@pome-sh/cli 0.23.37 → 0.23.39

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.
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "package": "pome-sh",
3
- "version": "0.23.37",
4
- "git_sha": "516d235ab8150a3823d3ac09d3cb667fb2fdd30a",
5
- "build_time": "2026-08-13T01:35:08.896Z"
3
+ "version": "0.23.39",
4
+ "git_sha": "7f1b85aeb0da4acd4547c537aa24c4347ad70b40",
5
+ "build_time": "2026-08-13T01:54:45.694Z"
6
6
  }
@@ -1,6 +1,6 @@
1
1
  import { getAvailablePort } from './chunk-XDU6TD4O.js';
2
2
  import { buildEgressAllowlist, readBlockedEgress } from './chunk-CBFKZZBR.js';
3
- import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-EKQQYW7E.js';
3
+ import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-SAYISSCP.js';
4
4
  import { createRecorder, bootTwin } from './chunk-YL4FHI3J.js';
5
5
  import { eventSchema } from './chunk-VBATFCWR.js';
6
6
  import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
@@ -1,5 +1,5 @@
1
1
  import { readManifest, normalizeManifestTwins } from './chunk-ABM3CMQB.js';
2
- import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-EKQQYW7E.js';
2
+ import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-SAYISSCP.js';
3
3
  import { MOUNTED_TWINS, HostedAuthError, HostedDiscardRefusedError, HostedQuotaError, HostedOrchError, HostedTrialError, agentResponseSchema } from './chunk-X66JOOO7.js';
4
4
  import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
5
5
  import { randomUUID, createHash } from 'node:crypto';
@@ -1248,13 +1248,15 @@ function normalizeAgentSdk(value) {
1248
1248
  return trimmed.length > 0 ? trimmed : null;
1249
1249
  }
1250
1250
 
1251
+ // ../packages/wire/dist/run-completeness.js
1252
+ var PRE_SATISFIED_REASON = "already_true_in_seed";
1253
+
1251
1254
  // src/hosted/evalResultView.ts
1252
1255
  function outcomeOf(result) {
1253
1256
  if (result.outcome) return result.outcome;
1254
1257
  if (result.skipped) return "skipped";
1255
1258
  return result.passed ? "passed" : "failed";
1256
1259
  }
1257
- var PRE_SATISFIED_REASON = "already_true_in_seed";
1258
1260
  function isPreSatisfied(result) {
1259
1261
  return result.skipped && result.reason === PRE_SATISFIED_REASON;
1260
1262
  }
@@ -1,9 +1,9 @@
1
1
  import { newGroupId, reassuranceBox, twinReadyLine, trialsHeaderLine, trialLine, summaryLines, evaluatingLine, criterionPhrase } from './chunk-RGZBC7NF.js';
2
2
  import { DemoCapacityError, capacityLabel, parseCapacityMarker, capacityKindFrom } from './chunk-ZX4WNSZ5.js';
3
- import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-7GHIB4HM.js';
3
+ import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-5UBTKN7K.js';
4
4
  import { getAvailablePort } from './chunk-XDU6TD4O.js';
5
5
  import './chunk-CBFKZZBR.js';
6
- import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-EKQQYW7E.js';
6
+ import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-SAYISSCP.js';
7
7
  import './chunk-NW7HGA2K.js';
8
8
  import { HostedQuotaError, HostedOrchError } from './chunk-X66JOOO7.js';
9
9
  import './chunk-KIUIKAVU.js';
@@ -1,7 +1,7 @@
1
1
  import { newGroupId, criterionPhrase } from './chunk-RGZBC7NF.js';
2
- import { runTaskHosted, resolveRunAgentIdentity } from './chunk-MUQX7EHA.js';
2
+ import { runTaskHosted, resolveRunAgentIdentity } from './chunk-G4B64MFJ.js';
3
3
  import './chunk-ABM3CMQB.js';
4
- import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-EKQQYW7E.js';
4
+ import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-SAYISSCP.js';
5
5
  import './chunk-NW7HGA2K.js';
6
6
  import { HostedQuotaError, HostedTrialError } from './chunk-X66JOOO7.js';
7
7
  import './chunk-KIUIKAVU.js';
@@ -1,10 +1,10 @@
1
1
  #!/usr/bin/env node
2
- 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-MUQX7EHA.js';
2
+ 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-G4B64MFJ.js';
3
3
  import { readManifest, writeManifest, MANIFEST_JSON, readRequiredManifest, normalizeManifestTwins } from '../../chunk-ABM3CMQB.js';
4
- import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-7GHIB4HM.js';
4
+ import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-5UBTKN7K.js';
5
5
  import '../../chunk-XDU6TD4O.js';
6
6
  import '../../chunk-CBFKZZBR.js';
7
- import { parseTaskFile, scoreStatus, runScoreLine, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, markerFor, criterionMarkerLabel, twinSkipSuffix, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-EKQQYW7E.js';
7
+ import { parseTaskFile, scoreStatus, runScoreLine, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, markerFor, criterionMarkerLabel, twinSkipSuffix, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-SAYISSCP.js';
8
8
  import '../../chunk-NW7HGA2K.js';
9
9
  import { MOUNTED_TWINS, deriveAgentSlug, exitCodeFor, HostedUsageError, HostedOrchError, HostedAuthError, HostedQuotaError } from '../../chunk-X66JOOO7.js';
10
10
  import { TAPE_ASSERTABLE_TOOLS } from '../../chunk-FKZZWWYC.js';
@@ -4580,7 +4580,7 @@ var DEFAULT_AGENT_COMMAND = `node ${DEFAULT_AGENT_FILE}`;
4580
4580
  var MANIFEST_SCHEMA_URL = "https://pome.sh/schemas/v1/pome.json";
4581
4581
  var MAX_UNREADABLE_PATHS_SHOWN = 5;
4582
4582
  function readPackageVersion() {
4583
- if ("0.23.37".length > 0) return "0.23.37";
4583
+ if ("0.23.39".length > 0) return "0.23.39";
4584
4584
  try {
4585
4585
  const here = dirname(fileURLToPath(import.meta.url));
4586
4586
  const candidates = [
@@ -5073,7 +5073,7 @@ function createProgram() {
5073
5073
  taskForRuns.config.runs
5074
5074
  );
5075
5075
  if (k > 1) {
5076
- const { runTrialGroup } = await import('../../runTrialGroup-6FWGZGNM.js');
5076
+ const { runTrialGroup } = await import('../../runTrialGroup-MD2WFAYT.js');
5077
5077
  const fileForRerun = relative(process.cwd(), file);
5078
5078
  const rerunCommand = defaultTask ? options.trials !== void 0 ? `pome run -n ${k}` : "pome run" : `pome run ${fileForRerun && !fileForRerun.startsWith("..") ? fileForRerun : file} -n ${k}`;
5079
5079
  const groupResult = await runTrialGroup({
@@ -5167,7 +5167,7 @@ function createProgram() {
5167
5167
  process.exitCode = 5;
5168
5168
  return;
5169
5169
  }
5170
- const { runDemo } = await import('../../runDemo-F722I6AW.js');
5170
+ const { runDemo } = await import('../../runDemo-2QCREGAM.js');
5171
5171
  const result = await runDemo({
5172
5172
  apiBase: opts.apiUrl.replace(/\/$/, ""),
5173
5173
  dashboardBase: process.env.POME_DASHBOARD_URL ?? DEFAULT_DASHBOARD_URL,
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@pome-sh/cli",
3
- "version": "0.23.37",
3
+ "version": "0.23.39",
4
4
  "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.",
5
5
  "keywords": [
6
6
  "ai",