root") ?? process.cwd()));
451
+ process.stdout.write(`Artifact paths checked: ${artifactResult.checked}\n`);
452
+ if (artifactResult.missing.length > 0) {
453
+ process.stdout.write("Missing artifact paths:\n");
454
+ for (const missing of artifactResult.missing) {
455
+ process.stdout.write(`- ${missing}\n`);
456
+ }
457
+ process.exitCode = 1;
458
+ }
459
+ }
460
+ }
320
461
  else {
321
462
  process.stdout.write(`Usage:
322
463
  agentcert init --subject my-browser-agent
323
464
  agentcert init --out agentcert.config.json --tripwire-config tripwire.yml --force
465
+ agentcert init --subject my-browser-agent --github-action
324
466
  agentcert report --mcpbench .mcpbench/latest/results.json --tripwire .tripwire/latest/tripwire-result.json --onegent .onegent/procurement/audit-packet.json --out .agentcert/latest --subject my-agent
325
467
  agentcert corpus ingest --tripwire .tripwire/latest/tripwire-result.json --out .agentcert/corpus/corpus.jsonl --subject my-agent
326
468
  agentcert corpus review --corpus .agentcert/corpus/corpus.jsonl --reviews .agentcert/corpus/failure-reviews.jsonl --pattern-key <failure-key> --type wrong_click --status corrected
@@ -330,8 +472,17 @@ else {
330
472
  agentcert monitor build --store sqlite --sqlite .agentcert/corpus/agentcert.sqlite --out .agentcert/latest/monitor.json --subject my-agent
331
473
  agentcert run --profile public-demo
332
474
  agentcert run --mcpbench .mcpbench/latest/results.json --tripwire .tripwire/latest/tripwire-result.json --onegent .onegent/procurement/audit-packet.json --out .agentcert/latest --corpus .agentcert/corpus/corpus.jsonl --monitor-out .agentcert/latest/monitor.json --reviewed-dataset-out .agentcert/latest/reviewed-failure-dataset.jsonl
475
+ agentcert release-gate --config agentcert.config.json --strict
476
+ agentcert release-gate --evidence .agentcert/latest/agentcert-evidence.json --baseline .agentcert/baselines/main.json
477
+ agentcert release-gate --evidence .agentcert/latest/agentcert-evidence.json --save-baseline .agentcert/baselines/main.json
478
+ agentcert release-gate --tripwire .tripwire/latest/tripwire-result.json --baseline .agentcert/baselines/main.json
479
+ agentcert evidence keygen --private-key .agentcert/keys/evidence-private.pem --public-key .agentcert/keys/evidence-public.pem
480
+ agentcert evidence sign .agentcert/latest/agentcert-evidence.json --private-key .agentcert/keys/evidence-private.pem
481
+ agentcert evidence verify .agentcert/latest/agentcert-evidence.json --signature .agentcert/latest/agentcert-evidence.json.sig.json --public-key .agentcert/keys/evidence-public.pem
333
482
  agentcert lab build --config examples/real-agents/robustness-lab/lab.config.json --out public-demo/real-agent-robustness/evidence/lab-snapshot.json
334
483
  agentcert serve --corpus .agentcert/corpus/corpus.jsonl --static public-demo/agentcert-monitor --artifact-root public-demo/browser-agent-robustness/evidence/tripwire-public-demo
484
+ agentcert validate .agentcert/latest/agentcert-evidence.json
485
+ agentcert validate .agentcert/latest/agentcert-evidence.json --check-artifacts
335
486
  agentcert schema validate --schema evidence-bundle --file .agentcert/latest/agentcert-evidence.json
336
487
  `);
337
488
  }
@@ -368,6 +519,32 @@ function readFlag(name) {
368
519
  const index = process.argv.indexOf(name);
369
520
  return index >= 0 ? process.argv[index + 1] : undefined;
370
521
  }
522
+ function readFirstPositionalAfterCommand() {
523
+ for (let index = 3; index < process.argv.length; index += 1) {
524
+ const value = process.argv[index];
525
+ if (!value)
526
+ continue;
527
+ if (value.startsWith("--")) {
528
+ index += 1;
529
+ continue;
530
+ }
531
+ return value;
532
+ }
533
+ return undefined;
534
+ }
535
+ function readPositional(startIndex) {
536
+ for (let index = startIndex; index < process.argv.length; index += 1) {
537
+ const value = process.argv[index];
538
+ if (!value)
539
+ continue;
540
+ if (value.startsWith("--")) {
541
+ index += 1;
542
+ continue;
543
+ }
544
+ return value;
545
+ }
546
+ return undefined;
547
+ }
371
548
  function readBoolFlag(name) {
372
549
  return process.argv.includes(name);
373
550
  }
@@ -443,6 +620,15 @@ function parseOptionalNonNegativeInteger(input, flagName) {
443
620
  }
444
621
  return value;
445
622
  }
623
+ function parseOptionalNonNegativeNumber(input, flagName) {
624
+ if (input === undefined)
625
+ return undefined;
626
+ const value = Number(input);
627
+ if (!Number.isFinite(value) || value < 0) {
628
+ throw new Error(`${flagName} must be a non-negative number.`);
629
+ }
630
+ return value;
631
+ }
446
632
  function readReviewsPath() {
447
633
  return readFlag("--reviews") ?? process.env.AGENTCERT_FAILURE_REVIEWS;
448
634
  }
@@ -467,6 +653,38 @@ async function writeStarterFile(path, content, force) {
467
653
  throw error;
468
654
  }
469
655
  }
656
+ function reportFatalError(error) {
657
+ process.stderr.write(formatCliError(error));
658
+ process.exit(1);
659
+ }
660
+ function formatCliError(error) {
661
+ const message = error instanceof Error ? error.message : String(error);
662
+ const code = error && typeof error === "object" && "code" in error ? String(error.code) : undefined;
