@pome-sh/cli 0.23.48 → 0.23.50

package/dist/build-info.json

@@ -1,6 +1,6 @@
 {
   "package": "pome-sh",
-  "version": "0.23.48",
-  "git_sha": "0dd8faf57062539f921e76f510d0f2cea847a104",
-  "build_time": "2026-08-13T20:30:07.412Z"
+  "version": "0.23.50",
+  "git_sha": "5bce2d403766321d74eda19b4281b05bd6f65037",
+  "build_time": "2026-08-14T19:50:13.287Z"
 }

package/dist/{checks-STMFOZA5.js → checks-YRMP2JX7.js}

@@ -183,7 +183,7 @@ async function checkTwinReachable(_configDir) {
   });
   let harness;
   try {
-    const { bootTwin } = await import('./twinHarness-WIWX5USC.js');
+    const { bootTwin } = await import('./twinHarness-5UJQR4QM.js');
     harness = await bootTwin({
       twin: "github",
       seedState: void 0,

package/dist/chunk-C4BVTUA3.js

@@ -0,0 +1,8 @@
+// ../packages/twin-github/dist/src/tape-assertable-tools.js
+var TAPE_ASSERTABLE_TOOLS = [
+  "create_commit_status",
+  "create_check_run",
+  "add_issue_comment"
+];
+
+export { TAPE_ASSERTABLE_TOOLS };

package/dist/{chunk-4YGE4L6O.js → chunk-ECMAQH5H.js}

@@ -1,5 +1,5 @@
 import { readManifest, normalizeManifestTwins } from './chunk-ABM3CMQB.js';
-import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-BZ3RQU56.js';
+import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-UZDZM67T.js';
 import { MOUNTED_TWINS, HostedAuthError, HostedDiscardRefusedError, HostedQuotaError, HostedOrchError, HostedTrialError, agentResponseSchema } from './chunk-X66JOOO7.js';
 import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
 import { randomUUID, createHash } from 'node:crypto';

package/dist/{chunk-XWQK4XXD.js → chunk-P6PGKXS5.js}

@@ -1,7 +1,7 @@
 import { getAvailablePort } from './chunk-XDU6TD4O.js';
 import { buildEgressAllowlist, readBlockedEgress } from './chunk-CBFKZZBR.js';
-import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-BZ3RQU56.js';
-import { createRecorder, bootTwin } from './chunk-WMHGTTRH.js';
+import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-UZDZM67T.js';
+import { createRecorder, bootTwin } from './chunk-SIJP5OCX.js';
 import { eventSchema } from './chunk-VBATFCWR.js';
 import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
 import { serve } from '@hono/node-server';

package/dist/{chunk-DZU4LAYD.js → chunk-R6YWMPGO.js}

@@ -1,6 +1,6 @@
 // ../packages/twin-github/package.json
 var package_default = {
-  version: "0.11.0"};
+  version: "0.12.0"};
 
 // ../packages/twin-slack/package.json
 var package_default2 = {
@@ -39,7 +39,7 @@ var TWIN_REGISTRY = {
         defaultSeedState,
         GitHubDomain,
         openGitHubCloneDatabase
-      } = await import('./src-7Z35KM4S.js');
+      } = await import('./src-BHCAQQF7.js');
       const db = openGitHubCloneDatabase();
       const domain = new GitHubDomain(db);
       domain.seed(seedState === void 0 ? defaultSeedState() : seedState);
@@ -173,7 +173,7 @@ var TWIN_REGISTRY = {
   }
 };
 async function createGitHubSmokeApp() {
-  const { createGitHubCloneApp } = await import('./src-7Z35KM4S.js');
+  const { createGitHubCloneApp } = await import('./src-BHCAQQF7.js');
   return createGitHubCloneApp();
 }
 function defaultPortFor(twin, env = process.env) {

package/dist/{chunk-WMHGTTRH.js → chunk-SIJP5OCX.js}

@@ -1,4 +1,4 @@
-import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-DZU4LAYD.js';
+import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-R6YWMPGO.js';
 import { createFileBackedRecorderStore, createRecorderStore } from './chunk-5EGHN6H4.js';
 
