@pome-sh/cli 0.26.4 → 0.26.6

package/README.md

@@ -116,7 +116,7 @@ Three rules CI must honor:
 npm install
 npm run typecheck
 npm run build
-npm test
+npx vitest run --project cli
 ```
 
 The package publishes the `pome` binary from `dist/src/cli/main.js`.

package/dist/build-info.json

@@ -1,6 +1,6 @@
 {
   "package": "pome-sh",
-  "version": "0.26.4",
-  "git_sha": "8e396fcfae65745a93a57f64ca4faaec941816b7",
-  "build_time": "2026-08-22T03:12:21.394Z"
+  "version": "0.26.6",
+  "git_sha": "9913100d227924c71a6a77307fd3c042197576f9",
+  "build_time": "2026-08-22T06:27:19.126Z"
 }

package/dist/{checks-LKGNPTOL.js → checks-7SWBCD5C.js}

@@ -183,7 +183,7 @@ async function checkTwinReachable(_configDir) {
   });
   let harness;
   try {
-    const { bootTwin } = await import('./twinHarness-AJHGWE5B.js');
+    const { bootTwin } = await import('./twinHarness-PEPV4RQX.js');
     harness = await bootTwin({
       twin: "github",
       seedState: void 0,

package/dist/{chunk-NSFNITIC.js → chunk-4H3JXPGZ.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-VDHVVSUJ.js';
-import { createRecorder, bootTwin } from './chunk-4AY5VHQK.js';
+import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-WERW3JMN.js';
+import { createRecorder, bootTwin } from './chunk-GJXH37KY.js';
 import { eventSchema } from './chunk-VBATFCWR.js';
 import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
 import { serve } from '@hono/node-server';

package/dist/{chunk-K37RXI3K.js → chunk-5D2TGFE3.js}

@@ -39,7 +39,7 @@ var TWIN_REGISTRY = {
         defaultSeedState,
         GitHubDomain,
         openGitHubCloneDatabase
-      } = await import('./src-ZGAM3HXJ.js');
+      } = await import('./src-TBV553KB.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-ZGAM3HXJ.js');
+  const { createGitHubCloneApp } = await import('./src-TBV553KB.js');
   return createGitHubCloneApp();
 }
 function defaultPortFor(twin, env = process.env) {

package/dist/{chunk-4AY5VHQK.js → chunk-GJXH37KY.js}

@@ -1,4 +1,4 @@
-import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-K37RXI3K.js';
+import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-5D2TGFE3.js';
 import { createFileBackedRecorderStore, createRecorderStore } from './chunk-WEKJIGUD.js';
 
 // src/recorder/recorder.ts

package/dist/{chunk-IBB4I5PM.js → chunk-M4IK4D4X.js}

@@ -1,5 +1,5 @@
 import { readManifest, normalizeManifestTwins } from './chunk-LA6BHJ5H.js';
-import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-VDHVVSUJ.js';
+import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-WERW3JMN.js';
 import { MOUNTED_TWINS, HostedAuthError, HostedDiscardRefusedError, HostedQuotaError, HostedOrchError, HostedTrialError, agentResponseSchema } from './chunk-QYCUNR2V.js';
 import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
 import { randomUUID, createHash } from 'node:crypto';

package/dist/{chunk-VDHVVSUJ.js → chunk-WERW3JMN.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-K37RXI3K.js';
+import { isTwinName, TWIN_REGISTRY } from './chunk-5D2TGFE3.js';
 import { toTwinHttpEventRow } from './chunk-WEKJIGUD.js';
 import { redactEvent, redactSecrets } from './chunk-SG6ZTIMT.js';
 import { mkdir, appendFile, writeFile, readFile } from 'node:fs/promises';

package/dist/{runDemo-NQV256MB.js → runDemo-EPTIRUX6.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-NSFNITIC.js';
+import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-4H3JXPGZ.js';
 import { getAvailablePort } from './chunk-XDU6TD4O.js';
 import './chunk-CBFKZZBR.js';
-import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-VDHVVSUJ.js';
+import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-WERW3JMN.js';
 import './chunk-NW7HGA2K.js';
 import { HostedQuotaError, HostedOrchError } from './chunk-QYCUNR2V.js';
 import './chunk-B22TYIEM.js';
 import './chunk-NJ246QPJ.js';
 import './chunk-IZIZC7ZW.js';
-import { bootTwin } from './chunk-4AY5VHQK.js';
-import './chunk-K37RXI3K.js';