663
+ const detail = [`AgentCert error: ${message}`];
664
+ const hint = errorHint(message, code);
665
+ if (hint) {
666
+ detail.push("", hint);
667
+ }
668
+ return `${detail.join("\n")}\n`;
669
+ }
670
+ function errorHint(message, code) {
671
+ if (code === "ENOENT") {
672
+ return "Hint: check the artifact path. Run `npx agentcert init` for starter config, then run Tripwire before `npx agentcert run`.";
673
+ }
674
+ if (message.includes("agentcert run requires")) {
675
+ return "Hint: run `npx agentcert init`, pass `--tripwire .tripwire/latest/tripwire-result.json`, or use `--config agentcert.config.json`.";
676
+ }
677
+ if (message.includes("No AgentCert artifacts were loaded") || message.includes("No input artifacts were provided")) {
678
+ return "Hint: provide at least one artifact with `--tripwire`, `--mcpbench`, `--onegent`, or configure `agentcert.config.json`.";
679
+ }
680
+ if (message.includes("Unexpected token") || message.includes("JSON")) {
681
+ return "Hint: AgentCert expected JSON. Check that the artifact file exists and was not replaced by logs or HTML.";
682
+ }
683
+ if (message.includes("already exists")) {
684
+ return "Hint: re-run with `--force` to overwrite starter files, or choose a different `--out` path.";
685
+ }
686
+ return undefined;
687
+ }
470
688
  function starterTripwireConfig(subject) {
471
689
  return `version: "0.1"
472
690
  project: ${JSON.stringify(subject)}
@@ -554,6 +772,8 @@ jobs:
554
772
  subject: ${JSON.stringify(subject)}
555
773
  agentcert-out: .agentcert/latest
556
774
  fail-on-verdict: "true"
775
+ release-gate: "true"
776
+ strict-release-gate: "false"
557
777
  # publish-pages: "true"
558
778
  `;
559
779
  }
@@ -0,0 +1,85 @@
1
+ import { createHash, createPrivateKey, createPublicKey, generateKeyPairSync, sign, verify } from "node:crypto";
2
+ import { mkdir, readFile, writeFile } from "node:fs/promises";
3
+ import { dirname, resolve } from "node:path";
4
+ export const AGENTCERT_SIGNATURE_SCHEMA_VERSION = "agentcert.evidence_signature.v0.1";
5
+ export async function generateEvidenceSigningKeyPair(privateKeyPath, publicKeyPath) {
6
+ const { privateKey, publicKey } = generateKeyPairSync("ed25519", {
7
+ privateKeyEncoding: { type: "pkcs8", format: "pem" },
8
+ publicKeyEncoding: { type: "spki", format: "pem" },
9
+ });
10
+ await Promise.all([
11
+ writePrivateFile(privateKeyPath, privateKey),
12
+ writePublicFile(publicKeyPath, publicKey),
13
+ ]);
14
+ }
15
+ export async function signEvidenceFile(artifactPath, privateKeyPath, signaturePath = `${artifactPath}.sig.json`) {
16
+ const [artifact, privatePem] = await Promise.all([readFile(resolve(artifactPath)), readFile(resolve(privateKeyPath), "utf8")]);
17
+ const privateKey = createPrivateKey(privatePem);
18
+ const publicKey = createPublicKey(privateKey);
19
+ const envelope = {
20
+ schemaVersion: AGENTCERT_SIGNATURE_SCHEMA_VERSION,
21
+ kind: "agentcert.evidence_signature",
22
+ algorithm: "Ed25519",
23
+ keyId: keyId(publicKey),
24
+ signedAt: new Date().toISOString(),
25
+ artifactPath: artifactPath.replace(/\\/g, "/"),
26
+ artifactSha256: sha256(artifact),
27
+ signature: sign(null, artifact, privateKey).toString("base64"),
28
+ };
29
+ const resolvedSignaturePath = resolve(signaturePath);
30
+ await mkdir(dirname(resolvedSignaturePath), { recursive: true });
31
+ await writeFile(resolvedSignaturePath, `${JSON.stringify(envelope, null, 2)}\n`);
32
+ return envelope;
33
+ }
34
+ export async function verifyEvidenceFile(artifactPath, signaturePath, publicKeyPath) {
35
+ const [artifact, signatureRaw, publicPem] = await Promise.all([
36
+ readFile(resolve(artifactPath)),
37
+ readFile(resolve(signaturePath), "utf8"),
38
+ readFile(resolve(publicKeyPath), "utf8"),
39
+ ]);
40
+ const envelope = JSON.parse(signatureRaw);
41
+ const errors = validateEvidenceSignature(envelope);
42
+ if (errors.length > 0)
43
+ throw new Error(`Invalid evidence signature: ${errors.join(" ")}`);
44
+ const publicKey = createPublicKey(publicPem);
45
+ const artifactSha256 = sha256(artifact);
46
+ const actualKeyId = keyId(publicKey);
47
+ const digestMatches = envelope.artifactSha256 === artifactSha256;
48
+ const keyIdMatches = envelope.keyId === actualKeyId;
49
+ const signatureMatches = verify(null, artifact, publicKey, Buffer.from(envelope.signature, "base64"));
50
+ return { valid: digestMatches && signatureMatches && keyIdMatches, digestMatches, signatureMatches, keyIdMatches, artifactSha256, keyId: actualKeyId };
51
+ }
52
+ export function validateEvidenceSignature(input) {
53
+ if (!input || typeof input !== "object" || Array.isArray(input))
54
+ return ["$ must be an object."];
55
+ const value = input;
56
+ const errors = [];
57
+ if (value.schemaVersion !== AGENTCERT_SIGNATURE_SCHEMA_VERSION)
58
+ errors.push(`schemaVersion must be ${JSON.stringify(AGENTCERT_SIGNATURE_SCHEMA_VERSION)}.`);
59
+ if (value.kind !== "agentcert.evidence_signature")
60
+ errors.push('kind must be "agentcert.evidence_signature".');
61
+ if (value.algorithm !== "Ed25519")
62
+ errors.push('algorithm must be "Ed25519".');
63
