@yemi33/minions 0.1.956 → 0.1.958

package/CHANGELOG.md

@@ -1,8 +1,9 @@
 # Changelog
 
-## 0.1.956 (2026-04-14)
+## 0.1.958 (2026-04-14)
 
 ### Features
+- surface in-flight tool calls in lastAction (#1064)
 - wire agentBusyReassignMs into settings UI and persist
 -  ADO throttle detection, poll guards, and dashboard banner (#1051)
 - gate auto-fix conflict dispatch behind autoFixConflicts flag
@@ -22,9 +23,11 @@
 - replace magic string 'active' with PR_STATUS.ACTIVE in lifecycle.js (#969)
 - fix dashboard plan-pause nested lock violation (#968)
 - add missing branch_name to central dispatch vars (#967)
-- fix dispatch.js mutator fallback to use nullish coalescing (#966)
 
 ### Fixes
+- enforce --timeout 1 on all azureauth calls to prevent agent orphans (#1063)
+- cancel steering kill watcher on resume spawn (#1052) (#1062)
+- suppress warn for optional template variables (#1061)
 - fix ENGINE_DEFAULTS ref and derive boolean settings from defaults
 - add autoFixConflicts to boolean settings persist list
 - tighten build query guard to require mergeCommitId
@@ -42,9 +45,6 @@
 - steering kill on no-session re-queues dispatch instead of erroring (#1015)
 - inject cached ADO token into spawned agents (#998) (#1012)
 -  qaAbort race + skip kill on completed doc-chat
--  doc-chat Stop button actually works — release server guard on disconnect
--  CC streaming auto-retries fresh session when resume fails
--  429 retry on abort race applies to all sends, not just queued
 
 ### Other
 - test(engine): add regression tests for autoFixConflicts flag and DEFAULTS alias

package/docs/auto-discovery.md

@@ -137,7 +137,7 @@ The engine directly polls the Azure DevOps REST API for **all** PR metadata: bui
 | `buildStatus` | PR statuses (codecoverage/deploy/build/ci contexts) | `passing` / `failing` / `running` / `none` |
 | `buildFailReason` | Failed status description | Set on failure, cleared otherwise |
 
-**Auth:** Bearer token via `azureauth ado token --output token` (cached 30 minutes).
+**Auth:** Bearer token via `azureauth ado token --output token --timeout 1` (cached 30 minutes). The `--timeout 1` flag is required — without it, azureauth can hang indefinitely in headless sessions.
 
 This feeds `discoverFromPrs` — when `buildStatus` flips to `"failing"`, the next discovery tick dispatches a fix agent. When `status` becomes `"merged"`, the PR drops out of active polling.
 

package/engine/playbook.js

@@ -214,6 +214,22 @@ function resolveTaskContext(item, config) {
 // are optional by design. Only variables that make the playbook non-functional
 // when absent are listed here.
 
+// ─── Optional Template Variables ────────────────────────────────────────────
+// Variables that legitimately resolve to empty string for ad-hoc work items.
+// These are silently filtered out of the empty-string warning to avoid
+// masking real warnings. Add new optional vars here when they are contextual
+// (plan-linked, checkpoint-dependent, etc.) and not required for the playbook
+// to function.
+
+const PLAYBOOK_OPTIONAL_VARS = new Set([
+  'source_plan',          // only set when work item is linked to a plan
+  'plan_slug',            // derived from source_plan
+  'additional_context',   // only set when item has a prompt
+  'references',           // only set when item.references has entries
+  'acceptance_criteria',  // only set when item.acceptanceCriteria has entries
+  'checkpoint_context',   // only set when resuming from a prior timeout
+]);
+
 const PLAYBOOK_REQUIRED_VARS = {
   'implement':            ['item_id', 'item_name', 'branch_name', 'project_path'],
   'implement-shared':     ['item_id', 'item_name', 'branch_name', 'worktree_path'],
@@ -364,9 +380,9 @@ function renderPlaybook(type, vars) {
     log('warn', `Playbook "${type}": substituted values contain unresolved {{...}} patterns (potential self-reference): ${selfRefVars.join(', ')}`);
   }
 
