@pome-sh/cli 0.23.9 → 0.23.11

package/dist/build-info.json

@@ -1,6 +1,6 @@
 {
   "package": "pome-sh",
-  "version": "0.23.9",
-  "git_sha": "67e57371305bb032008c357cd29fe817593b9fb8",
-  "build_time": "2026-08-11T00:20:04.114Z"
+  "version": "0.23.11",
+  "git_sha": "0dbb84dc7726b94d8e6a3d1199f64ac583166990",
+  "build_time": "2026-08-11T13:17:29.925Z"
 }

package/dist/{checks-V76LVIJE.js → checks-RGGGYUMP.js}

@@ -183,7 +183,7 @@ async function checkTwinReachable(_configDir) {
   });
   let harness;
   try {
-    const { bootTwin } = await import('./twinHarness-5ZQMXYHU.js');
+    const { bootTwin } = await import('./twinHarness-MA7NQHL2.js');
     harness = await bootTwin({
       twin: "github",
       seedState: void 0,

package/dist/{chunk-3SOI5EKJ.js → chunk-7I53HBHS.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-ZRDVM4QC.js';
-import { createRecorder, bootTwin } from './chunk-FLGJ5GBX.js';
+import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-UR3HUJDI.js';
+import { createRecorder, bootTwin } from './chunk-JT6MYL7P.js';
 import { eventSchema } from './chunk-VBATFCWR.js';
 import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
 import { serve } from '@hono/node-server';

package/dist/{chunk-GYKU53JU.js → chunk-FNLI5YX3.js}

@@ -1,9 +1,9 @@
 import { readManifest, normalizeManifestTwins } from './chunk-XOWIA7NR.js';
-import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-ZRDVM4QC.js';
+import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-UR3HUJDI.js';
 import { MOUNTED_TWINS, HostedAuthError, HostedDiscardRefusedError, HostedQuotaError, HostedOrchError, HostedTrialError, agentResponseSchema } from './chunk-7AMIVWUB.js';
 import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
 import { existsSync } from 'node:fs';
-import { readFile, rm, stat, mkdir, writeFile, chmod, mkdtemp, readdir } from 'node:fs/promises';
+import { rm, readFile, stat, mkdir, writeFile, chmod, mkdtemp, readdir } from 'node:fs/promises';
 import { join, dirname } from 'node:path';
 import { tmpdir, homedir } from 'node:os';
 import { execFile } from 'node:child_process';
@@ -11,6 +11,49 @@ import { promisify } from 'node:util';
 import { createInterface } from 'node:readline';
 import { randomUUID, createHash } from 'node:crypto';
 
