@pome-sh/cli 0.40.0 → 0.41.0

package/dist/build-info.json

@@ -1,6 +1,6 @@
 {
   "package": "pome-sh",
-  "version": "0.40.0",
-  "git_sha": "18ff79b115d837fa3081e5925664bd24f5843c8c",
-  "build_time": "2026-08-30T05:50:34.377Z"
+  "version": "0.41.0",
+  "git_sha": "ea56e5f03b661023c9cea61aca95323fa0a9eada",
+  "build_time": "2026-08-30T06:26:23.470Z"
 }

package/dist/{chunk-G7252TQ3.js → chunk-KD6OUKCT.js}

@@ -1,4 +1,4 @@
-import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-BS47AE55.js';
+import { createHostedClient, perTwinReturnedByCloud, parseTaskFile, runAgentCommand, writeRunArtifactsCore, toTwinHttpEvent, redactJsonl, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, evaluationCounts } from './chunk-STMGQ7GU.js';
 import { readSeedFileText, parseSeedFileText, soleTwinOf, seedsForTwins } from './chunk-VPXIGCZW.js';
 import { readManifest, normalizeManifestTwins } from './chunk-DFOQGAKS.js';
 import { MOUNTED_TWINS, HostedAuthError, HostedDiscardRefusedError, HostedQuotaError, HostedOrchError, HostedTrialError, agentResponseSchema, isMultiTwinSeedEnvelope } from './chunk-NNXVR46L.js';
@@ -73,9 +73,10 @@ async function writeVerdictArtifact(runDir, verdict) {
     throw err;
   }
 }
-function looksLikeVerdictArtifactBase(parsed) {
+function isVerdictArtifact(parsed) {
   if (typeof parsed !== "object" || parsed === null) return false;
   const v = parsed;
+  if (v.version !== VERDICT_ARTIFACT_VERSION) return false;
   if (v.source !== "cloud-finalize") return false;
   if (typeof v.session_id !== "string") return false;
   if (typeof v.task_name !== "string") return false;
@@ -84,6 +85,11 @@ function looksLikeVerdictArtifactBase(parsed) {
   if (typeof v.finalized_at !== "string") return false;
   if (typeof v.passed !== "boolean") return false;
   if (typeof v.score !== "number") return false;
+  if (typeof v.state !== "string" || !VALID_STATES.has(v.state)) return false;
+  if (typeof v.evaluated !== "number") return false;
+  if (typeof v.not_evaluated !== "number") return false;
+  if (typeof v.pre_satisfied !== "number") return false;
+  if (typeof v.total !== "number") return false;
   if (!Array.isArray(v.criteria_results)) return false;
   return v.criteria_results.every((r) => {
     if (typeof r !== "object" || r === null) return false;
@@ -92,18 +98,6 @@ function looksLikeVerdictArtifactBase(parsed) {
     return typeof criterion === "object" && criterion !== null && typeof criterion.text === "string" && typeof result.reason === "string" && typeof result.passed === "boolean" && typeof result.skipped === "boolean";
   });
 }
-function isVerdictArtifact(parsed) {
-  if (!looksLikeVerdictArtifactBase(parsed)) return false;
-  const v = parsed;
-  if (v.version !== VERDICT_ARTIFACT_VERSION) return false;
-  if (typeof v.task_path !== "string") return false;
-  if (typeof v.state !== "string" || !VALID_STATES.has(v.state)) return false;
-  if (typeof v.evaluated !== "number") return false;
-  if (typeof v.not_evaluated !== "number") return false;
-  if (typeof v.pre_satisfied !== "number") return false;
-  if (typeof v.total !== "number") return false;
-  return true;
-}
 var MISSING_ERROR_CODES = /* @__PURE__ */ new Set([
   "ENOENT",
   "ENOTDIR",