 // src/recorder/recorder.ts

package/dist/{chunk-BZ3RQU56.js → chunk-UZDZM67T.js}

@@ -2,7 +2,7 @@ import { criterionSchema, finalizeResponseSchema, HostedDiscardRefusedError, Hos
 import { seedSchema, parseSeed, defaultSeedState as defaultSeedState$2 } from './chunk-B22TYIEM.js';
 import { gmailSeedSchema, defaultSeedState } from './chunk-NJ246QPJ.js';
 import { linearSeedSchema, defaultSeedState as defaultSeedState$1 } from './chunk-IZIZC7ZW.js';
-import { isTwinName, TWIN_REGISTRY } from './chunk-DZU4LAYD.js';
+import { isTwinName, TWIN_REGISTRY } from './chunk-R6YWMPGO.js';
 import { toTwinHttpEventRow } from './chunk-5EGHN6H4.js';
 import { redactEvent, redactSecrets } from './chunk-SG6ZTIMT.js';
 import { mkdir, appendFile, writeFile, readFile } from 'node:fs/promises';

package/dist/{runDemo-63ZG7QKS.js → runDemo-EK4AMZVR.js}

@@ -1,16 +1,16 @@
 import { newGroupId, reassuranceBox, twinReadyLine, trialsHeaderLine, trialLine, summaryLines, evaluatingLine, criterionPhrase } from './chunk-RGZBC7NF.js';
 import { DemoCapacityError, capacityLabel, parseCapacityMarker, capacityKindFrom } from './chunk-ZX4WNSZ5.js';
-import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-XWQK4XXD.js';
+import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-P6PGKXS5.js';
 import { getAvailablePort } from './chunk-XDU6TD4O.js';
 import './chunk-CBFKZZBR.js';
-import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-BZ3RQU56.js';
+import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-UZDZM67T.js';
 import './chunk-NW7HGA2K.js';
 import { HostedQuotaError, HostedOrchError } from './chunk-X66JOOO7.js';
 import './chunk-B22TYIEM.js';
 import './chunk-NJ246QPJ.js';
 import './chunk-IZIZC7ZW.js';
-import { bootTwin } from './chunk-WMHGTTRH.js';
-import './chunk-DZU4LAYD.js';
+import { bootTwin } from './chunk-SIJP5OCX.js';
+import './chunk-R6YWMPGO.js';
 import './chunk-5EGHN6H4.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/dist/{runTrialGroup-SG4QQVTN.js → runTrialGroup-RLR2JD3T.js}

@@ -1,13 +1,13 @@
 import { newGroupId, criterionPhrase } from './chunk-RGZBC7NF.js';
-import { runTaskHosted, resolveRunAgentIdentity } from './chunk-4YGE4L6O.js';
+import { runTaskHosted, resolveRunAgentIdentity } from './chunk-ECMAQH5H.js';
 import './chunk-ABM3CMQB.js';
-import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-BZ3RQU56.js';
+import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-UZDZM67T.js';
 import './chunk-NW7HGA2K.js';
 import { HostedQuotaError, HostedTrialError } from './chunk-X66JOOO7.js';
 import './chunk-B22TYIEM.js';
 import './chunk-NJ246QPJ.js';
 import './chunk-IZIZC7ZW.js';
-import './chunk-DZU4LAYD.js';
+import './chunk-R6YWMPGO.js';
 import './chunk-5EGHN6H4.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/dist/src/cli/main.js