+ for (const field of ["keyId", "signedAt", "artifactPath", "artifactSha256", "signature"]) {
64
+ if (typeof value[field] !== "string" || value[field].length === 0)
65
+ errors.push(`${field} must be a non-empty string.`);
66
+ }
67
+ return errors;
68
+ }
69
+ function keyId(publicKey) {
70
+ const der = publicKey.export({ type: "spki", format: "der" });
71
+ return `sha256:${sha256(der).slice(0, 32)}`;
72
+ }
73
+ function sha256(input) {
74
+ return createHash("sha256").update(input).digest("hex");
75
+ }
76
+ async function writePrivateFile(path, content) {
77
+ const resolved = resolve(path);
78
+ await mkdir(dirname(resolved), { recursive: true });
79
+ await writeFile(resolved, content, { mode: 0o600, flag: "wx" });
80
+ }
81
+ async function writePublicFile(path, content) {
82
+ const resolved = resolve(path);
83
+ await mkdir(dirname(resolved), { recursive: true });
84
+ await writeFile(resolved, content, { flag: "wx" });
85
+ }
package/dist/index.js CHANGED
@@ -2,9 +2,11 @@ export * from "./bundle.js";
2
2
  export * from "./corpus.js";
3
3
  export * from "./corpus-store.js";
4
4
  export * from "./failure-review.js";
5
+ export * from "./evidence-signing.js";
5
6
  export * from "./local-server.js";
6
7
  export * from "./monitor.js";
7
8
  export * from "./normalizers.js";
8
9
  export * from "./report.js";
10
+ export * from "./release-gate.js";
9
11
  export * from "./schema-validator.js";
10
12
  export * from "./types.js";
@@ -94,6 +94,8 @@ export function normalizeOnegentAuditPacket(input, artifactPath) {
94
94
  const action = asRecord(value.actionIntent);
95
95
  const risk = asRecord(value.riskAssessment);
96
96
  const approval = asRecord(value.approvalRequest);
97
+ const principal = asRecord(action.principal);
98
+ const authorization = asRecord(value.authorizationDecision);
97
99
  const verification = asRecord(value.verificationResult);
98
100
  const auditEvents = Array.isArray(value.auditEvents) ? value.auditEvents : [];
99
101
  const verificationSuccess = verification.success === true;
@@ -101,6 +103,18 @@ export function normalizeOnegentAuditPacket(input, artifactPath) {
101
103
  const passed = verificationSuccess && approved;
102
104
  const riskLevel = stringValue(risk.riskLevel) ?? "UNKNOWN";
103
105
  const evidence = [
106
+ {
107
+ id: "onegent_principal",
108
+ kind: "runtime_identity",
109
+ severity: "info",
110
+ message: `Runtime action principal: ${stringValue(principal.id) ?? stringValue(action.sourceAgentName) ?? "UNKNOWN"}.`,
111
+ source: "onegent-runtime",
112
+ artifactPath,
113
+ metadata: {
114
+ principal,
115
+ requestedPermissions: action.requestedPermissions,
116
+ },
117
+ },
104
118
  {
105
119
  id: "onegent_risk_assessment",
106
120
  kind: "runtime_risk_assessment",
@@ -111,6 +125,17 @@ export function normalizeOnegentAuditPacket(input, artifactPath) {
111
125
  metadata: risk,
112
126
  },
113
127
  ];
128
+ if (authorization.decision) {
129
+ evidence.push({
130
+ id: "onegent_authorization",
131
+ kind: "authorization_decision",
132
+ severity: authorization.decision === "ALLOW" ? "info" : "critical",
133
+ message: `Runtime authorization decision: ${String(authorization.decision)}.`,
134
+ source: "onegent-runtime",
135
+ artifactPath,
136
+ metadata: authorization,
137
+ });
138
+ }
114
139
  if (approval.status) {
115
140
  evidence.push({
116
141
  id: "onegent_approval",
@@ -0,0 +1,457 @@
1
+ import { createHash } from "node:crypto";
2
+ import { access, mkdir, readFile, stat, writeFile } from "node:fs/promises";
3
+ import { resolve } from "node:path";
4
+ export const AGENTCERT_RELEASE_GATE_SCHEMA_VERSION = "agentcert.release_gate.v0.1";
5
+ export const RELEASE_GATE_CONTROL_IDS = [
6
+ "permission-boundary",
7
+ "data-boundary",
8
+ "tool-contract",
9
+ "state-verification",
10
+ "human-handoff",
11
+ "rate-loop-cost-limits",
12
+ "idempotency-retry-safety",
13
+ "observability-auditability",
14
+ "rollback-kill-switch",
15
+ "supply-chain-dependency-boundary",
16
+ ];
17
+ const CONTROL_DEFINITIONS = [
18
+ {
19
+ id: "permission-boundary",
20
+ name: "Identity, authorization, and least privilege",
21
+ mode: "evidence-required",
22
+ evaluate: evaluatePermissionBoundary,
23
+ },
24
+ {
25
+ id: "data-boundary",
26
+ name: "Sensitive data boundary",
27
+ mode: "evidence-required",
28
+ evaluate: evaluateDataBoundary,
29
+ },
30
+ {
31
+ id: "tool-contract",
32
+ name: "Tool and MCP contract",
33
+ mode: "automated",
34
+ evaluate: (bundle) => evaluateProduct(bundle, "mcpbench", "MCPBench tool contract evidence is not present."),
35
+ },
36
+ {
37
+ id: "state-verification",
38
+ name: "Observed state verification",
39
+ mode: "automated",
40
+ evaluate: evaluateStateVerification,
41
+ },
42
+ {
43
+ id: "human-handoff",
44
+ name: "High-risk action approval and handoff",
45
+ mode: "evidence-required",
46
+ evaluate: evaluateHumanHandoff,
47
+ },
48
+ {
49
+ id: "rate-loop-cost-limits",
50
+ name: "Rate, loop, timeout, and cost limits",
51
+ mode: "evidence-required",
52
+ evaluate: () => unresolved("needs-evidence", "Provide configured step, timeout, retry, and budget limits."),
53
+ },
54
+ {
55
+ id: "idempotency-retry-safety",
56
+ name: "Idempotency and retry safety",
