@pome-sh/cli 0.23.9 → 0.23.10

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": "0.23.9",
4
- "git_sha": "67e57371305bb032008c357cd29fe817593b9fb8",
5
- "build_time": "2026-08-11T00:20:04.114Z"
3
+ "version": "0.23.10",
4
+ "git_sha": "8ec8b1602b470c35df8c83c71caacb309070ccac",
5
+ "build_time": "2026-08-11T11:30:52.650Z"
6
6
  }
@@ -183,7 +183,7 @@ async function checkTwinReachable(_configDir) {
183
183
  });
184
184
  let harness;
185
185
  try {
186
- const { bootTwin } = await import('./twinHarness-5ZQMXYHU.js');
186
+ const { bootTwin } = await import('./twinHarness-MA7NQHL2.js');
187
187
  harness = await bootTwin({
188
188
  twin: "github",
189
189
  seedState: void 0,
@@ -1,7 +1,7 @@
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-ZRDVM4QC.js';
4
- import { createRecorder, bootTwin } from './chunk-FLGJ5GBX.js';
3
+ import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-UR3HUJDI.js';
4
+ import { createRecorder, bootTwin } from './chunk-JT6MYL7P.js';
5
5
  import { eventSchema } from './chunk-VBATFCWR.js';
6
6
  import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
7
7
  import { serve } from '@hono/node-server';
@@ -172,7 +172,7 @@ var noMessagePosted = defineCheck({
172
172
  });
173
173
  var noReactionAdded = defineCheck({
174
174
  id: "slack.no-reaction-added",
175
- description: "Resolves the named channel, then filters the TOP-LEVEL reactions list by that channel's id and this emoji name, asserting no row matches. Reactions are not nested under their channel in the export, so this is a join the predicate performs itself. It asserts nothing about which message was reacted to, or by whom.",
175
+ description: "Resolves the named channel, then filters the TOP-LEVEL reactions list by that channel's id and this emoji name, asserting no row matches. Reactions are not nested under their channel in the export, so this is a join the predicate performs itself. It asserts nothing about which message was reacted to, or by whom. An export carrying no `reactions` collection at all is SKIPPED, because absent is not the same as none.",
176
176
  template: 'No "{reaction}" reaction was added in the "{channel}" channel',
177
177
  params: { reaction: emojiName, channel: channelName },
178
178
  substrate: "final",
@@ -208,14 +208,20 @@ var noReactionAdded = defineCheck({
208
208
  const found = resolveChannel(final, channel);
209
209
  if ("missing" in found)
210
210
  return missSkip(found);
211
- const hit = (final.reactions ?? []).some((row) => row.channel_id === found.found.id && row.name === reaction);
212
- const joined = [found.path];
213
- if (final.reactions !== void 0)
214
- joined.push(statePath("reactions"));
211
+ if (final.reactions == null)
212
+ return STATE_INCOMPLETE;
213
+ const hit = final.reactions.some((row) => row.channel_id === found.found.id && row.name === reaction);
215
214
  return {
216
215
  passed: !hit,
217
216
  reason: hit ? `reaction "${reaction}" found in channel "${channel}"` : `no reaction "${reaction}" in channel "${channel}"`,
218
- evidenceStatePaths: joined
217
+ // BOTH sides of the join, because this predicate really does read two
218
+ // places and a reader who opens only one cannot check its work: the
219
+ // channel row is where the id came from, the top-level reactions list is
220
+ // what was filtered. Reactions are not nested under their channel in the
221
+ // export — that is the whole reason this check performs a join — so one
222
+ // pointer cannot say it. The guard above already proved `reactions` is
223
+ // present, so both pointers always resolve.