+// src/hosted/runSets.ts
+function groupRunSets(trials) {
+  const byKey = /* @__PURE__ */ new Map();
+  for (const trial of trials) {
+    const key = trial.verdict.group_id ?? `solo:${trial.verdict.session_id}`;
+    const bucket = byKey.get(key);
+    if (bucket) bucket.push(trial);
+    else byKey.set(key, [trial]);
+  }
+  const sets = [];
+  for (const bucket of byKey.values()) {
+    bucket.sort(
+      (a, b) => a.verdict.finalized_at.localeCompare(b.verdict.finalized_at)
+    );
+    const last = bucket[bucket.length - 1];
+    const hasFailed = bucket.some((t) => t.verdict.state === "fail");
+    const allPassed = bucket.every((t) => t.verdict.state === "pass");
+    sets.push({
+      groupId: bucket[0].verdict.group_id,
+      taskName: bucket[0].verdict.task_name,
+      taskPath: bucket[0].verdict.task_path,
+      trials: bucket,
+      latestFinalizedAt: last.verdict.finalized_at,
+      outcome: hasFailed ? "fail" : allPassed ? "pass" : "incomplete"
+    });
+  }
+  sets.sort((a, b) => a.latestFinalizedAt.localeCompare(b.latestFinalizedAt));
+  return sets;
+}
+function latestFailedRunSet(sets) {
+  for (let i = sets.length - 1; i >= 0; i -= 1) {
+    if (sets[i].outcome === "fail") return sets[i];
+  }
+  return null;
+}
+function latestIncompleteRunSet(sets) {
+  for (let i = sets.length - 1; i >= 0; i -= 1) {
+    if (sets[i].outcome === "incomplete") return sets[i];
+  }
+  return null;
+}
+
+// src/hosted/evalResultCache.ts
 var VERDICT_ARTIFACT_VERSION = 2;
 var VERDICT_FILENAME = "verdict.json";
 var VALID_STATES = /* @__PURE__ */ new Set([
@@ -80,11 +123,12 @@ async function readVerdictArtifactDetailed(runDir) {
 async function scanVerdictArtifactsDetailed(artifactsRoot) {
   const trials = [];
   const staleVersionDirs = [];
+  const unreadableDirs = [];
   let slugs;
   try {
     slugs = await readdir(artifactsRoot);
   } catch {
-    return { trials, staleVersionDirs };
+    return { trials, staleVersionDirs, unreadableDirs };
   }
   for (const slug of slugs) {
     const slugDir = join(artifactsRoot, slug);
@@ -99,49 +143,13 @@ async function scanVerdictArtifactsDetailed(artifactsRoot) {
       const result = await readVerdictArtifactDetailed(runDir);
       if (result.status === "ok") trials.push(result.trial);
       else if (result.status === "stale-version") staleVersionDirs.push(runDir);
+      else if (result.status === "unreadable" && existsSync(join(runDir, VERDICT_FILENAME))) {
+        unreadableDirs.push(runDir);
+      }
     }
   }
-  return { trials, staleVersionDirs };
-}
-function groupRunSets(trials) {
-  const byKey = /* @__PURE__ */ new Map();
-  for (const trial of trials) {
-    const key = trial.verdict.group_id ?? `solo:${trial.verdict.session_id}`;
-    const bucket = byKey.get(key);
-    if (bucket) bucket.push(trial);
-    else byKey.set(key, [trial]);
-  }
-  const sets = [];
-  for (const bucket of byKey.values()) {
-    bucket.sort(
-      (a, b) => a.verdict.finalized_at.localeCompare(b.verdict.finalized_at)
-    );
-    const last = bucket[bucket.length - 1];
-    const hasFailed = bucket.some((t) => t.verdict.state === "fail");
-    const allPassed = bucket.every((t) => t.verdict.state === "pass");
-    sets.push({
-      groupId: bucket[0].verdict.group_id,
-      taskName: bucket[0].verdict.task_name,
-      taskPath: bucket[0].verdict.task_path,
-      trials: bucket,
-      latestFinalizedAt: last.verdict.finalized_at,
-      outcome: hasFailed ? "fail" : allPassed ? "pass" : "incomplete"
-    });
-  }
-  sets.sort((a, b) => a.latestFinalizedAt.localeCompare(b.latestFinalizedAt));
-  return sets;
-}
-function latestFailedRunSet(sets) {