57
+ mode: "manual",
58
+ evaluate: () => unresolved("manual-review", "A named owner must review duplicate execution and partial failure behavior."),
59
+ },
60
+ {
61
+ id: "observability-auditability",
62
+ name: "Observability, incident trace, and accountability",
63
+ mode: "automated",
64
+ evaluate: evaluateObservability,
65
+ },
66
+ {
67
+ id: "rollback-kill-switch",
68
+ name: "Rollback and kill switch",
69
+ mode: "manual",
70
+ evaluate: () => unresolved("manual-review", "A named owner must provide a rollback or kill-switch runbook."),
71
+ },
72
+ {
73
+ id: "supply-chain-dependency-boundary",
74
+ name: "Supply-chain and dependency boundary",
75
+ mode: "manual",
76
+ evaluate: () => unresolved("manual-review", "A named owner must review models, packages, MCP servers, and adapters."),
77
+ },
78
+ ];
79
+ export async function buildReleaseGateReport(bundle, options = {}) {
80
+ const strict = options.strict ?? false;
81
+ const controls = CONTROL_DEFINITIONS.map((definition) => {
82
+ const evaluated = definition.evaluate(bundle);
83
+ return applyAttestation({ id: definition.id, name: definition.name, mode: definition.mode, ...evaluated }, options.attestations?.[definition.id]);
84
+ });
85
+ const regression = compareWithBaseline(bundle, options.baseline, {
86
+ requireBaseline: options.requireBaseline ?? false,
87
+ maxScoreDrop: options.maxScoreDrop ?? 0,
88
+ });
89
+ const cwd = resolve(options.cwd ?? process.cwd());
90
+ const sourceArtifacts = await digestArtifacts(options.sourceArtifacts ?? {}, cwd);
91
+ const evidenceBundle = options.evidenceBundlePath
92
+ ? await digestArtifact("agentcert-evidence", options.evidenceBundlePath, cwd)
93
+ : undefined;
94
+ const blockers = collectBlockers(bundle, controls, regression, strict, options.requireBaseline ?? false, sourceArtifacts, evidenceBundle);
95
+ return {
96
+ schemaVersion: AGENTCERT_RELEASE_GATE_SCHEMA_VERSION,
97
+ kind: "agentcert.release_gate",
98
+ runId: `release_gate_${createHash("sha256").update(`${bundle.runId}:${JSON.stringify(controls)}`).digest("hex").slice(0, 12)}`,
99
+ generatedAt: new Date().toISOString(),
100
+ subject: bundle.subject,
101
+ strict,
102
+ verdict: {
103
+ passed: blockers.length === 0,
104
+ blockers,
105
+ passedControls: controls.filter((control) => control.status === "pass").length,
106
+ failedControls: controls.filter((control) => control.status === "fail").length,
107
+ needsEvidenceControls: controls.filter((control) => control.status === "needs-evidence").length,
108
+ manualReviewControls: controls.filter((control) => control.status === "manual-review").length,
109
+ },
110
+ bundleVerdict: bundle.verdict,
111
+ controls,
112
+ regression,
113
+ provenance: { evidenceBundle, sourceArtifacts },
114
+ };
115
+ }
116
+ export async function writeReleaseGateArtifacts(outDir, report) {
117
+ const resolvedOutDir = resolve(outDir);
118
+ await mkdir(resolvedOutDir, { recursive: true });
119
+ const paths = {
120
+ json: resolve(resolvedOutDir, "agentcert-release-gate.json"),
121
+ markdown: resolve(resolvedOutDir, "agentcert-release-gate.md"),
122
+ html: resolve(resolvedOutDir, "agentcert-release-gate.html"),
123
+ junit: resolve(resolvedOutDir, "agentcert-release-gate-junit.xml"),
124
+ badge: resolve(resolvedOutDir, "release-gate-badge.svg"),
125
+ };
126
+ await Promise.all([
127
+ writeFile(paths.json, `${JSON.stringify(report, null, 2)}\n`),
128
+ writeFile(paths.markdown, renderReleaseGateMarkdown(report)),
129
+ writeFile(paths.html, renderReleaseGateHtml(report)),
130
+ writeFile(paths.junit, renderReleaseGateJunit(report)),
131
+ writeFile(paths.badge, renderReleaseGateBadge(report)),
132
+ ]);
133
+ return paths;
134
+ }
135
+ export function renderReleaseGateSummary(report) {
136
+ const lines = [
137
+ "# AgentCert Release Gate",
138
+ "",
139
+ `Subject: ${report.subject.name}`,
140
+ `Verdict: ${report.verdict.passed ? "PASS" : "FAIL"}`,
141
+ `Controls: ${report.verdict.passedControls}/10 passed, ${report.verdict.failedControls} failed, ${report.verdict.needsEvidenceControls} need evidence, ${report.verdict.manualReviewControls} need manual review`,
142
+ `Regression: ${report.regression.status}`,
143
+ ];
144
+ if (report.verdict.blockers.length > 0) {
145
+ lines.push("", "Blockers:", ...report.verdict.blockers.map((blocker) => `- ${blocker}`));
146
+ }
147
+ return `${lines.join("\n")}\n`;
148
+ }
149
+ export function validateReleaseGateReport(input) {
150
+ const errors = [];
151
+ if (!input || typeof input !== "object" || Array.isArray(input))
152
+ return ["$ must be an object."];
153
+ const value = input;
154
+ if (value.schemaVersion !== AGENTCERT_RELEASE_GATE_SCHEMA_VERSION) {
155
+ errors.push(`schemaVersion must be ${JSON.stringify(AGENTCERT_RELEASE_GATE_SCHEMA_VERSION)}.`);
156
+ }
157
+ if (value.kind !== "agentcert.release_gate")
158
+ errors.push('kind must be "agentcert.release_gate".');
159
+ if (typeof value.runId !== "string" || value.runId.length === 0)
160
+ errors.push("runId must be a non-empty string.");
161
+ if (!Array.isArray(value.controls) || value.controls.length !== RELEASE_GATE_CONTROL_IDS.length) {
162