-  // Warn on variables that resolved to empty string
+  // Warn on variables that resolved to empty string (skip known-optional vars)
   const emptyVars = Object.entries(allVars)
-    .filter(([, val]) => String(val) === '')
+    .filter(([key, val]) => String(val) === '' && !PLAYBOOK_OPTIONAL_VARS.has(key))
     .map(([key]) => key);
   if (emptyVars.length > 0) {
     log('warn', `Playbook "${type}": template variables resolved to empty string: ${emptyVars.join(', ')}`);
@@ -551,6 +567,7 @@ module.exports = {
   renderPlaybook,
   validatePlaybookVars,
   PLAYBOOK_REQUIRED_VARS,
+  PLAYBOOK_OPTIONAL_VARS,
   buildSystemPrompt,
   buildAgentContext,
   selectPlaybook,

package/engine/queries.js

@@ -42,6 +42,34 @@ function readHeadTail(filePath, bytes = 1024) {
   }
 }
 
+/**
+ * Detect in-flight tool calls from live-output.log tail content.
+ * Scans for task_started events with no matching task_notification (by task_id).
+ * Returns { description, taskId } for the most recent in-flight tool, or null.
+ */
+function detectInFlightTool(tail) {
+  if (!tail) return null;
+  const lines = tail.split('\n');
+  const completed = new Set();
+
+  // Reverse scan: collect task_notification ids first (most recent lines),
+  // then the first task_started not in completed is the in-flight tool.
+  for (let i = lines.length - 1; i >= 0; i--) {
+    const line = lines[i];
+    if (!line.includes('"task_')) continue; // fast skip non-task lines
+    try {
+      const parsed = JSON.parse(line);
+      if (parsed.type !== 'system') continue;
+      if (parsed.subtype === 'task_notification') {
+        completed.add(parsed.task_id);
+      } else if (parsed.subtype === 'task_started' && !completed.has(parsed.task_id)) {
+        return { description: parsed.description || null, taskId: parsed.task_id };
+      }
+    } catch { /* not valid JSON — skip heartbeats, headers, partial lines */ }
+  }
+  return null;
+}
+
 // ── Paths ───────────────────────────────────────────────────────────────────
 
 const MINIONS_DIR = shared.MINIONS_DIR;
@@ -220,7 +248,7 @@ function getAgentStatus(agentId) {
     if (active._blockingToolCall) {
       result._blockingToolCall = active._blockingToolCall;
     }
-    // Detect permission-waiting: read only head+tail of live-output.log (max 2KB total)
+    // Detect permission-waiting and in-flight tools: read only head+tail of live-output.log (max 2KB total)
     try {
       const liveLogPath = path.join(AGENTS_DIR, agentId, 'live-output.log');
       const { head, tail } = readHeadTail(liveLogPath, 1024);
@@ -239,6 +267,11 @@ function getAgentStatus(agentId) {
             result._warning = 'Possibly waiting for permission approval — agent is not in bypass mode';
           }
         }
+        // Detect in-flight tool calls (task_started with no task_notification)
+        const inFlight = detectInFlightTool(tail);
+        if (inFlight && inFlight.description) {
+          result._runningToolDescription = inFlight.description;
+        }
       }
     } catch { /* optional — don't block status */ }
     return result;
@@ -328,7 +361,7 @@ function getAgents(config) {
     const s = getAgentStatus(a.id); // derives from dispatch.json
 
     let lastAction = 'Waiting for assignment';
-    if (s.status === 'working') lastAction = `Working: ${s.task}`;
+    if (s.status === 'working') lastAction = s._runningToolDescription ? `Running: ${s._runningToolDescription}` : `Working: ${s.task}`;
     else if (s.status === 'done') lastAction = `Done: ${s.task}`;
     else if (s.status === 'error') lastAction = `Error: ${s.task}`;
     else if (inboxFiles.length > 0) {
@@ -1033,6 +1066,7 @@ module.exports = {
   // Helpers
   timeSince,
   readHeadTail, // exported for testing
+  detectInFlightTool, // exported for testing
   resetPrdInfoCache,
 
   // Core state

package/engine/shared.js

@@ -555,6 +555,7 @@ const ENGINE_DEFAULTS = {
   autoCompletePrs: false, // auto-merge PRs when builds green + review approved (opt-in)
   prMergeMethod: 'squash', // merge method: squash, merge, rebase
   ignoredCommentAuthors: [], // comments from these authors are auto-closed and never trigger fixes
+  agentBusyReassignMs: 600000, // 10min — reassign work item to another agent if preferred agent is busy beyond this threshold
   ccModel: 'sonnet', // model for Command Center and doc-chat (sonnet, haiku, opus)
   ccEffort: null, // effort level for CC/doc-chat (null, 'low', 'medium', 'high')
   heartbeatTimeouts: {}, // populated after WORK_TYPE is defined (below)

package/engine.js

@@ -1051,8 +1051,12 @@ async function spawnAgent(dispatchItem, config) {
         return;
       }
 
-      // Re-attach to existing tracking
-      activeProcesses.set(id, { proc: resumeProc, agentId, startedAt: procInfo.startedAt, sessionId: steerSessionId, _steeringAt: Date.now(), lastRealOutputAt: Date.now() });
+      // Re-attach to existing tracking — do NOT carry _steeringAt forward (#1052).
+      // The kill watcher in timeout.js fires 30s after _steeringAt is set. If we carry it
+      // into the resumed process, it kills the resumed session. The kill watcher only exists
+      // to handle cases where the original kill didn't take effect — once the process has
+      // exited and the resume is spawned, _steeringAt must not be present.
+      activeProcesses.set(id, { proc: resumeProc, agentId, startedAt: procInfo.startedAt, sessionId: steerSessionId, lastRealOutputAt: Date.now() });
 
