@yemi33/minions 0.1.163 → 0.1.165

package/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## 0.1.165 (2026-04-02)
+
+### Engine
+- engine.js
+
+## 0.1.164 (2026-04-02)
+
+### Engine
+- engine/lifecycle.js
+
 ## 0.1.163 (2026-04-02)
 
 ### Engine

package/engine/lifecycle.js

@@ -1188,6 +1188,11 @@ function runPostCompletionHooks(dispatchItem, agentId, code, stdout, config) {
         ? path.join(MINIONS_DIR, 'work-items.json')
         : meta.project?.name ? path.join(MINIONS_DIR, 'projects', meta.project.name, 'work-items.json') : null;
       if (wiPath) {
+        // Cost ceiling circuit breaker config — resolved outside lock for clarity
+        const engineCfg = config?.engine || {};
+        const evalMaxCost = engineCfg.evalMaxCost != null ? engineCfg.evalMaxCost : shared.ENGINE_DEFAULTS.evalMaxCost;
+
+        // Single lock: accumulate cost AND check ceiling atomically (no TOCTOU)
         mutateJsonFileLocked(wiPath, (items) => {
           if (!Array.isArray(items)) return items;
           const wi = items.find(i => i.id === meta.item.id);
@@ -1195,29 +1200,17 @@ function runPostCompletionHooks(dispatchItem, agentId, code, stdout, config) {
             wi._totalCostUsd = (wi._totalCostUsd || 0) + (taskUsage.costUsd || 0);
             wi._totalInputTokens = (wi._totalInputTokens || 0) + (taskUsage.inputTokens || 0);
             wi._totalOutputTokens = (wi._totalOutputTokens || 0) + (taskUsage.outputTokens || 0);
+
+            // Cost ceiling circuit breaker — treat like evalMaxIterations exceeded
+            if (evalMaxCost != null && evalMaxCost > 0 &&
+                wi._totalCostUsd > evalMaxCost && wi.status !== 'needs-human-review') {
+              wi.status = 'needs-human-review';
+              wi.failReason = `Cumulative cost $${wi._totalCostUsd.toFixed(2)} exceeds evalMaxCost ceiling $${evalMaxCost.toFixed(2)}`;
+              log('warn', `Work item ${meta.item.id} exceeded cost ceiling ($${wi._totalCostUsd.toFixed(2)} > $${evalMaxCost.toFixed(2)}) — needs-human-review`);
+            }
           }
           return items;
         }, { defaultValue: [] });
-
-        // Cost ceiling circuit breaker — treat like evalMaxIterations exceeded
-        const engineCfg = config?.engine || {};
-        const evalMaxCost = engineCfg.evalMaxCost != null ? engineCfg.evalMaxCost : shared.ENGINE_DEFAULTS.evalMaxCost;
-        if (evalMaxCost != null && evalMaxCost > 0) {
-          const freshItems = safeJson(wiPath) || [];
-          const wi = freshItems.find(i => i.id === meta.item.id);
-          if (wi && wi._totalCostUsd > evalMaxCost && wi.status !== 'needs-human-review') {
-            mutateJsonFileLocked(wiPath, (items) => {
-              if (!Array.isArray(items)) return items;
-              const target = items.find(i => i.id === meta.item.id);
-              if (target) {
-                target.status = 'needs-human-review';
-                target.failReason = `Cumulative cost $${wi._totalCostUsd.toFixed(2)} exceeds evalMaxCost ceiling $${evalMaxCost.toFixed(2)}`;
-                log('warn', `Work item ${meta.item.id} exceeded cost ceiling ($${wi._totalCostUsd.toFixed(2)} > $${evalMaxCost.toFixed(2)}) — needs-human-review`);
-              }
-              return items;
-            }, { defaultValue: [] });
-          }
-        }
       }
     } catch (err) { log('warn', `Cost accumulation: ${err.message}`); }
   }
@@ -1358,15 +1351,15 @@ function runPostCompletionHooks(dispatchItem, agentId, code, stdout, config) {
   }
 
   if (!isSuccess && meta?.item?.id) {
-    const wiPath = resolveWiPath(meta);
-    if (wiPath) {
-      let finalStatus = null;
-      try {
+    // Auto-retry with atomic read-modify-write (no TOCTOU race)
+    try {
+      const wiPath = resolveWiPath(meta);
+      if (wiPath) {
+        let retriesExhausted = false;
         mutateJsonFileLocked(wiPath, (items) => {
           if (!Array.isArray(items)) return items;
           const wi = items.find(i => i.id === meta.item.id);
           if (!wi) return items;
-
           const retries = wi._retryCount || 0;
           if (retries < 3) {
             log('info', `Agent failed for ${meta.item.id} — auto-retry ${retries + 1}/3`);
@@ -1374,22 +1367,28 @@ function runPostCompletionHooks(dispatchItem, agentId, code, stdout, config) {
             wi.status = 'pending';
             delete wi.dispatched_at;
             delete wi.dispatched_to;
-            finalStatus = 'pending';
+            if (type === 'decompose') delete wi._decomposing;
           } else {
-            wi.status = 'failed';
-            wi.failReason = 'Agent failed (3 retries exhausted)';
-            wi.failedAt = ts();
-            finalStatus = 'failed';
+            retriesExhausted = true;
           }
+          // Clear _decomposing flag on any decompose failure to prevent permanent stuck
           if (type === 'decompose') delete wi._decomposing;
           return items;
-        });
-      } catch (err) { log('warn', `Retry update: ${err.message}`); }
-      // Sync status to PRD outside the work-items lock
-      if (finalStatus) {
-        syncPrdItemStatus(meta.item.id, finalStatus, meta.item?.sourcePlan);
+        }, { defaultValue: [] });
+        if (retriesExhausted) {
+          updateWorkItemStatus(meta, 'failed', 'Agent failed (3 retries exhausted)');
+        }
+      } else {
+        // No wiPath — can't read retries, fall back to meta snapshot
+        const retries = meta.item._retryCount || 0;
+        if (retries < 3) {
+          log('info', `Agent failed for ${meta.item.id} — auto-retry ${retries + 1}/3`);
+          updateWorkItemStatus(meta, 'pending', '');
+        } else {
+          updateWorkItemStatus(meta, 'failed', 'Agent failed (3 retries exhausted)');
+        }
       }
-    }
+    } catch (err) { log('warn', `Retry/decompose update: ${err.message}`); }
   }
   // Meeting post-completion: collect findings/debate/conclusion
   if (type === 'meeting' && meta?.meetingId) {