+ errors.push(`controls must contain ${RELEASE_GATE_CONTROL_IDS.length} entries.`);
163
+ }
164
+ else {
165
+ const seen = new Set();
166
+ for (const [index, controlInput] of value.controls.entries()) {
167
+ const control = record(controlInput);
168
+ if (typeof control.id !== "string" || !RELEASE_GATE_CONTROL_IDS.includes(control.id)) {
169
+ errors.push(`controls[${index}].id is not a supported release control.`);
170
+ }
171
+ else {
172
+ seen.add(control.id);
173
+ }
174
+ if (!["automated", "evidence-required", "manual"].includes(control.mode)) {
175
+ errors.push(`controls[${index}].mode is invalid.`);
176
+ }
177
+ if (!["pass", "fail", "needs-evidence", "manual-review"].includes(control.status)) {
178
+ errors.push(`controls[${index}].status is invalid.`);
179
+ }
180
+ if (!Array.isArray(control.evidence))
181
+ errors.push(`controls[${index}].evidence must be an array.`);
182
+ }
183
+ if (seen.size !== RELEASE_GATE_CONTROL_IDS.length)
184
+ errors.push("controls must contain every release control exactly once.");
185
+ }
186
+ if (!value.verdict || typeof value.verdict !== "object" || Array.isArray(value.verdict)) {
187
+ errors.push("verdict must be an object.");
188
+ }
189
+ else if (typeof record(value.verdict).passed !== "boolean") {
190
+ errors.push("verdict.passed must be a boolean.");
191
+ }
192
+ if (!value.regression || typeof value.regression !== "object" || Array.isArray(value.regression))
193
+ errors.push("regression must be an object.");
194
+ if (!value.provenance || typeof value.provenance !== "object" || Array.isArray(value.provenance))
195
+ errors.push("provenance must be an object.");
196
+ return errors;
197
+ }
198
+ function evaluatePermissionBoundary(bundle) {
199
+ const decision = bundle.evidence.find((item) => item.kind === "authorization_decision");
200
+ if (!decision) {
201
+ return unresolved("needs-evidence", "No runtime authorization decision proves the principal and granted permissions.");
202
+ }
203
+ const metadata = record(decision.metadata);
204
+ const allowed = metadata.decision === "ALLOW";
205
+ return {
206
+ status: allowed ? "pass" : "fail",
207
+ summary: allowed ? "Runtime authorization allowed the recorded principal within granted permissions." : "Runtime authorization denied the action.",
208
+ evidence: evidenceReferences([decision]),
209
+ };
210
+ }
211
+ function evaluateDataBoundary(bundle) {
212
+ const exposures = bundle.evidence.filter((item) => item.kind === "sensitive_data_exposure" ||
213
+ item.kind === "secret_exposure" ||
214
+ /sensitive|secret|credential|pii/i.test(item.kind));
215
+ if (exposures.some((item) => item.severity === "critical" || item.severity === "high")) {
216
+ return { status: "fail", summary: "Sensitive-data evidence contains a blocking finding.", evidence: evidenceReferences(exposures) };
217
+ }
218
+ return unresolved("needs-evidence", "No blocking leak was observed, but an explicit data-boundary test or attestation is required.");
219
+ }
220
+ function evaluateProduct(bundle, product, missingSummary) {
221
+ const result = bundle.results.find((item) => item.product === product);
222
+ if (!result)
223
+ return unresolved("needs-evidence", missingSummary);
224
+ return {
225
+ status: result.passed ? "pass" : "fail",
226
+ summary: result.summary ?? `${product} ${result.passed ? "passed" : "failed"}.`,
227
+ evidence: resultEvidence(result),
228
+ };
229
+ }
230
+ function evaluateStateVerification(bundle) {
231
+ const results = bundle.results.filter((result) => result.product === "tripwire-ci" || result.product === "onegent-runtime");
232
+ if (results.length === 0) {
233
+ return unresolved("needs-evidence", "No Tripwire or Onegent result proves expected state against observed state.");
234
+ }
235
+ const failed = results.filter((result) => !result.passed);
236
+ return {
237
+ status: failed.length === 0 ? "pass" : "fail",
238
+ summary: failed.length === 0 ? "Configured browser/runtime state verification passed." : `${failed.map((item) => item.product).join(", ")} state verification failed.`,
239
+ evidence: results.flatMap(resultEvidence),
240
+ };
241
+ }
242
+ function evaluateHumanHandoff(bundle) {
243
+ const onegent = bundle.results.find((result) => result.product === "onegent-runtime");
244
+ if (!onegent)
245
+ return unresolved("needs-evidence", "No runtime approval or human-handoff record is present.");
246
+ const approval = onegent.evidence.find((item) => item.kind === "approval_record");
247
+ const risk = onegent.evidence.find((item) => item.kind === "runtime_risk_assessment");
248
+ const requiresApproval = record(risk?.metadata).requiresHumanApproval;
249
+ if (!approval && requiresApproval === false) {
250
+ return { status: "pass", summary: "Recorded risk assessment did not require human approval.", evidence: evidenceReferences(risk ? [risk] : []) };
251
+ }
252
+ if (!approval)
253
+ return unresolved("needs-evidence", "The runtime result does not contain an approval decision.");
254
+ const status = record(approval.metadata).status;
255
+ return {
256
+ status: status === "APPROVED" ? "pass" : "fail",
257
+ summary: status === "APPROVED" ? "The high-risk action has a recorded human approval." : `Approval status ${String(status ?? "UNKNOWN")} does not allow execution.`,