       // Reset output buffers so post-completion parsing only sees the resumed session
       stdout = '';
@@ -1300,6 +1304,7 @@ async function spawnAgent(dispatchItem, config) {
     const item = dispatch.pending.splice(idx, 1)[0];
     item.started_at = startedAt;
     delete item.skipReason;
+    delete item._agentBusySince;
     if (!dispatch.active.some(d => d.id === id)) {
       dispatch.active.push(item);
     }
@@ -3324,7 +3329,46 @@ async function tickInner() {
       log('warn', `Duplicate dispatch ID ${item.id} in pending queue — skipping`);
       continue;
     }
-    if (busyAgents.has(item.agent)) continue;
+    if (busyAgents.has(item.agent)) {
+      // Agent busy reassignment: if item has been waiting on a busy agent past the threshold,
+      // try to find an alternative agent via routing. Skip explicitly assigned items.
+      const reassignMs = config.engine?.agentBusyReassignMs ?? ENGINE_DEFAULTS.agentBusyReassignMs;
+      const isExplicitReassign = !!item.meta?.item?.agent;
+      if (!isExplicitReassign && reassignMs > 0 && item._agentBusySince) {
+        const busySinceMs = new Date(item._agentBusySince).getTime();
+        if (Date.now() - busySinceMs > reassignMs) {
+          const originalAgent = item.agent;
+          const altAgent = resolveAgent(item.type, config);
+          if (altAgent && altAgent !== originalAgent && !busyAgents.has(altAgent)) {
+            log('info', `Reassigning ${item.id} from ${originalAgent} to ${altAgent} — agent busy > ${reassignMs}ms`);
+            item.agent = altAgent;
+            item.agentName = config.agents[altAgent]?.name || tempAgents.get(altAgent)?.name || altAgent;
+            item.agentRole = config.agents[altAgent]?.role || tempAgents.get(altAgent)?.role || 'Agent';
+            delete item._agentBusySince;
+            delete item.skipReason;
+            // Persist reassignment to dispatch.json
+            mutateDispatch((dp) => {
+              const p = (dp.pending || []).find(d => d.id === item.id);
+              if (p) {
+                p.agent = altAgent;
+                p.agentName = item.agentName;
+                p.agentRole = item.agentRole;
+                delete p._agentBusySince;
+                delete p.skipReason;
+              }
+              return dp;
+            });
+            // Fall through to branch mutex / concurrency checks below
+          } else {
+            continue; // No alternative agent available — keep waiting
+          }
+        } else {
+          continue; // Below threshold — keep waiting
+        }
+      } else {
+        continue; // No _agentBusySince set yet or explicitly assigned — skip
+      }
+    }
     // Branch mutex: skip items targeting a branch already locked by an active or newly-dispatched task
     const itemBranch = item.meta?.branch ? sanitizeBranch(item.meta.branch) : null;
     if (itemBranch && lockedBranches.has(itemBranch)) continue;
@@ -3412,6 +3456,18 @@ async function tickInner() {
         reason = 'branch_locked';
       }
     }
+    // Track when item first became blocked on a busy agent for reassignment threshold
+    if (reason === 'agent_busy') {
+      if (!item._agentBusySince) {
+        item._agentBusySince = ts();
+        skipReasonChanged = true;
+      }
+    } else {
+      if (item._agentBusySince) {
+        delete item._agentBusySince;
+        skipReasonChanged = true;
+      }
+    }
     if (item.skipReason !== reason) {
       item.skipReason = reason;
       skipReasonChanged = true;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yemi33/minions",
-  "version": "0.1.956",
+  "version": "0.1.958",
   "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"

package/playbooks/shared-rules.md

@@ -54,3 +54,5 @@ node engine/ado-status.js <prNumber> --live --project MyProject
 Output is JSON with the same fields. Exit 0 on success, 1 if not found.
 
 **Never make raw `curl` calls to ADO APIs directly.** Use `node engine/ado-status.js` which routes through `ado.js` — authenticated, retried, circuit-broken. Raw `azureauth` + curl bypasses all of that.
+
+**If you must run `azureauth` directly, ALWAYS include `--timeout 1`.** Without this flag, `azureauth ado token` can hang indefinitely waiting for interactive broker UI that never appears in headless agent sessions. This causes the Claude Code process to silently exit and the engine to declare the agent orphaned. Example: `azureauth ado token --mode iwa --mode broker --output token --timeout 1`.