@@ -128,21 +122,17 @@ async function readVerdictArtifactDetailed(runDir) {
   } catch {
     return { status: "unreadable" };
   }
-  if (!looksLikeVerdictArtifactBase(parsed)) return { status: "unreadable" };
-  const version = typeof parsed.version === "number" ? parsed.version : null;
-  if (version !== VERDICT_ARTIFACT_VERSION) return { status: "stale-version", version };
   if (!isVerdictArtifact(parsed)) return { status: "unreadable" };
   return { status: "ok", trial: { runDir, verdict: parsed } };
 }
 async function scanVerdictArtifactsDetailed(artifactsRoot) {
   const trials = [];
-  const staleVersionDirs = [];
   const unreadableDirs = [];
   let slugs;
   try {
     slugs = await readdir(artifactsRoot);
   } catch {
-    return { trials, staleVersionDirs, unreadableDirs };
+    return { trials, unreadableDirs };
   }
   for (const slug of slugs) {
     const slugDir = join(artifactsRoot, slug);
@@ -156,33 +146,20 @@ async function scanVerdictArtifactsDetailed(artifactsRoot) {
       const runDir = join(slugDir, runId);
       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") unreadableDirs.push(runDir);
     }
   }
   unreadableDirs.sort();
-  return { trials, staleVersionDirs, unreadableDirs };
+  return { trials, unreadableDirs };
 }
 async function discoverRunSet(target) {
   const anchorResult = await readVerdictArtifactDetailed(target);
-  if (anchorResult.status === "stale-version") {
-    return {
-      kind: "trial-dir",
-      set: null,
-      incompleteSet: null,
-      totalSets: 0,
-      staleVersionCount: 1,
-      unreadableCount: 0,
-      unreadablePaths: []
-    };
-  }
   if (anchorResult.status === "unreadable") {
     return {
       kind: "trial-dir",
       set: null,
       incompleteSet: null,
       totalSets: 0,
-      staleVersionCount: 0,
       unreadableCount: 1,
       unreadablePaths: [target]
     };
@@ -190,7 +167,7 @@ async function discoverRunSet(target) {
   if (anchorResult.status === "ok") {
     const anchor = anchorResult.trial;
     const root = join(target, "..", "..");
-    const { trials: trials2, staleVersionDirs: staleVersionDirs2, unreadableDirs: unreadableDirs2 } = await scanVerdictArtifactsDetailed(root);
+    const { trials: trials2, 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
@@ -200,12 +177,11 @@ async function discoverRunSet(target) {
       set: own,
       incompleteSet: null,
       totalSets: Math.max(sets2.length, 1),
-      staleVersionCount: staleVersionDirs2.length,
       unreadableCount: unreadableDirs2.length,
       unreadablePaths: unreadableDirs2
     };
   }
-  const { trials, staleVersionDirs, unreadableDirs } = await scanVerdictArtifactsDetailed(target);
+  const { trials, unreadableDirs } = await scanVerdictArtifactsDetailed(target);
   const sets = groupRunSets(trials);
   const failedSet = latestFailedRunSet(sets);
   return {
@@ -213,7 +189,6 @@ async function discoverRunSet(target) {
     set: failedSet,
     incompleteSet: failedSet ? null : latestIncompleteRunSet(sets),
     totalSets: sets.length,
-    staleVersionCount: staleVersionDirs.length,
     unreadableCount: unreadableDirs.length,
     unreadablePaths: unreadableDirs
   };
@@ -1255,4 +1230,4 @@ async function abandonBestEffort(client, sessionId, errorCode) {
   }
 }
 
