@pome-sh/cli 0.23.10 → 0.23.11

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": "0.23.10",
4
- "git_sha": "8ec8b1602b470c35df8c83c71caacb309070ccac",
5
- "build_time": "2026-08-11T11:30:52.650Z"
3
+ "version": "0.23.11",
4
+ "git_sha": "0dbb84dc7726b94d8e6a3d1199f64ac583166990",
5
+ "build_time": "2026-08-11T13:17:29.925Z"
6
6
  }
@@ -3,7 +3,7 @@ import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentComm
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';
6
- import { readFile, rm, stat, mkdir, writeFile, chmod, mkdtemp, readdir } from 'node:fs/promises';
6
+ import { rm, readFile, stat, mkdir, writeFile, chmod, mkdtemp, readdir } from 'node:fs/promises';
7
7
  import { join, dirname } from 'node:path';
8
8
  import { tmpdir, homedir } from 'node:os';
9
9
  import { execFile } from 'node:child_process';
@@ -11,6 +11,49 @@ import { promisify } from 'node:util';
11
11
  import { createInterface } from 'node:readline';
12
12
  import { randomUUID, createHash } from 'node:crypto';
13
13
 
14
+ // src/hosted/runSets.ts
15
+ function groupRunSets(trials) {
16
+ const byKey = /* @__PURE__ */ new Map();
17
+ for (const trial of trials) {
18
+ const key = trial.verdict.group_id ?? `solo:${trial.verdict.session_id}`;
19
+ const bucket = byKey.get(key);
20
+ if (bucket) bucket.push(trial);
21
+ else byKey.set(key, [trial]);
22
+ }
23
+ const sets = [];
24
+ for (const bucket of byKey.values()) {
25
+ bucket.sort(
26
+ (a, b) => a.verdict.finalized_at.localeCompare(b.verdict.finalized_at)
27
+ );
28
+ const last = bucket[bucket.length - 1];
29
+ const hasFailed = bucket.some((t) => t.verdict.state === "fail");
30
+ const allPassed = bucket.every((t) => t.verdict.state === "pass");
31
+ sets.push({
32
+ groupId: bucket[0].verdict.group_id,
33
+ taskName: bucket[0].verdict.task_name,
34
+ taskPath: bucket[0].verdict.task_path,
35
+ trials: bucket,
36
+ latestFinalizedAt: last.verdict.finalized_at,
37
+ outcome: hasFailed ? "fail" : allPassed ? "pass" : "incomplete"
38
+ });
39
+ }
40
+ sets.sort((a, b) => a.latestFinalizedAt.localeCompare(b.latestFinalizedAt));
41
+ return sets;
42
+ }
43
+ function latestFailedRunSet(sets) {
44
+ for (let i = sets.length - 1; i >= 0; i -= 1) {
45
+ if (sets[i].outcome === "fail") return sets[i];
46
+ }
47
+ return null;
48
+ }
49
+ function latestIncompleteRunSet(sets) {
50
+ for (let i = sets.length - 1; i >= 0; i -= 1) {
51
+ if (sets[i].outcome === "incomplete") return sets[i];
52
+ }
53
+ return null;
54
+ }
55
+
56
+ // src/hosted/evalResultCache.ts
14
57
  var VERDICT_ARTIFACT_VERSION = 2;
15
58
  var VERDICT_FILENAME = "verdict.json";
16
59
  var VALID_STATES = /* @__PURE__ */ new Set([
@@ -80,11 +123,12 @@ async function readVerdictArtifactDetailed(runDir) {
80
123
  async function scanVerdictArtifactsDetailed(artifactsRoot) {
81
124
  const trials = [];
82
125
  const staleVersionDirs = [];
126
+ const unreadableDirs = [];
83
127
  let slugs;
84
128
  try {
85
129
  slugs = await readdir(artifactsRoot);
86
130
  } catch {
87
- return { trials, staleVersionDirs };
131
+ return { trials, staleVersionDirs, unreadableDirs };
88
132
  }
89
133
  for (const slug of slugs) {
90
134
  const slugDir = join(artifactsRoot, slug);