-  for (let i = sets.length - 1; i >= 0; i -= 1) {
-    if (sets[i].outcome === "fail") return sets[i];
-  }
-  return null;
-}
-function latestIncompleteRunSet(sets) {
-  for (let i = sets.length - 1; i >= 0; i -= 1) {
-    if (sets[i].outcome === "incomplete") return sets[i];
-  }
-  return null;
+  unreadableDirs.sort();
+  return { trials, staleVersionDirs, unreadableDirs };
 }
 async function discoverRunSet(target) {
   const anchorResult = await readVerdictArtifactDetailed(target);
@@ -151,13 +159,26 @@ async function discoverRunSet(target) {
       set: null,
       incompleteSet: null,
       totalSets: 0,
-      staleVersionCount: 1
+      staleVersionCount: 1,
+      unreadableCount: 0,
+      unreadablePaths: []
+    };
+  }
+  if (anchorResult.status === "unreadable" && existsSync(join(target, VERDICT_FILENAME))) {
+    return {
+      kind: "trial-dir",
+      set: null,
+      incompleteSet: null,
+      totalSets: 0,
+      staleVersionCount: 0,
+      unreadableCount: 1,
+      unreadablePaths: [target]
     };
   }
   if (anchorResult.status === "ok") {
     const anchor = anchorResult.trial;
     const root = join(target, "..", "..");
-    const { trials: trials2, staleVersionDirs: staleVersionDirs2 } = await scanVerdictArtifactsDetailed(root);
+    const { trials: trials2, staleVersionDirs: staleVersionDirs2, unreadableDirs: unreadableDirs2 } = await scanVerdictArtifactsDetailed(root);
     const sets2 = groupRunSets(trials2);
     const own = sets2.find(
       (s) => anchor.verdict.group_id !== null && s.groupId === anchor.verdict.group_id || anchor.verdict.group_id === null && s.trials.length === 1 && s.trials[0].verdict.session_id === anchor.verdict.session_id
@@ -167,7 +188,9 @@ async function discoverRunSet(target) {
       set: own,
       incompleteSet: null,
       totalSets: Math.max(sets2.length, 1),
-      staleVersionCount: staleVersionDirs2.length
+      staleVersionCount: staleVersionDirs2.length,
+      unreadableCount: unreadableDirs2.length,
+      unreadablePaths: unreadableDirs2
     };
   }
   if (!existsSync(target)) {
@@ -176,10 +199,12 @@ async function discoverRunSet(target) {
       set: null,
       incompleteSet: null,
       totalSets: 0,
-      staleVersionCount: 0
+      staleVersionCount: 0,
+      unreadableCount: 0,
+      unreadablePaths: []
     };
   }
-  const { trials, staleVersionDirs } = await scanVerdictArtifactsDetailed(target);
+  const { trials, staleVersionDirs, unreadableDirs } = await scanVerdictArtifactsDetailed(target);
   const sets = groupRunSets(trials);
   const failedSet = latestFailedRunSet(sets);
   return {
@@ -187,28 +212,11 @@ async function discoverRunSet(target) {
     set: failedSet,
     incompleteSet: failedSet ? null : latestIncompleteRunSet(sets),
     totalSets: sets.length,
-    staleVersionCount: staleVersionDirs.length
+    staleVersionCount: staleVersionDirs.length,
+    unreadableCount: unreadableDirs.length,
+    unreadablePaths: unreadableDirs
   };
 }
-async function loadTrialEvents(runDir) {
-  let raw;
-  try {
-    raw = await readFile(join(runDir, "events.jsonl"), "utf8");
-  } catch {
-    return [];
-  }
-  const events = [];
-  for (const line of raw.split("\n")) {
-    const trimmed = line.trim();
-    if (!trimmed) continue;
-    try {
-      const parsed = JSON.parse(trimmed);
-      if (typeof parsed === "object" && parsed !== null) events.push(parsed);
-    } catch {
-    }
-  }
-  return events;
-}
 var execFileAsync = promisify(execFile);
 var SERVICE = "sh.pome.cli";
 var ACCOUNT = "hosted";
@@ -1239,4 +1247,4 @@ async function abandonBestEffort(client, sessionId, errorCode) {
   }
 }
 