-export { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, DEFAULT_DOCS_SITE_ORIGIN, SESSION_TWIN_NAMES, VERDICT_ARTIFACT_VERSION, clearLocalCredentials, discoverRunSet, ensurePomeGitignored, friendlyHostedError, persistCredentialsAfterLogin, postAgentResolver, readLinkCache, resolveCachedAgentId, resolveCredentials, resolveRunAgentIdentity, resolveSeams, runSessionCreate, runSessionList, runSessionStop, runTaskHosted, writeLinkCache };
+export { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, DEFAULT_DOCS_SITE_ORIGIN, SESSION_TWIN_NAMES, clearLocalCredentials, discoverRunSet, ensurePomeGitignored, friendlyHostedError, persistCredentialsAfterLogin, postAgentResolver, readLinkCache, resolveCachedAgentId, resolveCredentials, resolveRunAgentIdentity, resolveSeams, runSessionCreate, runSessionList, runSessionStop, runTaskHosted, writeLinkCache };

package/dist/{chunk-BS47AE55.js → chunk-STMGQ7GU.js}

@@ -1525,7 +1525,7 @@ function scoreFromFinalizeResponse(finalized) {
 }
 function redactJsonl(body) {
   const lines = body.split("\n");
-  const redacted = lines.filter((line) => line.trim().length > 0).map((line) => {
+  const redacted = lines.map((line) => line.trim()).filter((line) => line.length > 0).map((line) => {
     try {
       return JSON.stringify(redactSecrets(JSON.parse(line)));
     } catch {

package/dist/{chunk-AEFXTGW3.js → chunk-WERG4AUH.js}

@@ -1,4 +1,4 @@
-import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-BS47AE55.js';
+import { parseTaskFile, seedStateForTwin, runAgentCommand, writeRunArtifactsCore } from './chunk-STMGQ7GU.js';
 import { createRecorder, bootTwin } from './chunk-5NPGY73F.js';
 import { getAvailablePort } from './chunk-XDU6TD4O.js';
 import { buildEgressAllowlist, readBlockedEgress } from './chunk-CBFKZZBR.js';

package/dist/{runDemo-B2SXM4OC.js → runDemo-DRDMCG25.js}

@@ -1,7 +1,7 @@
 import { newGroupId } from './chunk-JNXZBK3O.js';
 import { DemoCapacityError, capacityLabel, parseCapacityMarker, capacityKindFrom } from './chunk-ZX4WNSZ5.js';
-import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-AEFXTGW3.js';
-import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, isNarrated, outcomeOf, criterionPhrase, narratorReadingLines } from './chunk-BS47AE55.js';
+import { runTask, demoTaskPath, DEMO_TASK_NAME, DEMO_REPO } from './chunk-WERG4AUH.js';
+import { createHostedClient, parseTaskFile, uploadRunBlobs, scoreFromFinalizeResponse, scoreStatus, isNarrated, outcomeOf, criterionPhrase, narratorReadingLines } from './chunk-STMGQ7GU.js';
 import './chunk-NW7HGA2K.js';
 import './chunk-PASFBRK4.js';
 import './chunk-3FZY376K.js';

package/dist/{runTrialGroup-I6FCP2L5.js → runTrialGroup-OGTLUB2X.js}

@@ -1,6 +1,6 @@
 import { newGroupId } from './chunk-JNXZBK3O.js';
-import { runTaskHosted, resolveRunAgentIdentity } from './chunk-G7252TQ3.js';
-import { createHostedClient, parseTaskFile, outcomeOf, isNarrated, criterionPhrase, narratorReadingLines } from './chunk-BS47AE55.js';
+import { runTaskHosted, resolveRunAgentIdentity } from './chunk-KD6OUKCT.js';
+import { createHostedClient, parseTaskFile, outcomeOf, isNarrated, criterionPhrase, narratorReadingLines } from './chunk-STMGQ7GU.js';
 import './chunk-VPXIGCZW.js';
 import './chunk-NW7HGA2K.js';
 import './chunk-PASFBRK4.js';