@@ -99,49 +143,13 @@ async function scanVerdictArtifactsDetailed(artifactsRoot) {
99
143
  const result = await readVerdictArtifactDetailed(runDir);
100
144
  if (result.status === "ok") trials.push(result.trial);
101
145
  else if (result.status === "stale-version") staleVersionDirs.push(runDir);
146
+ else if (result.status === "unreadable" && existsSync(join(runDir, VERDICT_FILENAME))) {
147
+ unreadableDirs.push(runDir);
148
+ }
102
149
  }
103
150
  }
104
- return { trials, staleVersionDirs };
105
- }
106
- function groupRunSets(trials) {
107
- const byKey = /* @__PURE__ */ new Map();
108
- for (const trial of trials) {
109
- const key = trial.verdict.group_id ?? `solo:${trial.verdict.session_id}`;
110
- const bucket = byKey.get(key);
111
- if (bucket) bucket.push(trial);
112
- else byKey.set(key, [trial]);
113
- }
114
- const sets = [];
115
- for (const bucket of byKey.values()) {
116
- bucket.sort(
117
- (a, b) => a.verdict.finalized_at.localeCompare(b.verdict.finalized_at)
118
- );
119
- const last = bucket[bucket.length - 1];
120
- const hasFailed = bucket.some((t) => t.verdict.state === "fail");
121
- const allPassed = bucket.every((t) => t.verdict.state === "pass");
122
- sets.push({
123
- groupId: bucket[0].verdict.group_id,
124
- taskName: bucket[0].verdict.task_name,
125
- taskPath: bucket[0].verdict.task_path,
126
- trials: bucket,
127
- latestFinalizedAt: last.verdict.finalized_at,
128
- outcome: hasFailed ? "fail" : allPassed ? "pass" : "incomplete"
129
- });
130
- }
131
- sets.sort((a, b) => a.latestFinalizedAt.localeCompare(b.latestFinalizedAt));
132
- return sets;
133
- }
134
- function latestFailedRunSet(sets) {
135
- for (let i = sets.length - 1; i >= 0; i -= 1) {
136
- if (sets[i].outcome === "fail") return sets[i];
137
- }
138
- return null;
139
- }
140
- function latestIncompleteRunSet(sets) {
141
- for (let i = sets.length - 1; i >= 0; i -= 1) {
142
- if (sets[i].outcome === "incomplete") return sets[i];
143
- }
144
- return null;
151
+ unreadableDirs.sort();
152
+ return { trials, staleVersionDirs, unreadableDirs };
145
153
  }
146
154
  async function discoverRunSet(target) {
147
155
  const anchorResult = await readVerdictArtifactDetailed(target);
@@ -151,13 +159,26 @@ async function discoverRunSet(target) {
151
159
  set: null,
152
160
  incompleteSet: null,
153
161
  totalSets: 0,
154
- staleVersionCount: 1
162
+ staleVersionCount: 1,
163
+ unreadableCount: 0,
164
+ unreadablePaths: []
165
+ };
166
+ }
167
+ if (anchorResult.status === "unreadable" && existsSync(join(target, VERDICT_FILENAME))) {
168
+ return {
169
+ kind: "trial-dir",
170
+ set: null,
171
+ incompleteSet: null,
172
+ totalSets: 0,
173
+ staleVersionCount: 0,
174
+ unreadableCount: 1,
175
+ unreadablePaths: [target]
155
176
  };
156
177
  }
157
178
  if (anchorResult.status === "ok") {
158
179
  const anchor = anchorResult.trial;
159
180
  const root = join(target, "..", "..");
160
- const { trials: trials2, staleVersionDirs: staleVersionDirs2 } = await scanVerdictArtifactsDetailed(root);
181
+ const { trials: trials2, staleVersionDirs: staleVersionDirs2, unreadableDirs: unreadableDirs2 } = await scanVerdictArtifactsDetailed(root);
161
182
  const sets2 = groupRunSets(trials2);
162
183
  const own = sets2.find(