@@ -1,13 +1,13 @@
 #!/usr/bin/env node
-import { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, clearLocalCredentials, friendlyHostedError, runSessionCreate, runSessionList, runSessionStop, resolveCredentials, runTaskHosted, discoverRunSet, VERDICT_ARTIFACT_VERSION, persistCredentialsAfterLogin, DEFAULT_DOCS_SITE_ORIGIN, resolveSeams, resolveCachedAgentId, readLinkCache, postAgentResolver, writeLinkCache, ensurePomeGitignored } from '../../chunk-4YGE4L6O.js';
+import { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, clearLocalCredentials, friendlyHostedError, runSessionCreate, runSessionList, runSessionStop, resolveCredentials, runTaskHosted, discoverRunSet, VERDICT_ARTIFACT_VERSION, persistCredentialsAfterLogin, DEFAULT_DOCS_SITE_ORIGIN, resolveSeams, resolveCachedAgentId, readLinkCache, postAgentResolver, writeLinkCache, ensurePomeGitignored } from '../../chunk-ECMAQH5H.js';
 import { readManifest, writeManifest, MANIFEST_JSON, readRequiredManifest, normalizeManifestTwins } from '../../chunk-ABM3CMQB.js';
-import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-XWQK4XXD.js';
+import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-P6PGKXS5.js';
 import '../../chunk-XDU6TD4O.js';
 import '../../chunk-CBFKZZBR.js';
-import { parseTaskFile, scoreStatus, runScoreLine, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, markerFor, criterionMarkerLabel, twinSkipSuffix, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-BZ3RQU56.js';
+import { parseTaskFile, scoreStatus, runScoreLine, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, markerFor, criterionMarkerLabel, twinSkipSuffix, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-UZDZM67T.js';
 import '../../chunk-NW7HGA2K.js';
 import { MOUNTED_TWINS, deriveAgentSlug, exitCodeFor, HostedUsageError, HostedOrchError, HostedAuthError, HostedQuotaError } from '../../chunk-X66JOOO7.js';
-import { TAPE_ASSERTABLE_TOOLS } from '../../chunk-FKZZWWYC.js';
+import { TAPE_ASSERTABLE_TOOLS } from '../../chunk-C4BVTUA3.js';
 import { seedSchema } from '../../chunk-B22TYIEM.js';
 import { SLACK_CHECKS } from '../../chunk-IDNSKKEC.js';
 import { GMAIL_CHECKS } from '../../chunk-JM6VS62R.js';
@@ -15,8 +15,8 @@ import '../../chunk-NJ246QPJ.js';
 import { LINEAR_CHECKS } from '../../chunk-XBT6ZLBG.js';
 import '../../chunk-IZIZC7ZW.js';
 import { oneOf, defineCheck, repoRef, VACUITY_SENTINEL_NUMBER, childStatePath, VACUITY_SENTINEL, statePath, templateSlots, renderCheck, checksDigest, checkPattern, checkNearMissPattern } from '../../chunk-JWJYNAWI.js';
-import '../../chunk-WMHGTTRH.js';
-import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-DZU4LAYD.js';
+import '../../chunk-SIJP5OCX.js';
+import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-R6YWMPGO.js';
 import '../../chunk-5EGHN6H4.js';
 import { isLegacyEventRow, eventSchema } from '../../chunk-VBATFCWR.js';
 import { redactEvent, redactSecrets } from '../../chunk-SG6ZTIMT.js';
@@ -850,7 +850,7 @@ async function throwForStatus(res) {
 