224
+ evidenceStatePaths: [found.path, statePath("reactions")]
219
225
  };
220
226
  }
221
227
  });
@@ -1,5 +1,5 @@
1
1
  import { readManifest, normalizeManifestTwins } from './chunk-XOWIA7NR.js';
2
- import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-ZRDVM4QC.js';
2
+ import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-UR3HUJDI.js';
3
3
  import { MOUNTED_TWINS, HostedAuthError, HostedDiscardRefusedError, HostedQuotaError, HostedOrchError, HostedTrialError, agentResponseSchema } from './chunk-7AMIVWUB.js';
4
4
  import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
5
5
  import { existsSync } from 'node:fs';
@@ -1,4 +1,4 @@
1
- import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-IG76HVJS.js';
1
+ import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-VO3FOSAZ.js';
2
2
  import { createFileBackedRecorderStore, createRecorderStore } from './chunk-TV5S6WQV.js';
3
3
 
4
4
  // src/recorder/recorder.ts
@@ -2,7 +2,7 @@ import { criterionSchema, finalizeResponseSchema, HostedDiscardRefusedError, Hos
2
2
  import { seedSchema, parseSeed, defaultSeedState as defaultSeedState$2 } from './chunk-KIUIKAVU.js';
3
3
  import { gmailSeedSchema, defaultSeedState } from './chunk-NJ246QPJ.js';
4
4
  import { linearSeedSchema, defaultSeedState as defaultSeedState$1 } from './chunk-ZKID2HS3.js';
5
- import { isTwinName, TWIN_REGISTRY } from './chunk-IG76HVJS.js';
5
+ import { isTwinName, TWIN_REGISTRY } from './chunk-VO3FOSAZ.js';
6
6
  import { toTwinHttpEventRow } from './chunk-TV5S6WQV.js';
7
7
  import { redactEvent, redactSecrets } from './chunk-SG6ZTIMT.js';
8
8
  import { mkdir, appendFile, writeFile, readFile } from 'node:fs/promises';
@@ -59,9 +59,9 @@ var TWIN_REGISTRY = {
59
59
  envName: "SLACK",
60
60
  defaultPort: 3333,
61
61
  version: package_default2.version,
62
- defaultSeed: async () => (await import('./src-D3CNAM6U.js')).defaultSeedState(),
62
+ defaultSeed: async () => (await import('./src-ZRMSHGVJ.js')).defaultSeedState(),
63
63
  async boot({ seedState, runId, recorder }) {
64
- const { createSlackTwinApp, openSlackTwinDatabase, SlackDomain } = await import('./src-D3CNAM6U.js');
64
+ const { createSlackTwinApp, openSlackTwinDatabase, SlackDomain } = await import('./src-ZRMSHGVJ.js');
65
65
  const db = openSlackTwinDatabase(":memory:");
66
66
  const domain = new SlackDomain(db);
67
67
  domain.applySeed(seedState);
@@ -1,16 +1,16 @@
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-3SOI5EKJ.js';
3
+ import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-7I53HBHS.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-ZRDVM4QC.js';
6
+ import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-UR3HUJDI.js';
7
7
  import './chunk-NW7HGA2K.js';
8
8
  import { HostedQuotaError, HostedOrchError } from './chunk-7AMIVWUB.js';
9
9
  import './chunk-KIUIKAVU.js';
10
10
  import './chunk-NJ246QPJ.js';
11
11
  import './chunk-ZKID2HS3.js';
12
- import { bootTwin } from './chunk-FLGJ5GBX.js';
13
- import './chunk-IG76HVJS.js';
12
+ import { bootTwin } from './chunk-JT6MYL7P.js';
13
+ import './chunk-VO3FOSAZ.js';
14
14
  import './chunk-TV5S6WQV.js';
15
15
  import './chunk-VBATFCWR.js';
16
16
  import './chunk-SG6ZTIMT.js';
@@ -1,13 +1,13 @@
1
1
  import { newGroupId, criterionPhrase } from './chunk-RGZBC7NF.js';
2
- import { runTaskHosted, resolveRunAgentIdentity } from './chunk-GYKU53JU.js';
2
+ import { runTaskHosted, resolveRunAgentIdentity } from './chunk-IXWYWNXO.js';
3
3
  import './chunk-XOWIA7NR.js';
4
- import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-ZRDVM4QC.js';
4
+ import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-UR3HUJDI.js';
5
5
  import './chunk-NW7HGA2K.js';
6
6
  import { HostedQuotaError, HostedTrialError } from './chunk-7AMIVWUB.js';
7
7
  import './chunk-KIUIKAVU.js';
8
8