package/dist/src/cli/main.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
-import { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, clearLocalCredentials, friendlyHostedError, SESSION_TWIN_NAMES, runSessionCreate, runSessionList, runSessionStop, resolveCredentials, runTaskHosted, discoverRunSet, VERDICT_ARTIFACT_VERSION, persistCredentialsAfterLogin, DEFAULT_DOCS_SITE_ORIGIN, resolveSeams, resolveCachedAgentId, readLinkCache, postAgentResolver, writeLinkCache, ensurePomeGitignored } from '../../chunk-G7252TQ3.js';
-import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-AEFXTGW3.js';
-import { parseTaskFile, scoreStatus, runScoreLine, narratorReadingLines, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, criterionRowLine, readCodeCriteria, createHostedClient, toTwinHttpEvent, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-BS47AE55.js';
+import { DEFAULT_CONTROL_PLANE_URL, DEFAULT_DASHBOARD_URL, clearLocalCredentials, friendlyHostedError, SESSION_TWIN_NAMES, runSessionCreate, runSessionList, runSessionStop, resolveCredentials, runTaskHosted, discoverRunSet, persistCredentialsAfterLogin, DEFAULT_DOCS_SITE_ORIGIN, resolveSeams, resolveCachedAgentId, readLinkCache, postAgentResolver, writeLinkCache, ensurePomeGitignored } from '../../chunk-KD6OUKCT.js';
+import { assetPath, runTask, resolvePackageRoot, DEMO_TASK_NAME, demoTaskPath } from '../../chunk-WERG4AUH.js';
+import { parseTaskFile, scoreStatus, runScoreLine, narratorReadingLines, readLatestRun, readMetaSummary, outcomeOf, readConfigTwins, scoreCountsSummary, criterionRowLine, readCodeCriteria, createHostedClient, redactJsonl, scoreFromFinalizeResponse, parseGitHubSeedState, uploadRunBlobs, isPreSatisfied } from '../../chunk-STMGQ7GU.js';
 import '../../chunk-VPXIGCZW.js';
 import '../../chunk-NW7HGA2K.js';
 import { GMAIL_CHECKS } from '../../chunk-JM6VS62R.js';
@@ -21,7 +21,7 @@ import { oneOf, defineCheck, repoRef, VACUITY_SENTINEL_NUMBER, childStatePath, V
 import '../../chunk-YBWG5JK2.js';
 import '../../chunk-HRAD7MRX.js';
 import { eventSchema, isLegacyEventRow } from '../../chunk-6KJC4BTO.js';
-import { redactEvent, redactSecrets } from '../../chunk-SG6ZTIMT.js';
+import { redactSecrets, redactEvent } from '../../chunk-SG6ZTIMT.js';
 import '../../chunk-2K6BJ3PI.js';
 import '../../chunk-FBSA5L36.js';
 import { Command } from 'commander';
@@ -38,17 +38,37 @@ import Anthropic from '@anthropic-ai/sdk';
 import { zodOutputFormat } from '@anthropic-ai/sdk/helpers/zod';
 