-export { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, DEFAULT_DOCS_SITE_ORIGIN, VERDICT_ARTIFACT_VERSION, clearLocalCredentials, discoverRunSet, ensurePomeGitignored, friendlyHostedError, loadTrialEvents, persistCredentialsAfterLogin, postAgentResolver, readLinkCache, resolveCachedAgentId, resolveCredentials, resolveRunAgentIdentity, resolveSeams, runSessionCreate, runSessionList, runSessionStop, runTaskHosted, writeLinkCache };
+export { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, DEFAULT_DOCS_SITE_ORIGIN, VERDICT_ARTIFACT_VERSION, clearLocalCredentials, discoverRunSet, ensurePomeGitignored, friendlyHostedError, persistCredentialsAfterLogin, postAgentResolver, readLinkCache, resolveCachedAgentId, resolveCredentials, resolveRunAgentIdentity, resolveSeams, runSessionCreate, runSessionList, runSessionStop, runTaskHosted, writeLinkCache };

package/dist/{chunk-6OT4IW25.js → chunk-IDNSKKEC.js}

@@ -172,7 +172,7 @@ var noMessagePosted = defineCheck({
 });
 var noReactionAdded = defineCheck({
   id: "slack.no-reaction-added",
-  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.",
+  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.",
   template: 'No "{reaction}" reaction was added in the "{channel}" channel',
   params: { reaction: emojiName, channel: channelName },
   substrate: "final",
@@ -208,14 +208,20 @@ var noReactionAdded = defineCheck({
     const found = resolveChannel(final, channel);
     if ("missing" in found)
       return missSkip(found);
-    const hit = (final.reactions ?? []).some((row) => row.channel_id === found.found.id && row.name === reaction);
-    const joined = [found.path];
-    if (final.reactions !== void 0)
-      joined.push(statePath("reactions"));
+    if (final.reactions == null)
+      return STATE_INCOMPLETE;
+    const hit = final.reactions.some((row) => row.channel_id === found.found.id && row.name === reaction);
     return {
       passed: !hit,
       reason: hit ? `reaction "${reaction}" found in channel "${channel}"` : `no reaction "${reaction}" in channel "${channel}"`,
-      evidenceStatePaths: joined
+      // BOTH sides of the join, because this predicate really does read two
+      // places and a reader who opens only one cannot check its work: the
+      // channel row is where the id came from, the top-level reactions list is
+      // what was filtered. Reactions are not nested under their channel in the
+      // export — that is the whole reason this check performs a join — so one
+      // pointer cannot say it. The guard above already proved `reactions` is
+      // present, so both pointers always resolve.
+      evidenceStatePaths: [found.path, statePath("reactions")]
     };
   }
 });

package/dist/{chunk-FLGJ5GBX.js → chunk-JT6MYL7P.js}

@@ -1,4 +1,4 @@
-import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-IG76HVJS.js';
+import { TWIN_NAMES, isTwinName, TWIN_REGISTRY } from './chunk-VO3FOSAZ.js';
 import { createFileBackedRecorderStore, createRecorderStore } from './chunk-TV5S6WQV.js';
 
 // src/recorder/recorder.ts

package/dist/{chunk-ZRDVM4QC.js → chunk-UR3HUJDI.js}

@@ -2,7 +2,7 @@ import { criterionSchema, finalizeResponseSchema, HostedDiscardRefusedError, Hos
 import { seedSchema, parseSeed, defaultSeedState as defaultSeedState$2 } from './chunk-KIUIKAVU.js';
 import { gmailSeedSchema, defaultSeedState } from './chunk-NJ246QPJ.js';
 import { linearSeedSchema, defaultSeedState as defaultSeedState$1 } from './chunk-ZKID2HS3.js';
-import { isTwinName, TWIN_REGISTRY } from './chunk-IG76HVJS.js';
+import { isTwinName, TWIN_REGISTRY } from './chunk-VO3FOSAZ.js';
 import { toTwinHttpEventRow } from './chunk-TV5S6WQV.js';
 import { redactEvent, redactSecrets } from './chunk-SG6ZTIMT.js';
 import { mkdir, appendFile, writeFile, readFile } from 'node:fs/promises';