 // src/task/seed-verifier.ts
 async function verifySeedWithTwin(seed) {
-  const { GitHubDomain, openGitHubCloneDatabase } = await import('../../src-7Z35KM4S.js');
+  const { GitHubDomain, openGitHubCloneDatabase } = await import('../../src-BHCAQQF7.js');
   const db = openGitHubCloneDatabase(":memory:");
   try {
     new GitHubDomain(db).seed(seed);
@@ -2171,7 +2171,7 @@ var toolNeverCalled = defineCheck({
   // that a verified fact rather than an assumption.
   subject: (args) => args.tool,
   // Null, and admitted in `HONEST_NULL_MUTANTS`. The only slot is a closed set,
-  // so there is no value guaranteed to be false: a mutant naming the OTHER
+  // so there is no value guaranteed to be false: a mutant naming ANOTHER
   // assertable action asserts something that may well also be true, and a value
   // outside the set does not re-bind at all — which reads as "the verdict moved"
   // and would bless the very criterion the probe exists to catch.
@@ -2221,9 +2221,9 @@ var toolWasCalled = defineCheck({
   // nothing, i.e. failing an agent that did the work.
   subject: (args) => args.tool,
   // Null, and admitted in `HONEST_NULL_MUTANTS`. Same closed-set argument as the
-  // sibling: the only substitutable value is the OTHER assertable action, which
-  // an agent may well have called too, and a value outside the set does not
-  // re-bind at all.
+  // sibling: every substitutable value is ANOTHER assertable action, which an
+  // agent may well have called too, and a value outside the set does not re-bind
+  // at all.
   vacuityMutant: () => null,
   // A POSITIVE check, so the passing world is the one where the action WAS
   // called. The failing world is deliberately NOT an empty tape: `[]` fails
@@ -4651,7 +4651,7 @@ var DEFAULT_AGENT_COMMAND = `node ${DEFAULT_AGENT_FILE}`;
 var MANIFEST_SCHEMA_URL = "https://pome.sh/schemas/v1/pome.json";
 var MAX_UNREADABLE_PATHS_SHOWN = 5;
 function readPackageVersion() {
-  if ("0.23.48".length > 0) return "0.23.48";
+  if ("0.23.50".length > 0) return "0.23.50";
   try {
     const here = dirname(fileURLToPath(import.meta.url));
     const candidates = [
@@ -5106,7 +5106,7 @@ function createProgram() {
         return;
       }
       {
-        const { runDoctorChecks } = await import('../../checks-STMFOZA5.js');
+        const { runDoctorChecks } = await import('../../checks-YRMP2JX7.js');
         const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
         const doctorReport = await runDoctorChecks({ mode: useLocal ? "full" : "hosted" });
         if (!doctorReport.ok) {
@@ -5144,7 +5144,7 @@ function createProgram() {
               taskForRuns.config.runs
             );
             if (k > 1) {
-              const { runTrialGroup } = await import('../../runTrialGroup-SG4QQVTN.js');
+              const { runTrialGroup } = await import('../../runTrialGroup-RLR2JD3T.js');
               const fileForRerun = relative(process.cwd(), file);
               const rerunCommand = defaultTask ? options.trials !== void 0 ? `pome run -n ${k}` : "pome run" : `pome run ${fileForRerun && !fileForRerun.startsWith("..") ? fileForRerun : file} -n ${k}`;
               const groupResult = await runTrialGroup({
@@ -5238,7 +5238,7 @@ function createProgram() {
         process.exitCode = 5;
         return;
       }
-      const { runDemo } = await import('../../runDemo-63ZG7QKS.js');
+      const { runDemo } = await import('../../runDemo-EK4AMZVR.js');
       const result = await runDemo({
         apiBase: opts.apiUrl.replace(/\/$/, ""),
         dashboardBase: process.env.POME_DASHBOARD_URL ?? DEFAULT_DASHBOARD_URL,
@@ -5256,7 +5256,7 @@ function createProgram() {
   program.command("doctor").description(
     "Check the agent\u2194twin wiring: pome.json (or pome.yaml) present + valid, the local twin boots + serves, requests routed to the twin (not a hardcoded production host), egress floor active. On failure prints one named cause (file:line where knowable) + one concrete fix and exits non-zero."
   ).action(async () => {
-    const { runDoctorChecks } = await import('../../checks-STMFOZA5.js');
+    const { runDoctorChecks } = await import('../../checks-YRMP2JX7.js');
     const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
     const report = await runDoctorChecks();
     for (const line of renderDoctorReport(report, { passNote: true })) console.error(line);
@@ -5432,7 +5432,7 @@ function createProgram() {
   ).description(
     "Start a standalone twin as a long-lived foreground server (Ctrl-C to stop)"
   ).action(async (name, options) => {
-    const { runTwinStartCommand } = await import('../../twinStart-Q4EJ4U2C.js');
+    const { runTwinStartCommand } = await import('../../twinStart-BF375EBE.js');
     await runTwinStartCommand(name, options);
   });
   twin.command("reset").argument("[name]", "Twin name (default: github)", "github").description("Reset standalone twin state").action(async (name) => {