 async function readEventsJsonl(runDir) {
+  const file = resolve(runDir, "events.jsonl");
   let raw;
   try {
-    raw = await readFile(join(runDir, "events.jsonl"), "utf8");
+    raw = await readFile(file, "utf8");
   } catch (err) {
     if (err.code === "ENOENT") {
       return { kind: "missing" };
     }
     throw err;
   }
-  const rows = raw.split("\n").map((line) => line.trim()).filter((line) => line.length > 0).map((line) => JSON.parse(line));
-  const events = rows.map((row) => eventSchema.parse(row));
+  const events = [];
+  const lines = raw.split("\n");
+  for (let i = 0; i < lines.length; i += 1) {
+    const line = lines[i].trim();
+    if (line.length === 0) continue;
+    let parsed;
+    try {
+      parsed = JSON.parse(line);
+    } catch (err) {
+      throw new Error(
+        `pome inspect: ${file} line ${i + 1} is not valid JSON \u2014 ${err instanceof Error ? err.message : String(err)}. Re-run the task to record the trace again.`
+      );
+    }
+    const result = eventSchema.safeParse(parsed);
+    if (!result.success) {
+      throw new Error(
+        `pome inspect: ${file} line ${i + 1} is not a recorded event \u2014 ${result.error.issues[0]?.message ?? "unrecognized shape"}. Re-run the task to record the trace again.`
+      );
+    }
+    events.push(result.data);
+  }
   return { kind: "events", events };
 }
 function computeTraceHealth(input) {
@@ -290,8 +310,8 @@ async function startCallbackServer(expectedState) {
   let rejectCode = null;
   let settled = false;
   let deliveredCode = false;
-  const codePromise = new Promise((resolve5, reject) => {
-    resolveCode = resolve5;
+  const codePromise = new Promise((resolve6, reject) => {
+    resolveCode = resolve6;
     rejectCode = reject;
   });
   const fail = (err) => {
@@ -335,9 +355,9 @@ async function startCallbackServer(expectedState) {
     succeed(code);
     void closeServer(server);
   });
-  await new Promise((resolve5, reject) => {
+  await new Promise((resolve6, reject) => {
     server?.once("error", reject);
-    server?.listen(0, "127.0.0.1", resolve5);
+    server?.listen(0, "127.0.0.1", resolve6);
   });
   const address = server.address();
   if (!address || typeof address === "string") {
@@ -374,21 +394,21 @@ async function startCallbackServer(expectedState) {
   };
 }
 function closeServer(server) {
-  return new Promise((resolve5) => {
-    server?.close(() => resolve5());
+  return new Promise((resolve6) => {
+    server?.close(() => resolve6());
   });
 }
 async function openBrowser(url) {
   const command = process.platform === "darwin" ? "open" : process.platform === "win32" ? "powershell.exe" : "xdg-open";
   const args = process.platform === "win32" ? ["-NoProfile", "-Command", "Start-Process", url] : [url];
-  await new Promise((resolve5) => {
+  await new Promise((resolve6) => {
     execFile(command, args, (error) => {
       if (error) {
         console.error(
           "Could not open a browser automatically \u2014 copy the URL above into a browser."
         );
       }
-      resolve5();
+      resolve6();
     });
   });
 }
@@ -591,7 +611,7 @@ function topicMatchesFilter(t, filter) {
   return t.id.includes(f) || t.title.toLowerCase().includes(f) || t.keywords.some((k) => k.toLowerCase().includes(f)) || t.path.toLowerCase().includes(f);
 }
 async function promptLine(rl, q) {
-  return await new Promise((resolve5) => rl.question(q, resolve5));
+  return await new Promise((resolve6) => rl.question(q, resolve6));
 }
 function printTopicUrl(topic, site) {
   console.log(`${site}${topic.path}`);
@@ -3605,13 +3625,19 @@ function validateJsonl(name, raw) {
     const line = lines[i].trim();
     if (line.length === 0) continue;
     nonEmpty += 1;
+    let parsed;