package/dist/{chunk-IG76HVJS.js → chunk-VO3FOSAZ.js}

@@ -59,9 +59,9 @@ var TWIN_REGISTRY = {
     envName: "SLACK",
     defaultPort: 3333,
     version: package_default2.version,
-    defaultSeed: async () => (await import('./src-D3CNAM6U.js')).defaultSeedState(),
+    defaultSeed: async () => (await import('./src-ZRMSHGVJ.js')).defaultSeedState(),
     async boot({ seedState, runId, recorder }) {
-      const { createSlackTwinApp, openSlackTwinDatabase, SlackDomain } = await import('./src-D3CNAM6U.js');
+      const { createSlackTwinApp, openSlackTwinDatabase, SlackDomain } = await import('./src-ZRMSHGVJ.js');
       const db = openSlackTwinDatabase(":memory:");
       const domain = new SlackDomain(db);
       domain.applySeed(seedState);

package/dist/{runDemo-QWCVL772.js → runDemo-SU5X6P27.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-3SOI5EKJ.js';
+import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-7I53HBHS.js';
 import { getAvailablePort } from './chunk-XDU6TD4O.js';
 import './chunk-CBFKZZBR.js';
-import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-ZRDVM4QC.js';
+import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, outcomeOf } from './chunk-UR3HUJDI.js';
 import './chunk-NW7HGA2K.js';
 import { HostedQuotaError, HostedOrchError } from './chunk-7AMIVWUB.js';
 import './chunk-KIUIKAVU.js';
 import './chunk-NJ246QPJ.js';
 import './chunk-ZKID2HS3.js';
-import { bootTwin } from './chunk-FLGJ5GBX.js';
-import './chunk-IG76HVJS.js';
+import { bootTwin } from './chunk-JT6MYL7P.js';
+import './chunk-VO3FOSAZ.js';
 import './chunk-TV5S6WQV.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/dist/{runTrialGroup-2NYQU3CM.js → runTrialGroup-VR2RS2RH.js}

@@ -1,13 +1,13 @@
 import { newGroupId, criterionPhrase } from './chunk-RGZBC7NF.js';
-import { runTaskHosted, resolveRunAgentIdentity } from './chunk-GYKU53JU.js';
+import { runTaskHosted, resolveRunAgentIdentity } from './chunk-FNLI5YX3.js';
 import './chunk-XOWIA7NR.js';
-import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-ZRDVM4QC.js';
+import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-UR3HUJDI.js';
 import './chunk-NW7HGA2K.js';
 import { HostedQuotaError, HostedTrialError } from './chunk-7AMIVWUB.js';
 import './chunk-KIUIKAVU.js';
 import './chunk-NJ246QPJ.js';
 import './chunk-ZKID2HS3.js';
-import './chunk-IG76HVJS.js';
+import './chunk-VO3FOSAZ.js';
 import './chunk-TV5S6WQV.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/dist/src/cli/main.js

@@ -1,22 +1,22 @@
 #!/usr/bin/env node
-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';
+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-FNLI5YX3.js';
 import { readManifest, writeManifest, MANIFEST_JSON, readRequiredManifest, normalizeManifestTwins } from '../../chunk-XOWIA7NR.js';
-import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-3SOI5EKJ.js';
+import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-7I53HBHS.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-ZRDVM4QC.js';
+import { parseTaskFile, scoreStatus, runScoreLine, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, markerFor, criterionMarkerLabel, twinSkipSuffix, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-UR3HUJDI.js';
 import '../../chunk-NW7HGA2K.js';
 import { MOUNTED_TWINS, deriveAgentSlug, exitCodeFor, HostedUsageError, HostedOrchError, HostedAuthError, HostedQuotaError } from '../../chunk-7AMIVWUB.js';
 import { TAPE_ASSERTABLE_TOOLS } from '../../chunk-FKZZWWYC.js';
 import { seedSchema } from '../../chunk-KIUIKAVU.js';
-import { SLACK_CHECKS } from '../../chunk-6OT4IW25.js';
+import { SLACK_CHECKS } from '../../chunk-IDNSKKEC.js';
 import { GMAIL_CHECKS } from '../../chunk-SE44FKWH.js';
 import '../../chunk-NJ246QPJ.js';
 import { LINEAR_CHECKS } from '../../chunk-XBT6ZLBG.js';
 import '../../chunk-ZKID2HS3.js';
 import { oneOf, defineCheck, repoRef, VACUITY_SENTINEL_NUMBER, childStatePath, VACUITY_SENTINEL, statePath, templateSlots, renderCheck, checksDigest, checkPattern, checkNearMissPattern } from '../../chunk-JWJYNAWI.js';
-import '../../chunk-FLGJ5GBX.js';
-import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-IG76HVJS.js';
+import '../../chunk-JT6MYL7P.js';
+import { TWIN_NAME_LIST, isTwinName, createGitHubSmokeApp } from '../../chunk-VO3FOSAZ.js';
 import '../../chunk-TV5S6WQV.js';
 import { isLegacyEventRow, eventSchema } from '../../chunk-VBATFCWR.js';
 import { redactEvent, redactSecrets } from '../../chunk-SG6ZTIMT.js';
@@ -4513,6 +4513,25 @@ function buildGroupFixPrompt(ctx) {
 