258
+ evidence: evidenceReferences([approval]),
259
+ };
260
+ }
261
+ function evaluateObservability(bundle) {
262
+ const missingArtifacts = bundle.results.filter((result) => Object.values(result.artifacts).filter(Boolean).length === 0);
263
+ if (!bundle.runId || !bundle.generatedAt || bundle.results.length === 0 || missingArtifacts.length > 0) {
264
+ return { status: "fail", summary: "The bundle is missing traceable run metadata or product artifacts.", evidence: [] };
265
+ }
266
+ return {
267
+ status: "pass",
268
+ summary: "Run identity, timestamps, normalized results, findings, and artifact pointers are present.",
269
+ evidence: bundle.results.flatMap(resultEvidence),
270
+ };
271
+ }
272
+ function applyAttestation(control, attestation) {
273
+ if (!attestation)
274
+ return control;
275
+ const attestationEvidence = Array.isArray(attestation.evidence) ? attestation.evidence.filter(Boolean) : [];
276
+ const complete = Boolean(attestation.owner && attestation.reviewedAt && attestationEvidence.length > 0);
277
+ if (!complete) {
278
+ return {
279
+ ...control,
280
+ status: control.status === "fail" ? "fail" : control.mode === "manual" ? "manual-review" : "needs-evidence",
281
+ summary: `${control.summary} The configured attestation is incomplete; owner, reviewedAt, and evidence are required.`,
282
+ };
283
+ }
284
+ if (control.status === "fail") {
285
+ return { ...control, evidence: [...control.evidence, ...attestationEvidence], owner: attestation.owner, reviewedAt: attestation.reviewedAt };
286
+ }
287
+ if (control.mode === "automated" && control.status !== "pass") {
288
+ return {
289
+ ...control,
290
+ summary: `${control.summary} An attestation cannot replace missing automated evidence.`,
291
+ evidence: [...control.evidence, ...attestationEvidence],
292
+ owner: attestation.owner,
293
+ reviewedAt: attestation.reviewedAt,
294
+ };
295
+ }
296
+ return {
297
+ ...control,
298
+ status: attestation.status,
299
+ summary: attestation.note ?? `Control attested ${attestation.status} by ${attestation.owner}.`,
300
+ evidence: [...control.evidence, ...attestationEvidence],
301
+ owner: attestation.owner,
302
+ reviewedAt: attestation.reviewedAt,
303
+ };
304
+ }
305
+ function compareWithBaseline(current, baseline, options) {
306
+ if (!baseline) {
307
+ return {
308
+ status: options.requireBaseline ? "fail" : "not-configured",
309
+ currentRunId: current.runId,
310
+ maxScoreDrop: options.maxScoreDrop,
311
+ regressions: options.requireBaseline ? ["A baseline evidence bundle is required but was not provided."] : [],
312
+ };
313
+ }
314
+ const regressions = [];
315
+ const scoreDelta = current.verdict.score - baseline.verdict.score;
316
+ if (current.subject.name !== baseline.subject.name) {
317
+ regressions.push(`Baseline subject ${baseline.subject.name} does not match current subject ${current.subject.name}.`);
318
+ }
319
+ if (baseline.verdict.passed && !current.verdict.passed)
320
+ regressions.push("Overall verdict changed from pass to fail.");
321
+ if (scoreDelta < -options.maxScoreDrop) {
322
+ regressions.push(`Overall score dropped by ${Math.abs(scoreDelta)} points; allowed drop is ${options.maxScoreDrop}.`);
323
+ }
324
+ for (const baselineResult of baseline.results) {
325
+ const currentResult = current.results.find((item) => item.product === baselineResult.product);
326
+ if (!currentResult) {
327
+ regressions.push(`${baselineResult.product} evidence is missing from the current run.`);
328
+ }
329
+ else if (baselineResult.passed && !currentResult.passed) {
330
+ regressions.push(`${baselineResult.product} changed from pass to fail.`);
331
+ }
332
+ }
333
+ return {
334
+ status: regressions.length === 0 ? "pass" : "fail",
335
+ baselineRunId: baseline.runId,
336
+ currentRunId: current.runId,
337
+ scoreDelta,
338
+ maxScoreDrop: options.maxScoreDrop,
339
+ regressions,
340
+ };
341
+ }
342
+ function collectBlockers(bundle, controls, regression, strict, requireBaseline, sourceArtifacts, evidenceBundle) {
343
+ const blockers = [];
344
+ if (!bundle.verdict.passed)
345
+ blockers.push("The unified AgentCert evidence bundle verdict failed.");
346
+ for (const control of controls) {
347
+ if (control.status === "fail")
348
+ blockers.push(`${control.name}: ${control.summary}`);
349
+ if (strict && control.status === "needs-evidence")
350
+ blockers.push(`${control.name}: evidence is required in strict mode.`);
351
+ if (strict && control.status === "manual-review")
352
+ blockers.push(`${control.name}: manual review is required in strict mode.`);
353
+ }
354
+ if (regression.status === "fail" && (regression.baselineRunId || requireBaseline))
355
+ blockers.push(...regression.regressions);
356
+ for (const artifact of [...sourceArtifacts, ...(evidenceBundle ? [evidenceBundle] : [])]) {
357
+ if (artifact.status === "missing")
358
+ blockers.push(`Referenced artifact is missing: ${artifact.path}`);
359
+ }
360
+ return [...new Set(blockers)];
361
+ }
362
+ async function digestArtifacts(input, cwd) {
363
+ const entries = Object.entries(input).filter((entry) => Boolean(entry[1]));
364