     try {
-      JSON.parse(line);
+      parsed = JSON.parse(line);
     } catch {
       throw new HostedUsageError(
         `pome eval: ${name} is corrupt \u2014 line ${i + 1} is not valid JSON.`
       );
     }
+    if (name === "events.jsonl" && isLegacyEventRow(parsed)) {
+      throw new HostedUsageError(
+        `pome eval: events.jsonl line ${i + 1} has no "kind" \u2014 it was recorded before the unified event shape, and the control plane refuses it. Re-run the task to record the trace again.`
+      );
+    }
   }
   if (name === "events.jsonl" && nonEmpty === 0) {
     throw new HostedUsageError(
@@ -3744,11 +3770,7 @@ async function runEval(options) {
     baseUrl: options.hosted.baseUrl,
     apiKey: options.hosted.apiKey
   });
-  const eventsJsonl = artifacts.eventsJsonl.split("\n").filter((line) => line.trim().length > 0).map((line) => {
-    const parsed = JSON.parse(line);
-    const event2 = isLegacyEventRow(parsed) ? toTwinHttpEvent(parsed) : parsed;
-    return JSON.stringify(redactEvent(event2));
-  }).join("\n") + "\n";
+  const eventsJsonl = redactJsonl(artifacts.eventsJsonl);
   const blobs = {
     eventsJsonl,
     stateInitialJson: JSON.stringify(
@@ -4575,7 +4597,7 @@ function firstSentence(description) {
 function resolveExampleRef(env = process.env) {
   const override = env.POME_EXAMPLE_REF?.trim();
   if (override) return override;
-  const baked = "18ff79b115d837fa3081e5925664bd24f5843c8c".trim() ;
+  const baked = "ea56e5f03b661023c9cea61aca95323fa0a9eada".trim() ;
   return FULL_SHA.test(baked) ? baked : "main";
 }
 function rawUrlFor(example, file, ref) {
@@ -4929,7 +4951,7 @@ 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 ("0.40.0".length > 0) return "0.40.0";
+  if ("0.41.0".length > 0) return "0.41.0";
   try {
     const here = dirname(fileURLToPath(import.meta.url));
     const candidates = [
@@ -5405,7 +5427,7 @@ function createProgram() {
               taskForRuns.config.runs
             );
             if (k > 1) {
-              const { runTrialGroup } = await import('../../runTrialGroup-I6FCP2L5.js');
+              const { runTrialGroup } = await import('../../runTrialGroup-OGTLUB2X.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 ${k}`;
               const groupResult = await runTrialGroup({
@@ -5498,7 +5520,7 @@ function createProgram() {
         process.exitCode = 5;
         return;
       }
-      const { runDemo } = await import('../../runDemo-B2SXM4OC.js');
+      const { runDemo } = await import('../../runDemo-DRDMCG25.js');
       const result = await runDemo({
         apiBase: globals(cmd).apiUrl.replace(/\/$/, ""),
         dashboardBase: process.env.POME_DASHBOARD_URL ?? DEFAULT_DASHBOARD_URL,
@@ -5590,14 +5612,9 @@ function createProgram() {
     }
     const root = target ?? "runs";
     const discovery = await discoverRunSet(resolve(root));
-    if (discovery.staleVersionCount > 0) {
-      console.error(
-        `${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 ${root} could not be read (truncated, hand-edited, or not a verdict artifact) and were skipped:`
+        `${discovery.unreadableCount} verdict.json file(s) under ${root} could not be read (truncated, hand-edited, written by an older CLI, or not a verdict artifact) and were skipped \u2014 re-run \`pome run\` to record those trials again:`
       );
       for (const path of discovery.unreadablePaths.slice(0, MAX_UNREADABLE_PATHS_SHOWN)) {
         console.error(`  - ${path}`);
@@ -5608,7 +5625,7 @@ function createProgram() {
       }
     }
     if (discovery.totalSets === 0) {
-      if (discovery.staleVersionCount === 0 && discovery.unreadableCount === 0) {
+      if (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).`
         );