163
184
  (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) {
167
188
  set: own,
168
189
  incompleteSet: null,
169
190
  totalSets: Math.max(sets2.length, 1),
170
- staleVersionCount: staleVersionDirs2.length
191
+ staleVersionCount: staleVersionDirs2.length,
192
+ unreadableCount: unreadableDirs2.length,
193
+ unreadablePaths: unreadableDirs2
171
194
  };
172
195
  }
173
196
  if (!existsSync(target)) {
@@ -176,10 +199,12 @@ async function discoverRunSet(target) {
176
199
  set: null,
177
200
  incompleteSet: null,
178
201
  totalSets: 0,
179
- staleVersionCount: 0
202
+ staleVersionCount: 0,
203
+ unreadableCount: 0,
204
+ unreadablePaths: []
180
205
  };
181
206
  }
182
- const { trials, staleVersionDirs } = await scanVerdictArtifactsDetailed(target);
207
+ const { trials, staleVersionDirs, unreadableDirs } = await scanVerdictArtifactsDetailed(target);
183
208
  const sets = groupRunSets(trials);
184
209
  const failedSet = latestFailedRunSet(sets);
185
210
  return {
@@ -187,28 +212,11 @@ async function discoverRunSet(target) {
187
212
  set: failedSet,
188
213
  incompleteSet: failedSet ? null : latestIncompleteRunSet(sets),
189
214
  totalSets: sets.length,
190
- staleVersionCount: staleVersionDirs.length
215
+ staleVersionCount: staleVersionDirs.length,
216
+ unreadableCount: unreadableDirs.length,
217
+ unreadablePaths: unreadableDirs
191
218
  };
192
219
  }
193
- async function loadTrialEvents(runDir) {
194
- let raw;
195
- try {
196
- raw = await readFile(join(runDir, "events.jsonl"), "utf8");
197
- } catch {
198
- return [];
199
- }
200
- const events = [];
201
- for (const line of raw.split("\n")) {
202
- const trimmed = line.trim();
203
- if (!trimmed) continue;
204
- try {
205
- const parsed = JSON.parse(trimmed);
206
- if (typeof parsed === "object" && parsed !== null) events.push(parsed);
207
- } catch {
208
- }
209
- }
210
- return events;
211
- }
212
220
  var execFileAsync = promisify(execFile);
213
221
  var SERVICE = "sh.pome.cli";
214
222
  var ACCOUNT = "hosted";
@@ -1239,4 +1247,4 @@ async function abandonBestEffort(client, sessionId, errorCode) {
1239
1247
  }
1240
1248
  }
1241
1249
 
1242
- 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 };
1250
+ 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 };
@@ -1,5 +1,5 @@
1
1
  import { newGroupId, criterionPhrase } from './chunk-RGZBC7NF.js';
2
- import { runTaskHosted, resolveRunAgentIdentity } from './chunk-IXWYWNXO.js';
2
+ import { runTaskHosted, resolveRunAgentIdentity } from './chunk-FNLI5YX3.js';
3
3
  import './chunk-XOWIA7NR.js';
4
4
  import { createHostedClient, parseTaskFile, outcomeOf } from './chunk-UR3HUJDI.js';
5
5
  import './chunk-NW7HGA2K.js';
@@ -1,5 +1,5 @@
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-IXWYWNXO.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-FNLI5YX3.js';
3
3
  import { readManifest, writeManifest, MANIFEST_JSON, readRequiredManifest, normalizeManifestTwins } from '../../chunk-XOWIA7NR.js';
4
4
  import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-7I53HBHS.js';
5
5
  import '../../chunk-XDU6TD4O.js';
@@ -4513,6 +4513,25 @@ function buildGroupFixPrompt(ctx) {
4513
4513
 
4514
4514
  ${buildGroupFixUserPrompt(ctx)}`;
4515
4515
  }
4516