+ return Promise.all(entries.map(([name, path]) => digestArtifact(name, path, cwd)));
365
+ }
366
+ async function digestArtifact(name, path, cwd) {
367
+ if (/^[a-z][a-z0-9+.-]*:\/\//i.test(path))
368
+ return { name, path, status: "remote" };
369
+ const fullPath = resolve(cwd, path);
370
+ try {
371
+ await access(fullPath);
372
+ if ((await stat(fullPath)).isDirectory()) {
373
+ return { name, path: path.replace(/\\/g, "/"), status: "verified" };
374
+ }
375
+ const bytes = await readFile(fullPath);
376
+ return { name, path: path.replace(/\\/g, "/"), sha256: createHash("sha256").update(bytes).digest("hex"), status: "verified" };
377
+ }
378
+ catch {
379
+ return { name, path: path.replace(/\\/g, "/"), status: "missing" };
380
+ }
381
+ }
382
+ function resultEvidence(result) {
383
+ const evidence = evidenceReferences(result.evidence);
384
+ const artifacts = Object.values(result.artifacts).filter(Boolean);
385
+ return [...new Set([...evidence, ...artifacts])];
386
+ }
387
+ function evidenceReferences(evidence) {
388
+ return evidence.map((item) => item.artifactPath ?? `${item.source ?? "agentcert"}:${item.id}`);
389
+ }
390
+ function unresolved(status, summary) {
391
+ return { status, summary, evidence: [] };
392
+ }
393
+ function record(value) {
394
+ return value && typeof value === "object" && !Array.isArray(value) ? value : {};
395
+ }
396
+ function renderReleaseGateMarkdown(report) {
397
+ const lines = [
398
+ "# AgentCert Release Gate",
399
+ "",
400
+ `Subject: ${report.subject.name}`,
401
+ `Generated: ${report.generatedAt}`,
402
+ `Verdict: ${report.verdict.passed ? "PASS" : "FAIL"}`,
403
+ `Mode: ${report.strict ? "strict" : "advisory"}`,
404
+ "",
405
+ "## Controls",
406
+ "",
407
+ "| Control | Mode | Status | Summary |",
408
+ "|---|---|---|---|",
409
+ ...report.controls.map((control) => `| ${control.name} | ${control.mode} | ${control.status} | ${control.summary.replaceAll("|", "\\|")} |`),
410
+ "",
411
+ "## Regression",
412
+ "",
413
+ `Status: ${report.regression.status}`,
414
+ ...(report.regression.regressions.length > 0 ? report.regression.regressions.map((item) => `- ${item}`) : ["No regression detected or no baseline configured."]),
415
+ "",
416
+ "## Blockers",
417
+ "",
418
+ ...(report.verdict.blockers.length > 0 ? report.verdict.blockers.map((item) => `- ${item}`) : ["No release blockers."]),
419
+ "",
420
+ "## Provenance",
421
+ "",
422
+ ...[...(report.provenance.evidenceBundle ? [report.provenance.evidenceBundle] : []), ...report.provenance.sourceArtifacts].map((artifact) => `- ${artifact.name}: \`${artifact.path}\` (${artifact.status}${artifact.sha256 ? `, sha256 ${artifact.sha256}` : ""})`),
423
+ ];
424
+ return `${lines.join("\n")}\n`;
425
+ }
426
+ function renderReleaseGateHtml(report) {
427
+ const rows = report.controls.map((control) => `<tr><td>${escapeHtml(control.name)}</td><td>${escapeHtml(control.mode)}</td><td><strong class="${control.status}">${escapeHtml(control.status)}</strong></td><td>${escapeHtml(control.summary)}</td></tr>`).join("");
428
+ const blockers = report.verdict.blockers.length > 0 ? report.verdict.blockers.map((item) => `<li>${escapeHtml(item)}</li>`).join("") : "<li>No release blockers.</li>";
429
+ return `<!doctype html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>AgentCert Release Gate</title><style>:root{--ink:#132033;--muted:#617083;--line:#d7dee8;--bg:#f5f7fa;--pass:#087f5b;--fail:#c92a2a;--warn:#a45a00}body{margin:0;background:var(--bg);color:var(--ink);font:15px/1.5 system-ui,sans-serif}main{width:min(1100px,calc(100% - 32px));margin:auto;padding:36px 0}h1{font-size:42px;margin:0}.verdict{font-size:28px;color:${report.verdict.passed ? "var(--pass)" : "var(--fail)"}}section{background:#fff;border:1px solid var(--line);border-radius:8px;padding:20px;margin-top:16px;overflow:auto}table{width:100%;border-collapse:collapse;min-width:760px}th,td{text-align:left;padding:10px;border-bottom:1px solid var(--line);vertical-align:top}.pass{color:var(--pass)}.fail{color:var(--fail)}.needs-evidence,.manual-review{color:var(--warn)}p{color:var(--muted)}</style></head><body><main><p>AgentCert Assurance Control Plane</p><h1>${escapeHtml(report.subject.name)}</h1><strong class="verdict">${report.verdict.passed ? "RELEASE READY" : "RELEASE BLOCKED"}</strong><p>${escapeHtml(report.generatedAt)} · ${report.strict ? "strict" : "advisory"} mode · baseline ${escapeHtml(report.regression.status)}</p><section><h2>Ten release controls</h2><table><thead><tr><th>Control</th><th>Mode</th><th>Status</th><th>Decision</th></tr></thead><tbody>${rows}</tbody></table></section><section><h2>Blockers</h2><ul>${blockers}</ul></section></main></body></html>\n`;
430
+ }
431
+ function renderReleaseGateJunit(report) {
432
+ const failures = report.controls.filter((control) => control.status === "fail" || (report.strict && control.status !== "pass")).length + (report.regression.status === "fail" ? 1 : 0);
433
+ const skipped = (report.strict ? 0 : report.controls.filter((control) => control.status === "needs-evidence" || control.status === "manual-review").length) + (report.regression.status === "not-configured" ? 1 : 0);
434
+ const tests = report.controls.length + 1;