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pome-sh/cli",
-  "version": "0.40.0",
+  "version": "0.41.0",
   "description": "Test AI agents against digital twins of real SaaS APIs. Records tool-call traces and scores them on pome.sh.",
   "keywords": [
     "ai",

package/tasks/14-stripe-refund-retry.expected-events.jsonl

@@ -1,3 +0,0 @@
-{"ts":"2026-05-11T22:21:54.256Z","run_id":"m0-1-refund-retry","twin":"stripe","request_id":"req_f2e53238-8783-4cfd-b5de-efd60f442be8","correlation_id":"req_f2e53238-8783-4cfd-b5de-efd60f442be8","scenario_step_id":null,"step_id":null,"tool_call_id":null,"method":"POST","path":"/s/default/v1/refunds","request_body":"{\"charge\":\"ch_q9dVWLmN4flQFRzqYRe7iAu1\",\"amount\":7500}","status":402,"response_body":{"error":{"type":"card_error","code":"card_declined","message":"Simulated lost-response failure: refund persisted server-side, but response delivery to the client failed."}},"latency_ms":0,"fidelity":"semantic","state_mutation":true,"state_delta":{"before":null,"after":{"id":"re_LOW1fo91HP1TYIe31f04Kpcn","account_id":"acct_default","charge_id":"ch_q9dVWLmN4flQFRzqYRe7iAu1","payment_intent_id":"pi_pKYzqMO9yYaY2jesB8VcRRxc","amount":7500,"currency":"usd","status":"succeeded","reason":null,"idempotency_key":null,"created":1778538114}},"error":"Simulated lost-response failure: refund persisted server-side, but response delivery to the client failed."}
-{"ts":"2026-05-11T22:21:54.257Z","run_id":"m0-1-refund-retry","twin":"stripe","request_id":"req_c0774431-a954-458c-9b38-8f416cfde198","correlation_id":"req_c0774431-a954-458c-9b38-8f416cfde198","scenario_step_id":null,"step_id":null,"tool_call_id":null,"method":"POST","path":"/s/default/v1/refunds","request_body":"{\"charge\":\"ch_q9dVWLmN4flQFRzqYRe7iAu1\",\"amount\":7500}","status":200,"response_body":{"id":"re_L5jCYOtsrPiLpbCCiQZtnOmv","object":"refund","amount":7500,"balance_transaction":null,"charge":"ch_q9dVWLmN4flQFRzqYRe7iAu1","created":1778538114,"currency":"usd","metadata":{},"payment_intent":"pi_pKYzqMO9yYaY2jesB8VcRRxc","reason":null,"receipt_number":null,"source_transfer_reversal":null,"status":"succeeded","transfer_reversal":null},"latency_ms":1,"fidelity":"semantic","state_mutation":true,"state_delta":{"before":null,"after":{"id":"re_L5jCYOtsrPiLpbCCiQZtnOmv","account_id":"acct_default","charge_id":"ch_q9dVWLmN4flQFRzqYRe7iAu1","payment_intent_id":"pi_pKYzqMO9yYaY2jesB8VcRRxc","amount":7500,"currency":"usd","status":"succeeded","reason":null,"idempotency_key":null,"created":1778538114}},"error":null}
-{"ts":"2026-05-11T22:21:54.257Z","run_id":"m0-1-refund-retry","twin":"stripe","request_id":"req_37c6074d-7b44-4c06-8b69-86567aa2ae8a","correlation_id":"req_37c6074d-7b44-4c06-8b69-86567aa2ae8a","scenario_step_id":null,"step_id":null,"tool_call_id":null,"method":"GET","path":"/s/default/v1/charges/ch_q9dVWLmN4flQFRzqYRe7iAu1","request_body":null,"status":200,"response_body":{"id":"ch_q9dVWLmN4flQFRzqYRe7iAu1","object":"charge","amount":20000,"amount_captured":20000,"amount_refunded":15000,"application":null,"application_fee":null,"application_fee_amount":null,"balance_transaction":"txn_nuda9v4Uks4931U8xOShQZsV","billing_details":{"address":{"city":null,"country":null,"line1":null,"line2":null,"postal_code":null,"state":null},"email":null,"name":null,"phone":null},"calculated_statement_descriptor":null,"captured":true,"created":1778538114,"currency":"usd","customer":null,"description":null,"disputed":false,"failure_balance_transaction":null,"failure_code":null,"failure_message":null,"fraud_details":{},"invoice":null,"livemode":false,"metadata":{},"on_behalf_of":null,"outcome":{"network_status":"approved_by_network","reason":null,"risk_level":"normal","seller_message":"Payment complete.","type":"authorized"},"paid":true,"payment_intent":"pi_pKYzqMO9yYaY2jesB8VcRRxc","payment_method":null,"payment_method_details":{"type":"crypto","crypto":{"buyer_address":null,"network":"base","token_currency":"usdc","transaction_hash":null}},"receipt_email":null,"receipt_number":null,"receipt_url":null,"refunded":false,"review":null,"shipping":null,"source_transfer":null,"statement_descriptor":null,"statement_descriptor_suffix":null,"status":"succeeded","transfer_data":null,"transfer_group":null},"latency_ms":0,"fidelity":"semantic","state_mutation":false,"state_delta":null,"error":null}