@pome-sh/cli 0.23.32 → 0.23.34

package/dist/build-info.json

@@ -1,6 +1,6 @@
 {
   "package": "pome-sh",
-  "version": "0.23.32",
-  "git_sha": "07c526747a7654a4aea3b1d8f417da3c9d653238",
-  "build_time": "2026-08-13T01:00:51.116Z"
+  "version": "0.23.34",
+  "git_sha": "0c6c21c02b475a025c5b83bc77f158832cd0dfe2",
+  "build_time": "2026-08-13T01:18:05.857Z"
 }

package/dist/{chunk-A2LGBYJG.js → chunk-ARAYJGDT.js}

@@ -2,14 +2,13 @@ import { readManifest, normalizeManifestTwins } from './chunk-ABM3CMQB.js';
 import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-ZTINFLSV.js';
 import { MOUNTED_TWINS, HostedAuthError, HostedDiscardRefusedError, HostedQuotaError, HostedOrchError, HostedTrialError, agentResponseSchema } from './chunk-X66JOOO7.js';
 import { redactSecrets, redactEvent } from './chunk-SG6ZTIMT.js';
-import { existsSync } from 'node:fs';
-import { rm, readFile, stat, mkdir, writeFile, chmod, mkdtemp, readdir } from 'node:fs/promises';
+import { randomUUID, createHash } from 'node:crypto';
+import { rm, readFile, stat, mkdir, writeFile, chmod, mkdtemp, readdir, rename } from 'node:fs/promises';
 import { join, dirname } from 'node:path';
 import { tmpdir, homedir } from 'node:os';
 import { execFile } from 'node:child_process';
 import { promisify } from 'node:util';
 import { createInterface } from 'node:readline';
-import { randomUUID, createHash } from 'node:crypto';
 
 // src/hosted/runSets.ts
 function groupRunSets(trials) {
@@ -62,12 +61,16 @@ var VALID_STATES = /* @__PURE__ */ new Set([
   "incomplete"
 ]);
 async function writeVerdictArtifact(runDir, verdict) {
-  await writeFile(
-    join(runDir, VERDICT_FILENAME),
-    `${JSON.stringify(verdict, null, 2)}
-`,
-    "utf8"
-  );
+  const tmpPath = join(runDir, `.${VERDICT_FILENAME}.${process.pid}.${randomUUID()}.tmp`);
+  try {
+    await writeFile(tmpPath, `${JSON.stringify(verdict, null, 2)}
+`, "utf8");
+    await rename(tmpPath, join(runDir, VERDICT_FILENAME));
+  } catch (err) {
+    await rm(tmpPath, { force: true }).catch(() => {
+    });
+    throw err;
+  }
 }
 function looksLikeVerdictArtifactBase(parsed) {
   if (typeof parsed !== "object" || parsed === null) return false;
@@ -100,13 +103,23 @@ function isVerdictArtifact(parsed) {
   if (typeof v.total !== "number") return false;
   return true;
 }
+var MISSING_ERROR_CODES = /* @__PURE__ */ new Set([
+  "ENOENT",
+  "ENOTDIR",
+  "ELOOP",
+  "ENAMETOOLONG"
+]);
+function isMissingFileError(err) {
+  const code = err?.code;
+  return typeof code === "string" && MISSING_ERROR_CODES.has(code);
+}
 async function readVerdictArtifactDetailed(runDir) {
   const path = join(runDir, VERDICT_FILENAME);
   let raw;
   try {
     raw = await readFile(path, "utf8");
-  } catch {
-    return { status: "unreadable" };
+  } catch (err) {
+    return isMissingFileError(err) ? { status: "missing" } : { status: "unreadable" };
   }
   let parsed;
   try {
@@ -143,9 +156,7 @@ 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);
-      }
+      else if (result.status === "unreadable") unreadableDirs.push(runDir);
     }
   }
   unreadableDirs.sort();
@@ -164,7 +175,7 @@ async function discoverRunSet(target) {
       unreadablePaths: []
     };
   }
-  if (anchorResult.status === "unreadable" && existsSync(join(target, VERDICT_FILENAME))) {
+  if (anchorResult.status === "unreadable") {
     return {
       kind: "trial-dir",
       set: null,
@@ -193,17 +204,6 @@ async function discoverRunSet(target) {
       unreadablePaths: unreadableDirs2
     };
   }
-  if (!existsSync(target)) {
-    return {
-      kind: "root",
-      set: null,
-      incompleteSet: null,
-      totalSets: 0,
-      staleVersionCount: 0,
-      unreadableCount: 0,
-      unreadablePaths: []
-    };
-  }
   const { trials, staleVersionDirs, unreadableDirs } = await scanVerdictArtifactsDetailed(target);
   const sets = groupRunSets(trials);
   const failedSet = latestFailedRunSet(sets);