  import './chunk-NJ246QPJ.js';
9
9
  import './chunk-ZKID2HS3.js';
10
- import './chunk-IG76HVJS.js';
10
+ import './chunk-VO3FOSAZ.js';
11
11
  import './chunk-TV5S6WQV.js';
12
12
  import './chunk-VBATFCWR.js';
13
13
  import './chunk-SG6ZTIMT.js';
@@ -1,22 +1,22 @@
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, loadTrialEvents, persistCredentialsAfterLogin, DEFAULT_DOCS_SITE_ORIGIN, resolveSeams, resolveCachedAgentId, readLinkCache, postAgentResolver, writeLinkCache, ensurePomeGitignored } from '../../chunk-GYKU53JU.js';
2
+ import { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, clearLocalCredentials, friendlyHostedError, runSessionCreate, runSessionList, runSessionStop, resolveCredentials, runTaskHosted, discoverRunSet, VERDICT_ARTIFACT_VERSION, loadTrialEvents, persistCredentialsAfterLogin, DEFAULT_DOCS_SITE_ORIGIN, resolveSeams, resolveCachedAgentId, readLinkCache, postAgentResolver, writeLinkCache, ensurePomeGitignored } from '../../chunk-IXWYWNXO.js';
3
3
  import { readManifest, writeManifest, MANIFEST_JSON, readRequiredManifest, normalizeManifestTwins } from '../../chunk-XOWIA7NR.js';
4
- import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-3SOI5EKJ.js';
4
+ import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-7I53HBHS.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-ZRDVM4QC.js';
7
+ import { parseTaskFile, scoreStatus, runScoreLine, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, markerFor, criterionMarkerLabel, twinSkipSuffix, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-UR3HUJDI.js';
8
8
  import '../../chunk-NW7HGA2K.js';
9
9
  import { MOUNTED_TWINS, deriveAgentSlug, exitCodeFor, HostedUsageError, HostedOrchError, HostedAuthError, HostedQuotaError } from '../../chunk-7AMIVWUB.js';
10
10
  import { TAPE_ASSERTABLE_TOOLS } from '../../chunk-FKZZWWYC.js';
11
11
  import { seedSchema } from '../../chunk-KIUIKAVU.js';
12
- import { SLACK_CHECKS } from '../../chunk-6OT4IW25.js';
12
+ import { SLACK_CHECKS } from '../../chunk-IDNSKKEC.js';
13
13
  import { GMAIL_CHECKS } from '../../chunk-SE44FKWH.js';
14
14
  import '../../chunk-NJ246QPJ.js';
15
15
  import { LINEAR_CHECKS } from '../../chunk-XBT6ZLBG.js';
16
16
  import '../../chunk-ZKID2HS3.js';
17
17
  import { oneOf, defineCheck, repoRef, VACUITY_SENTINEL_NUMBER, childStatePath, VACUITY_SENTINEL, statePath, templateSlots, renderCheck, checksDigest, checkPattern, checkNearMissPattern } from '../../chunk-JWJYNAWI.js';
18
- import '../../chunk-FLGJ5GBX.js';
19
- import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-IG76HVJS.js';
18
+ import '../../chunk-JT6MYL7P.js';
19
+ import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-VO3FOSAZ.js';
20
20
  import '../../chunk-TV5S6WQV.js';
21
21
  import { isLegacyEventRow, eventSchema } from '../../chunk-VBATFCWR.js';
22
22
  import { redactEvent, redactSecrets } from '../../chunk-SG6ZTIMT.js';
@@ -4521,7 +4521,7 @@ var DEFAULT_AGENT_FILE = "examples/agents/scripted-triage-agent.ts";
4521
4521
  var DEFAULT_AGENT_COMMAND = `node ${DEFAULT_AGENT_FILE}`;
4522
4522
  var MANIFEST_SCHEMA_URL = "https://pome.sh/schemas/v1/pome.json";
4523
4523
  function readPackageVersion() {
4524
- if ("0.23.9".length > 0) return "0.23.9";
4524
+ if ("0.23.10".length > 0) return "0.23.10";
4525
4525
  try {
4526
4526
  const here = dirname(fileURLToPath(import.meta.url));
4527
4527
  const candidates = [
@@ -4976,7 +4976,7 @@ function createProgram() {
4976
4976
  return;
4977
4977
  }
4978
4978
  {
4979
- const { runDoctorChecks } = await import('../../checks-V76LVIJE.js');
4979
+ const { runDoctorChecks } = await import('../../checks-RGGGYUMP.js');
4980
4980
  const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
4981
4981
  const doctorReport = await runDoctorChecks({ mode: useLocal ? "full" : "hosted" });
4982
4982