 ${buildGroupFixUserPrompt(ctx)}`;
 }
+async function loadTrialEvents(runDir) {
+  let raw;
+  try {
+    raw = await readFile(join(runDir, "events.jsonl"), "utf8");
+  } catch {
+    return [];
+  }
+  const events = [];
+  for (const line of raw.split("\n")) {
+    const trimmed = line.trim();
+    if (!trimmed) continue;
+    try {
+      const parsed = JSON.parse(trimmed);
+      if (typeof parsed === "object" && parsed !== null) events.push(parsed);
+    } catch {
+    }
+  }
+  return events;
+}
 
 // src/cli/main.ts
 var PACKAGE_VERSION = readPackageVersion();
@@ -4520,8 +4539,9 @@ var SESSION_CREATE_FORMATS = /* @__PURE__ */ new Set(["text", "json", "env"]);
 var DEFAULT_AGENT_FILE = "examples/agents/scripted-triage-agent.ts";
 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 ("0.23.9".length > 0) return "0.23.9";
+  if ("0.23.11".length > 0) return "0.23.11";
   try {
     const here = dirname(fileURLToPath(import.meta.url));
     const candidates = [
@@ -4976,7 +4996,7 @@ function createProgram() {
         return;
       }
       {
-        const { runDoctorChecks } = await import('../../checks-V76LVIJE.js');
+        const { runDoctorChecks } = await import('../../checks-RGGGYUMP.js');
         const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
         const doctorReport = await runDoctorChecks({ mode: useLocal ? "full" : "hosted" });
         if (!doctorReport.ok) {
@@ -5014,7 +5034,7 @@ function createProgram() {
               taskForRuns.config.runs
             );
             if (k > 1) {
-              const { runTrialGroup } = await import('../../runTrialGroup-2NYQU3CM.js');
+              const { runTrialGroup } = await import('../../runTrialGroup-VR2RS2RH.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 ${k}`;
               const groupResult = await runTrialGroup({
@@ -5108,7 +5128,7 @@ function createProgram() {
         process.exitCode = 5;
         return;
       }
-      const { runDemo } = await import('../../runDemo-QWCVL772.js');
+      const { runDemo } = await import('../../runDemo-SU5X6P27.js');
       const result = await runDemo({
         apiBase: opts.apiUrl.replace(/\/$/, ""),
         dashboardBase: process.env.POME_DASHBOARD_URL ?? DEFAULT_DASHBOARD_URL,
@@ -5126,7 +5146,7 @@ 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-V76LVIJE.js');
+    const { runDoctorChecks } = await import('../../checks-RGGGYUMP.js');
     const { renderDoctorReport } = await import('../../render-ZQQ4UMNO.js');
     const report = await runDoctorChecks();
     for (const line of renderDoctorReport(report, { passNote: true })) console.error(line);
@@ -5218,8 +5238,20 @@ function createProgram() {
         `${discovery.staleVersionCount} verdict.json file(s) under ${root} are not artifact version ${VERDICT_ARTIFACT_VERSION} (the only version this CLI reads) and were skipped \u2014 re-run \`pome run\` to record those trials again.`
       );
     }
+    if (discovery.unreadableCount > 0) {
+      console.error(
+        `${discovery.unreadableCount} verdict.json file(s) under ${root} could not be read (truncated, hand-edited, or not a verdict artifact) and were skipped:`
+      );
+      for (const path of discovery.unreadablePaths.slice(0, MAX_UNREADABLE_PATHS_SHOWN)) {
+        console.error(`  - ${path}`);
+      }
+      const omitted = discovery.unreadablePaths.length - MAX_UNREADABLE_PATHS_SHOWN;
+      if (omitted > 0) {
+        console.error(`  (${omitted} more omitted \u2014 kept first ${MAX_UNREADABLE_PATHS_SHOWN})`);
+      }
+    }
     if (discovery.totalSets === 0) {
-      if (discovery.staleVersionCount === 0) {
+      if (discovery.staleVersionCount === 0 && discovery.unreadableCount === 0) {
         console.error(
           `No finalized run sets under ${root} \u2014 hosted \`pome run\` records a verdict.json per trial; run one first (or point fix-prompt at your artifacts dir).`
         );
@@ -5290,7 +5322,7 @@ 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-DI4PUPQ2.js');
+    const { runTwinStartCommand } = await import('../../twinStart-HKIRLN4U.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-D3CNAM6U.js → src-ZRMSHGVJ.js}