+ async function loadTrialEvents(runDir) {
4517
+ let raw;
4518
+ try {
4519
+ raw = await readFile(join(runDir, "events.jsonl"), "utf8");
4520
+ } catch {
4521
+ return [];
4522
+ }
4523
+ const events = [];
4524
+ for (const line of raw.split("\n")) {
4525
+ const trimmed = line.trim();
4526
+ if (!trimmed) continue;
4527
+ try {
4528
+ const parsed = JSON.parse(trimmed);
4529
+ if (typeof parsed === "object" && parsed !== null) events.push(parsed);
4530
+ } catch {
4531
+ }
4532
+ }
4533
+ return events;
4534
+ }
4516
4535
 
4517
4536
  // src/cli/main.ts
4518
4537
  var PACKAGE_VERSION = readPackageVersion();
@@ -4520,8 +4539,9 @@ var SESSION_CREATE_FORMATS = /* @__PURE__ */ new Set(["text", "json", "env"]);
4520
4539
  var DEFAULT_AGENT_FILE = "examples/agents/scripted-triage-agent.ts";
4521
4540
  var DEFAULT_AGENT_COMMAND = `node ${DEFAULT_AGENT_FILE}`;
4522
4541
  var MANIFEST_SCHEMA_URL = "https://pome.sh/schemas/v1/pome.json";
4542
+ var MAX_UNREADABLE_PATHS_SHOWN = 5;
4523
4543
  function readPackageVersion() {
4524
- if ("0.23.10".length > 0) return "0.23.10";
4544
+ if ("0.23.11".length > 0) return "0.23.11";
4525
4545
  try {
4526
4546
  const here = dirname(fileURLToPath(import.meta.url));
4527
4547
  const candidates = [
@@ -5014,7 +5034,7 @@ function createProgram() {
5014
5034
  taskForRuns.config.runs
5015
5035
  );
5016
5036
  if (k > 1) {
5017
- const { runTrialGroup } = await import('../../runTrialGroup-NY33WGT2.js');
5037
+ const { runTrialGroup } = await import('../../runTrialGroup-VR2RS2RH.js');
5018
5038
  const fileForRerun = relative(process.cwd(), file);
5019
5039
  const rerunCommand = defaultTask ? options.trials !== void 0 ? `pome run -n ${k}` : "pome run" : `pome run ${fileForRerun && !fileForRerun.startsWith("..") ? fileForRerun : file} -n ${k}`;
5020
5040
  const groupResult = await runTrialGroup({
@@ -5218,8 +5238,20 @@ function createProgram() {
5218
5238
  `${discovery.staleVersionCount} verdict.json file(s) under ${root} are not artifact version ${VERDICT_ARTIFACT_VERSION} (the only version this CLI reads) and were skipped \u2014 re-run \`pome run\` to record those trials again.`
5219
5239
  );
5220
5240
  }
5241
+ if (discovery.unreadableCount > 0) {
5242
+ console.error(
5243
+ `${discovery.unreadableCount} verdict.json file(s) under ${root} could not be read (truncated, hand-edited, or not a verdict artifact) and were skipped:`
5244
+ );
5245
+ for (const path of discovery.unreadablePaths.slice(0, MAX_UNREADABLE_PATHS_SHOWN)) {
5246
+ console.error(` - ${path}`);
5247
+ }
5248
+ const omitted = discovery.unreadablePaths.length - MAX_UNREADABLE_PATHS_SHOWN;
5249
+ if (omitted > 0) {
5250
+ console.error(` (${omitted} more omitted \u2014 kept first ${MAX_UNREADABLE_PATHS_SHOWN})`);
5251
+ }
5252
+ }
5221
5253
  if (discovery.totalSets === 0) {
5222
- if (discovery.staleVersionCount === 0) {
5254
+ if (discovery.staleVersionCount === 0 && discovery.unreadableCount === 0) {
5223
5255
  console.error(
5224
5256
  `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).`
5225
5257
  );
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@pome-sh/cli",
3
- "version": "0.23.10",
3
+ "version": "0.23.11",
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",