+import { bootTwin } from './chunk-GJXH37KY.js';
+import './chunk-5D2TGFE3.js';
 import './chunk-WEKJIGUD.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/dist/{runTrialGroup-S7CRQPIM.js → runTrialGroup-Z4PB3WGJ.js}

@@ -1,13 +1,13 @@
 import { newGroupId, criterionPhrase } from './chunk-RGZBC7NF.js';
-import { runTaskHosted, resolveRunAgentIdentity } from './chunk-IBB4I5PM.js';
+import { runTaskHosted, resolveRunAgentIdentity } from './chunk-M4IK4D4X.js';
 import './chunk-LA6BHJ5H.js';
-import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-VDHVVSUJ.js';
+import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-WERW3JMN.js';
 import './chunk-NW7HGA2K.js';
 import { HostedQuotaError, HostedTrialError } from './chunk-QYCUNR2V.js';
 import './chunk-B22TYIEM.js';
 import './chunk-NJ246QPJ.js';
 import './chunk-IZIZC7ZW.js';
-import './chunk-K37RXI3K.js';
+import './chunk-5D2TGFE3.js';
 import './chunk-WEKJIGUD.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/dist/src/cli/main.js

@@ -1,10 +1,10 @@
 #!/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-IBB4I5PM.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-M4IK4D4X.js';
 import { readManifest, writeManifest, MANIFEST_JSON, readRequiredManifest, normalizeManifestTwins } from '../../chunk-LA6BHJ5H.js';
-import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-NSFNITIC.js';
+import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-4H3JXPGZ.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-VDHVVSUJ.js';
+import { parseTaskFile, scoreStatus, runScoreLine, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, markerFor, criterionMarkerLabel, twinSkipSuffix, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-WERW3JMN.js';
 import '../../chunk-NW7HGA2K.js';
 import { MOUNTED_TWINS, deriveAgentSlug, exitCodeFor, HostedUsageError, HostedOrchError, HostedAuthError, HostedQuotaError } from '../../chunk-QYCUNR2V.js';
 import { TAPE_ASSERTABLE_TOOLS } from '../../chunk-C4BVTUA3.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-4AY5VHQK.js';
-import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-K37RXI3K.js';
+import '../../chunk-GJXH37KY.js';
+import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-5D2TGFE3.js';
 import '../../chunk-WEKJIGUD.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-ZGAM3HXJ.js');
+  const { GitHubDomain, openGitHubCloneDatabase } = await import('../../src-TBV553KB.js');
   const db = openGitHubCloneDatabase(":memory:");
   try {
     new GitHubDomain(db).seed(seed);
@@ -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 ("0.26.4".length > 0) return "0.26.4";
+  if ("0.26.6".length > 0) return "0.26.6";
   try {
     const here = dirname(fileURLToPath(import.meta.url));
     const candidates = [
@@ -5096,7 +5096,7 @@ function createProgram() {
         return;
       }
       {
-        const { runDoctorChecks } = await import('../../checks-LKGNPTOL.js');
+        const { runDoctorChecks } = await import('../../checks-7SWBCD5C.js');
         const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
         const doctorReport = await runDoctorChecks({ mode: useLocal ? "full" : "hosted" });
         if (!doctorReport.ok) {
@@ -5134,7 +5134,7 @@ function createProgram() {
               taskForRuns.config.runs
             );
             if (k > 1) {
-              const { runTrialGroup } = await import('../../runTrialGroup-S7CRQPIM.js');
+              const { runTrialGroup } = await import('../../runTrialGroup-Z4PB3WGJ.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 ${k}`;
               const groupResult = await runTrialGroup({
@@ -5228,7 +5228,7 @@ function createProgram() {
         process.exitCode = 5;
         return;
       }
-      const { runDemo } = await import('../../runDemo-NQV256MB.js');
+      const { runDemo } = await import('../../runDemo-EPTIRUX6.js');
       const result = await runDemo({
         apiBase: opts.apiUrl.replace(/\/$/, ""),