  if (!doctorReport.ok) {
@@ -5014,7 +5014,7 @@ function createProgram() {
5014
5014
  taskForRuns.config.runs
5015
5015
  );
5016
5016
  if (k > 1) {
5017
- const { runTrialGroup } = await import('../../runTrialGroup-2NYQU3CM.js');
5017
+ const { runTrialGroup } = await import('../../runTrialGroup-NY33WGT2.js');
5018
5018
  const fileForRerun = relative(process.cwd(), file);
5019
5019
  const rerunCommand = defaultTask ? options.trials !== void 0 ? `pome run -n ${k}` : "pome run" : `pome run ${fileForRerun && !fileForRerun.startsWith("..") ? fileForRerun : file} -n ${k}`;
5020
5020
  const groupResult = await runTrialGroup({
@@ -5108,7 +5108,7 @@ function createProgram() {
5108
5108
  process.exitCode = 5;
5109
5109
  return;
5110
5110
  }
5111
- const { runDemo } = await import('../../runDemo-QWCVL772.js');
5111
+ const { runDemo } = await import('../../runDemo-SU5X6P27.js');
5112
5112
  const result = await runDemo({
5113
5113
  apiBase: opts.apiUrl.replace(/\/$/, ""),
5114
5114
  dashboardBase: process.env.POME_DASHBOARD_URL ?? DEFAULT_DASHBOARD_URL,
@@ -5126,7 +5126,7 @@ function createProgram() {
5126
5126
  program.command("doctor").description(
5127
5127
  "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."
5128
5128
  ).action(async () => {
5129
- const { runDoctorChecks } = await import('../../checks-V76LVIJE.js');
5129
+ const { runDoctorChecks } = await import('../../checks-RGGGYUMP.js');
5130
5130
  const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
5131
5131
  const report = await runDoctorChecks();
5132
5132
  for (const line of renderDoctorReport(report, { passNote: true })) console.error(line);
@@ -5290,7 +5290,7 @@ function createProgram() {
5290
5290
  ).description(
5291
5291
  "Start a standalone twin as a long-lived foreground server (Ctrl-C to stop)"
5292
5292
  ).action(async (name, options) => {
5293
- const { runTwinStartCommand } = await import('../../twinStart-DI4PUPQ2.js');
5293
+ const { runTwinStartCommand } = await import('../../twinStart-HKIRLN4U.js');
5294
5294
  await runTwinStartCommand(name, options);
5295
5295
  });
5296
5296
  twin.command("reset").argument("[name]", "Twin name (default: github)", "github").description("Reset standalone twin state").action(async (name) => {
@@ -1,4 +1,4 @@
1
- export { SLACK_CHECKS } from './chunk-6OT4IW25.js';
1
+ export { SLACK_CHECKS } from './chunk-IDNSKKEC.js';
2
2
  import './chunk-JWJYNAWI.js';
3
3
  import { routeInputDeclarer, booleanInput, integerInput, mountDeclaredRoute, UndeclaredInputError } from './chunk-2Q3P45LK.js';
4
4
  import { loadMcpToolFixture, defineTwin, queryTokenResolver, formTokenResolver, deriveMcpToolTable, UnknownToolError, openTwinDatabase, createApp } from './chunk-TV5S6WQV.js';
@@ -0,0 +1,5 @@
1
+ export { UnsupportedTwinError, bootTwin } from './chunk-JT6MYL7P.js';
2
+ export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-VO3FOSAZ.js';
3
+ import './chunk-TV5S6WQV.js';
4
+ import './chunk-VBATFCWR.js';
5
+ import './chunk-SG6ZTIMT.js';
@@ -1,5 +1,5 @@
1
- import { bootTwin } from './chunk-FLGJ5GBX.js';
2
- import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-IG76HVJS.js';
1
+ import { bootTwin } from './chunk-JT6MYL7P.js';
2
+ import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-VO3FOSAZ.js';
3
3
  import './chunk-TV5S6WQV.js';
4
4
  import './chunk-VBATFCWR.js';
5
5
  import './chunk-SG6ZTIMT.js';
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@pome-sh/cli",
3
- "version": "0.23.9",
3
+ "version": "0.23.10",
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",
@@ -1,5 +0,0 @@
1
- export { UnsupportedTwinError, bootTwin } from './chunk-FLGJ5GBX.js';
2
- export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-IG76HVJS.js';
3
- import './chunk-TV5S6WQV.js';
4
- import './chunk-VBATFCWR.js';
5
- import './chunk-SG6ZTIMT.js';