@@ -1,4 +1,4 @@
-export { SLACK_CHECKS } from './chunk-6OT4IW25.js';
+export { SLACK_CHECKS } from './chunk-IDNSKKEC.js';
 import './chunk-JWJYNAWI.js';
 import { routeInputDeclarer, booleanInput, integerInput, mountDeclaredRoute, UndeclaredInputError } from './chunk-2Q3P45LK.js';
 import { loadMcpToolFixture, defineTwin, queryTokenResolver, formTokenResolver, deriveMcpToolTable, UnknownToolError, openTwinDatabase, createApp } from './chunk-TV5S6WQV.js';

package/dist/twinHarness-MA7NQHL2.js

@@ -0,0 +1,5 @@
+export { UnsupportedTwinError, bootTwin } from './chunk-JT6MYL7P.js';
+export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-VO3FOSAZ.js';
+import './chunk-TV5S6WQV.js';
+import './chunk-VBATFCWR.js';
+import './chunk-SG6ZTIMT.js';

package/dist/{twinStart-DI4PUPQ2.js → twinStart-HKIRLN4U.js}

@@ -1,5 +1,5 @@
-import { bootTwin } from './chunk-FLGJ5GBX.js';
-import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-IG76HVJS.js';
+import { bootTwin } from './chunk-JT6MYL7P.js';
+import { isTwinName, TWIN_NAMES, defaultPortFor, TWIN_REGISTRY } from './chunk-VO3FOSAZ.js';
 import './chunk-TV5S6WQV.js';
 import './chunk-VBATFCWR.js';
 import './chunk-SG6ZTIMT.js';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/cli",
-  "version": "0.23.9",
+  "version": "0.23.11",
   "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",

package/dist/twinHarness-5ZQMXYHU.js

@@ -1,5 +0,0 @@
-export { UnsupportedTwinError, bootTwin } from './chunk-FLGJ5GBX.js';
-export { STRIPE_LOCAL_ACCOUNT_ID } from './chunk-IG76HVJS.js';
-import './chunk-TV5S6WQV.js';
-import './chunk-VBATFCWR.js';
-import './chunk-SG6ZTIMT.js';