         dashboardBase: process.env.POME_DASHBOARD_URL ?? DEFAULT_DASHBOARD_URL,
@@ -5246,7 +5246,7 @@ 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-LKGNPTOL.js');
+    const { runDoctorChecks } = await import('../../checks-7SWBCD5C.js');
     const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
     const report = await runDoctorChecks();
     for (const line of renderDoctorReport(report, { passNote: true })) console.error(line);
@@ -5422,7 +5422,7 @@ 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-QXSCA577.js');
+    const { runTwinStartCommand } = await import('../../twinStart-CP4UAY5W.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-ZGAM3HXJ.js → src-TBV553KB.js}

@@ -623,7 +623,7 @@ var mcp_tools_list_meta_default = {
     'What this file does NOT say is what the twin ACCEPTS. These are GitHub\'s advertised schemas; the twin\'s zod validators are what actually admit or refuse an examinee\'s call, and the two can disagree \u2014 GitHub advertises `perPage` where the validator also takes `per_page`, and advertises state ["OPEN","CLOSED"] where the validator still takes lowercase. src/tool-schema-conformance.ts is the report on that gap. Reading this fixture as a statement about behaviour is the mistake it exists to make visible.',
     "Dropping a tool from this listing does NOT drop the twin's REST route for the same operation. Fourteen of the thirty-four names retired by F-1376 (milestones, commit statuses, check runs, issue-comment edit/delete) were GitHub REST operations that were never MCP tools upstream, and they remain reachable exactly the way an examinee could always have reached them at GitHub: over HTTP."
   ],
-  canonicalFileSha256: "ccd957048004c010791e19999d1aff507791171dec4882d5a3e171e22482f13b",
+  canonicalFileSha256: "223917a5b2669df70622bbd48cf566f6b8258c902dbff5cc274a5318ddf1264a",
   files: {
     raw: "mcp-tools-list.raw.json",
     canonical: "mcp-tools-list.canonical.json"
@@ -631,7 +631,7 @@ var mcp_tools_list_meta_default = {
   configuration: {
     toolsets: "default",
     whyThisToolset: [
-      "examples/support-triage declares mcp_servers[name=github].url = https://api.githubcopilot.com/mcp/ (agents/support-triage-v1.yaml and -v2.yaml, byte-identical on that line).",
+      "agent-examples/support-triage declares mcp_servers[name=github].url = https://api.githubcopilot.com/mcp/ (agents/support-triage-v1.yaml and -v2.yaml, byte-identical on that line).",
       "docs/remote-server.md at the pinned commit maps that exact URL to the `default` toolset. `all` is a DIFFERENT URL: https://api.githubcopilot.com/mcp/x/all.",
       "Measured at the pinned commit: --toolsets=default returns 44 tools, --toolsets=all returns 85. Reading the union would invent 41 tools no examinee of ours can call."
     ],

package/dist/twinHarness-PEPV4RQX.js

@@ -0,0 +1,5 @@
+export { UnsupportedTwinError, bootTwin } from './chunk-GJXH37KY.js';
+export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-5D2TGFE3.js';
+import './chunk-WEKJIGUD.js';
+import './chunk-VBATFCWR.js';
+import './chunk-SG6ZTIMT.js';

package/dist/{twinStart-QXSCA577.js → twinStart-CP4UAY5W.js}

@@ -1,5 +1,5 @@
-import { bootTwin } from './chunk-4AY5VHQK.js';
-import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-K37RXI3K.js';
+import { bootTwin } from './chunk-GJXH37KY.js';
+import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-5D2TGFE3.js';
 import './chunk-WEKJIGUD.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/cli",
-  "version": "0.26.4",
+  "version": "0.26.6",
   "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",
@@ -48,8 +48,6 @@
     "pome": "node dist/src/cli/main.js",
     "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/twinHarness-AJHGWE5B.js

@@ -1,5 +0,0 @@
-export { UnsupportedTwinError, bootTwin } from './chunk-4AY5VHQK.js';
-export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-K37RXI3K.js';
-import './chunk-WEKJIGUD.js';
-import './chunk-VBATFCWR.js';
-import './chunk-SG6ZTIMT.js';