435
+ const cases = report.controls.map((control) => {
436
+ const body = control.status === "fail" || (report.strict && control.status !== "pass")
437
+ ? `<failure message="${escapeXml(control.summary)}"/>`
438
+ : control.status === "needs-evidence" || control.status === "manual-review"
439
+ ? `<skipped message="${escapeXml(control.summary)}"/>`
440
+ : "";
441
+ return `<testcase classname="agentcert.release-gate" name="${escapeXml(control.id)}">${body}</testcase>`;
442
+ });
443
+ const regressionBody = report.regression.status === "fail" ? `<failure message="${escapeXml(report.regression.regressions.join(" "))}"/>` : report.regression.status === "not-configured" ? '<skipped message="No baseline configured."/>' : "";
444
+ cases.push(`<testcase classname="agentcert.release-gate" name="continuous-regression">${regressionBody}</testcase>`);
445
+ return `<?xml version="1.0" encoding="UTF-8"?>\n<testsuite name="AgentCert release gate" tests="${tests}" failures="${failures}" skipped="${skipped}">\n ${cases.join("\n ")}\n</testsuite>\n`;
446
+ }
447
+ function renderReleaseGateBadge(report) {
448
+ const status = report.verdict.passed ? "ready" : "blocked";
449
+ const color = report.verdict.passed ? "#087f5b" : "#c92a2a";
450
+ return `<svg xmlns="http://www.w3.org/2000/svg" width="188" height="20" role="img" aria-label="agentcert release: ${status}"><title>agentcert release: ${status}</title><rect width="112" height="20" rx="3" fill="#132033"/><rect x="109" width="79" height="20" rx="3" fill="${color}"/><g fill="#fff" text-anchor="middle" font-family="Verdana,sans-serif" font-size="11"><text x="56" y="14">agentcert release</text><text x="149" y="14">${status}</text></g></svg>\n`;
451
+ }
452
+ function escapeHtml(value) {
453
+ return String(value).replaceAll("&", "&amp;").replaceAll("<", "&lt;").replaceAll(">", "&gt;").replaceAll('"', "&quot;");
454
+ }
455
+ function escapeXml(value) {
456
+ return escapeHtml(value).replaceAll("'", "&apos;");
457
+ }
@@ -1,3 +1,6 @@
1
+ import { AGENTCERT_EVIDENCE_SCHEMA_SEMVER, AGENTCERT_EVIDENCE_SCHEMA_VERSION } from "./types.js";
2
+ import { validateEvidenceSignature } from "./evidence-signing.js";
3
+ import { validateReleaseGateReport } from "./release-gate.js";
1
4
  export function parseSchemaId(input) {
2
5
  const value = input ?? "evidence-bundle";
3
6
  if (value === "evidence-bundle" ||
@@ -6,13 +9,23 @@ export function parseSchemaId(input) {
6
9
  value === "failure-review" ||
7
10
  value === "classifier-eval" ||
8
11
  value === "monitor-snapshot" ||
9
- value === "robustness-lab") {
12
+ value === "robustness-lab" ||
13
+ value === "release-gate" ||
14
+ value === "evidence-signature") {
10
15
  return value;
11
16
  }
12
- throw new Error(`Unsupported schema "${value}". Use evidence-bundle, result, corpus-record, failure-review, classifier-eval, monitor-snapshot, or robustness-lab.`);
17
+ throw new Error(`Unsupported schema "${value}". Use evidence-bundle, result, corpus-record, failure-review, classifier-eval, monitor-snapshot, robustness-lab, release-gate, or evidence-signature.`);
13
18
  }
14
19
  export function validateAgentCertSchema(schema, input) {
15
20
  const errors = [];
21
+ if (schema === "release-gate") {
22
+ errors.push(...validateReleaseGateReport(input));
23
+ return { schema, valid: errors.length === 0, errors };
24
+ }
25
+ if (schema === "evidence-signature") {
26
+ errors.push(...validateEvidenceSignature(input));
27
+ return { schema, valid: errors.length === 0, errors };
28
+ }
16
29
  const value = object(input, "$", errors);
17
30
  if (value) {
18
31
  if (schema === "evidence-bundle")
@@ -34,8 +47,8 @@ export function validateAgentCertSchema(schema, input) {
34
47
  }
35
48
  function validateEvidenceBundle(value, errors) {
36
49
  requiredConst(value, "schemaName", "agentcert.evidence_bundle", errors);
37
- requiredConst(value, "schemaVersion", "1", errors);
38
- requiredConst(value, "schemaSemver", "1.0.0", errors);
50
+ requiredConst(value, "schemaVersion", AGENTCERT_EVIDENCE_SCHEMA_VERSION, errors);
51
+ requiredConst(value, "schemaSemver", AGENTCERT_EVIDENCE_SCHEMA_SEMVER, errors);
39
52
  requiredConst(value, "kind", "agentcert.evidence_bundle", errors);
40
53
  requiredString(value, "runId", errors);
41
54
  requiredString(value, "generatedAt", errors);
package/dist/types.js CHANGED
@@ -1 +1,2 @@
1
- export {};
1
+ export const AGENTCERT_EVIDENCE_SCHEMA_VERSION = "agentcert.evidence.v0.1";
2
+ export const AGENTCERT_EVIDENCE_SCHEMA_SEMVER = "0.1.0";
package/package.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "agentcert",
3
- "version": "0.1.1",
4
- "description": "Regression CI, evidence reports, and failure corpus tooling for browser agents.",
3
+ "version": "0.2.0",
4
+ "description": "Assurance release gates, runtime evidence, and regression CI for AI agents.",
5
5
  "type": "module",
6
6
  "license": "Apache-2.0",
7
7
  "homepage": "https://github.com/Kakarottoooo/agentcert#readme",
@@ -22,7 +22,10 @@
22
22
  "llm",
23
23
  "mcp",
24
24
  "tripwire",
25
- "evidence"
25
+ "evidence",
26
+ "assurance",
27
+ "authorization",
28
+ "audit"
26
29
  ],
27
30
  "bin": {
28
31
  "agentcert": "dist/cli.js"