package/dist/{runTrialGroup-WXDRCVUE.js → runTrialGroup-VT7UGDJU.js}

@@ -1,5 +1,5 @@
 import { newGroupId, criterionPhrase } from './chunk-RGZBC7NF.js';
-import { runTaskHosted, resolveRunAgentIdentity } from './chunk-A2LGBYJG.js';
+import { runTaskHosted, resolveRunAgentIdentity } from './chunk-ARAYJGDT.js';
 import './chunk-ABM3CMQB.js';
 import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-ZTINFLSV.js';
 import './chunk-NW7HGA2K.js';

package/dist/src/cli/main.js

@@ -1,5 +1,5 @@
 #!/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-A2LGBYJG.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-ARAYJGDT.js';
 import { readManifest, writeManifest, MANIFEST_JSON, readRequiredManifest, normalizeManifestTwins } from '../../chunk-ABM3CMQB.js';
 import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-GEC6MJV5.js';
 import '../../chunk-XDU6TD4O.js';
@@ -3064,7 +3064,22 @@ async function runChecksCommand(twinArg, opts) {
 // src/task/insertCriterion.ts
 var HEADING_RE = /^##\s+Success Criteria\s*$/;
 var NEXT_HEADING_RE = /^##\s+/;
-var CRITERION_RE = /^[-*]\s+\[(?:code|model)(?::[a-z][a-z0-9_-]*)?(?:\s+always-scored)?\]\s+.+$/;
+var CRITERION_RE = /^[-*]\s+\[(code|model)(?::([a-z][a-z0-9_-]*))?(\s+always-scored)?\]\s+(.+)$/;
+function identityKey(criterion) {
+  return `${criterion.kind}\0${criterion.twin ?? ""}\0${criterion.text}`;
+}
+function readCriterionLine(line) {
+  const match = line.trim().match(CRITERION_RE);
+  if (!match) return void 0;
+  return { kind: match[1], tag: match[2], text: match[4].trim() };
+}
+function primaryTwin(source) {
+  try {
+    return readConfigTwins(source)[0];
+  } catch {
+    return void 0;
+  }
+}
 var MissingCriteriaSectionError = class extends Error {
   constructor(path) {
     super(
@@ -3074,11 +3089,15 @@ var MissingCriteriaSectionError = class extends Error {
   }
 };
 var DuplicateCriterionError = class extends Error {
-  constructor(path, line) {
+  constructor(path, line, existing = line) {
+    const differs = existing.trim() !== line.trim();
     super(
       `${path} already has this criterion:
-  ${line}
-Adding it again would score it twice and inflate the task's denominator.`
+  ${existing.trim()}
+` + (differs ? `which is the same check as the one being added:
+  ${line.trim()}
+\u2014 a marker annotation or a twin tag does not make it a second check.
+` : "") + `Adding it again would score it twice and inflate the task's denominator.`
     );
     this.name = "DuplicateCriterionError";
   }
@@ -3094,13 +3113,33 @@ function insertCriterion(source, line, path = "this task file") {
       break;
     }
   }
+  const primary = primaryTwin(source);
+  const declared = /* @__PURE__ */ new Map();
+  for (const criterion of readCodeCriteria(source)) {
+    declared.set(
+      identityKey({ kind: "code", twin: criterion.twin, text: criterion.text }),
+      `- ${criterion.marker} ${criterion.text}`
+    );
+  }
   let insertAt = -1;
   for (let i = start + 1; i < end; i += 1) {
-    const existing = lines[i].trim();
-    if (!CRITERION_RE.test(existing)) continue;
-    if (existing === line.trim()) throw new DuplicateCriterionError(path, line);
+    const parsed = readCriterionLine(lines[i]);
+    if (!parsed) continue;
+    if (parsed.kind === "model") {
+      declared.set(
+        identityKey({ kind: "model", twin: parsed.tag ?? primary, text: parsed.text }),
+        lines[i].trim()
+      );
+    }
     insertAt = i + 1;
   }
+  const incoming = readCriterionLine(line);
+  if (incoming) {
+    const stored = declared.get(
+      identityKey({ kind: incoming.kind, twin: incoming.tag ?? primary, text: incoming.text })
+    );
+    if (stored !== void 0) throw new DuplicateCriterionError(path, line, stored);
+  }
   if (insertAt === -1) {
     insertAt = lines[start + 1]?.trim() === "" ? start + 2 : start + 1;
   }
@@ -4541,7 +4580,7 @@ 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 ("0.23.32".length > 0) return "0.23.32";
+  if ("0.23.34".length > 0) return "0.23.34";
   try {
     const here = dirname(fileURLToPath(import.meta.url));
     const candidates = [
@@ -5034,7 +5073,7 @@ function createProgram() {
               taskForRuns.config.runs
             );
             if (k > 1) {
-              const { runTrialGroup } = await import('../../runTrialGroup-WXDRCVUE.js');
+              const { runTrialGroup } = await import('../../runTrialGroup-VT7UGDJU.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({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/cli",
-  "version": "0.23.32",
+  "version": "0.23.34",
   "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",