package/dist/{src-7Z35KM4S.js → src-BHCAQQF7.js}

@@ -1,4 +1,4 @@
-import './chunk-FKZZWWYC.js';
+import './chunk-C4BVTUA3.js';
 import { defaultSeedState, parseSeed } from './chunk-B22TYIEM.js';
 export { defaultSeedState, parseSeed, seedSchema } from './chunk-B22TYIEM.js';
 import { routeInputDeclarer, integerInput, mountDeclaredRoute, UndeclaredInputError, MalformedBodyError } from './chunk-2Q3P45LK.js';
@@ -5408,7 +5408,7 @@ function registerGitHubRoutes(session, { domain, recorder }) {
   route(GITHUB_ROUTES.addIssueComment, ({ path, body }) => {
     const { value, delta } = captureDelta((onDelta) => domain.addIssueComment({ ...issueRef(path), ...body }, onDelta));
     return created(value, delta);
-  });
+  }, "add_issue_comment");
   route(GITHUB_ROUTES.listRepositoryLabels, ({ path }) => ok(domain.listRepositoryLabels(path)));
   route(GITHUB_ROUTES.createRepositoryLabel, ({ path, body }) => {
     const { value, delta } = captureDelta((onDelta) => domain.createRepositoryLabel({ ...path, ...body }, onDelta));

package/dist/twinHarness-5UJQR4QM.js

@@ -0,0 +1,5 @@
+export { UnsupportedTwinError, bootTwin } from './chunk-SIJP5OCX.js';
+export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-R6YWMPGO.js';
+import './chunk-5EGHN6H4.js';
+import './chunk-VBATFCWR.js';
+import './chunk-SG6ZTIMT.js';

package/dist/{twinStart-Q4EJ4U2C.js → twinStart-BF375EBE.js}

@@ -1,5 +1,5 @@
-import { bootTwin } from './chunk-WMHGTTRH.js';
-import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-DZU4LAYD.js';
+import { bootTwin } from './chunk-SIJP5OCX.js';
+import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-R6YWMPGO.js';
 import './chunk-5EGHN6H4.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/cli",
-  "version": "0.23.48",
+  "version": "0.23.50",
   "description": "Digital-twin testing for AI agents \u2014 run tasks against resettable local or hosted twins and record tool-call traces for evaluation on pome.sh.",
   "keywords": [
     "ai",
@@ -49,6 +49,7 @@
     "prepublishOnly": "node scripts/assert-publishable.mjs && npm run build",
     "typecheck": "tsc -p tsconfig.json --noEmit",
     "test": "vitest run",
+    "gate:golden": "vitest run test/golden",
     "gate:recorder-overhead": "tsx scripts/recorder-overhead-gate.ts",
     "gate:agent-trace-overhead": "tsx scripts/overhead-gate.ts",
     "gate:cas-adapter-acceptance": "tsx scripts/cas-adapter-acceptance.ts",

package/dist/chunk-FKZZWWYC.js

@@ -1,4 +0,0 @@
-// ../packages/twin-github/dist/src/tape-assertable-tools.js
-var TAPE_ASSERTABLE_TOOLS = ["create_commit_status", "create_check_run"];
-
-export { TAPE_ASSERTABLE_TOOLS };

package/dist/twinHarness-WIWX5USC.js

@@ -1,5 +0,0 @@
-export { UnsupportedTwinError, bootTwin } from './chunk-WMHGTTRH.js';
-export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-DZU4LAYD.js';
-import './chunk-5EGHN6H4.js';
-import './chunk-VBATFCWR.js';
-import './chunk-SG6ZTIMT.js';