package/engine.js

@@ -1802,6 +1802,41 @@ function discoverCentralWorkItems(config) {
         assignedProject: projects.length > 0 ? projects[i % projects.length] : null
       }));
 
+      // Inject checkpoint context if agent left a checkpoint.json from a prior run
+      let fanOutCheckpointContext = '';
+      try {
+        const fanFirstProject = projects[0];
+        const fanBranch = item.branch || `work/${item.id}`;
+        const fanWtPath = fanFirstProject?.localPath
+          ? path.resolve(fanFirstProject.localPath, config.engine?.worktreeRoot || '../worktrees', fanBranch)
+          : '';
+        const fanCpPath = fanWtPath ? path.join(fanWtPath, 'checkpoint.json') : '';
+        if (fanCpPath && fs.existsSync(fanCpPath)) {
+          const fanCpData = JSON.parse(fs.readFileSync(fanCpPath, 'utf8'));
+          const fanCpCount = (item._checkpointCount || 0) + 1;
+          if (fanCpCount > 3) {
+            log('warn', `Work item ${item.id} exceeded 3 checkpoint-resumes — marking as needs-human-review`);
+            item.status = 'needs-human-review';
+            item._checkpointCount = fanCpCount;
+            needsWrite = true;
+            continue;
+          }
+          item._checkpointCount = fanCpCount;
+          const fanCpSummary = [
+            `## Checkpoint (Resume #${fanCpCount}/3)`,
+            '',
+            'A previous agent run timed out but left a checkpoint. Continue from where it left off.',
+            '',
+            Array.isArray(fanCpData.completed) && fanCpData.completed.length > 0 ? `### Completed\n${fanCpData.completed.map(s => '- ' + s).join('\n')}` : '',
+            Array.isArray(fanCpData.remaining) && fanCpData.remaining.length > 0 ? `### Remaining\n${fanCpData.remaining.map(s => '- ' + s).join('\n')}` : '',
+            Array.isArray(fanCpData.blockers) && fanCpData.blockers.length > 0 ? `### Blockers\n${fanCpData.blockers.map(s => '- ' + s).join('\n')}` : '',
+            fanCpData.branch_state ? `### Branch State\n${fanCpData.branch_state}` : '',
+          ].filter(Boolean).join('\n');
+          fanOutCheckpointContext = fanCpSummary;
+          log('info', `Injecting checkpoint context for ${item.id} (resume #${fanCpCount})`);
+        }
+      } catch (e) { log('warn', `checkpoint read for ${item.id}: ${e.message}`); }
+
       for (const { agent, assignedProject } of assignments) {
         const fanKey = `${key}-${agent.id}`;
         if (isAlreadyDispatched(fanKey)) continue;
@@ -1833,6 +1868,10 @@ function discoverCentralWorkItems(config) {
           ? '## Push Branch\n\n**PR creation is skipped for this work item.** Push your branch and report the branch name.\n\n```bash\ngit push -u origin ' + fanBranch + '\n```\n\nInclude the branch name in your completion summary.'
           : '## Create PR (MANDATORY)\n\n**Your task is NOT complete until a pull request exists.** If PR creation fails, retry up to 3 times before reporting the error.\n\n{{pr_create_instructions}}\n- sourceRefName: `refs/heads/' + fanBranch + '`\n- targetRefName: `refs/heads/{{main_branch}}`\n- title: `{{commit_message}}`\n- labels: `["minions:{{agent_id}}"]`\n\nInclude in the PR description:\n- What was built and why\n- Files changed\n- How to build and test, browser URL if applicable\n- Test plan\n\n## Post self-review on PR\n\n{{pr_comment_instructions}}\n- pullRequestId: `<from PR creation>`\n- Re-read your own diff critically before posting\n- Sign: `Built by Minions ({{agent_name}} — {{agent_role}})`';
 
+        // Inject checkpoint context (computed once above the loop)
+        vars.checkpoint_context = fanOutCheckpointContext;
+
+
         if (workType === 'ask') {
           vars.question = item.title + (item.description ? '\n\n' + item.description : '');
           vars.task_id = item.id;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yemi33/minions",
-  "version": "0.1.163",
+  "version": "0.1.165",
   "description": "Multi-agent AI dev team that runs from ~/.minions/ — five autonomous agents share a single engine, dashboard, and knowledge base",
   "bin": {
     "minions": "bin/minions.js"