@yemi33/minions 0.1.809 → 0.1.811

package/CHANGELOG.md

@@ -1,8 +1,9 @@
 # Changelog
 
-## 0.1.809 (2026-04-10)
+## 0.1.811 (2026-04-10)
 
 ### Features
+- Restructure decompose.md sub-task output format (#655)
 -  CC race condition fixes, doc-chat expand, settings save fix, metric rate
 -  configurable ignored comment authors — auto-filtered, never trigger fixes
 -  add open-source community sources to agentic research pipeline
@@ -10,6 +11,7 @@
 -  cap review→fix cycles at evalMaxIterations (default 3)
 
 ### Fixes
+-  allow agents to explain rationale instead of blindly fixing review feedback
 -  don't URL-encode ADO branchName filter (refs/heads/ format expected raw)
 -  ADO build query filters by source branch (not all org builds)
 -  adoFetch supports method/body for PATCH requests (auto-complete)
@@ -29,7 +31,6 @@
 -  flush queued CC messages as single combined request
 -  human feedback fixes are never capped (only minion review→fix loop is)
 -  5 bugs from comprehensive audit
--  stale ADO build detection + merge conflict auto-fix for both platforms
 
 ## 0.1.782 (2026-04-10)
 

package/dashboard/js/command-center.js

@@ -378,11 +378,14 @@ async function ccSend() {
   }
   var wasAborted = await _ccDoSend(message);
 
-  // Flush queued messages — combine into single request
-  if (tab._queue && tab._queue.length > 0) {
-    var combined = tab._queue.splice(0).join('\n\n');
+  // Drain queued messages one at a time with abort delay
+  while (tab._queue && tab._queue.length > 0) {
+    if (wasAborted) {
+      await new Promise(function(r) { setTimeout(r, 600); });
+    }
+    var next = tab._queue.shift();
     _renderQueueIndicator();
-    await _ccDoSend(combined);
+    wasAborted = await _ccDoSend(next);
   }
 }
 

package/engine/lifecycle.js

@@ -186,7 +186,7 @@ function checkPlanCompletion(meta, config) {
 
   // 4. Create verification work item (build, test, start webapp, write testing guide)
   const existingVerify = allWorkItems.find(w => w.sourcePlan === planFile && w.itemType === 'verify');
-  if (!existingVerify && doneItems.length > 0 && failedItems.length === 0) {
+  if (!existingVerify && doneItems.length > 0) {
     const verifyId = 'PL-' + shared.uid();
     const planSlug = planFile.replace('.json', '');
 
@@ -1352,7 +1352,7 @@ function handleDecompositionResult(stdout, meta, config) {
       // Create child work items
       for (const sub of subItems) {
         if (data.some(i => i.id === sub.id)) continue; // dedupe
-        data.push({
+        const childItem = {
           id: sub.id,
           title: sub.name || sub.title || `Sub-task of ${parentId}`,
           type: (sub.estimated_complexity === 'large') ? 'implement:large' : 'implement',
@@ -1367,7 +1367,15 @@ function handleDecompositionResult(stdout, meta, config) {
           featureBranch: p.featureBranch,
           created: ts(),
           createdBy: 'decomposition',
-        });
+        };
+        // Persist structured fields from decompose output (additive — safe if absent)
+        if (Array.isArray(sub.acceptance_criteria) && sub.acceptance_criteria.length > 0) {
+          childItem.acceptance_criteria = sub.acceptance_criteria;
+        }
+        if (Array.isArray(sub.scope_boundaries) && sub.scope_boundaries.length > 0) {
+          childItem.scope_boundaries = sub.scope_boundaries;
+        }
+        data.push(childItem);
       }
       return data;
     });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yemi33/minions",
-  "version": "0.1.809",
+  "version": "0.1.811",
   "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/decompose.md

@@ -48,13 +48,13 @@ Write the decomposition result as a JSON code block in your response:
     {
       "id": "{{item_id}}-a",
       "name": "Short descriptive name",
-      "objective": "One-sentence goal: what this sub-task achieves.",
+      "objective": "One-sentence goal — what this sub-task achieves in the overall plan.",
       "description": "What to build, where, and how. Be specific enough that an engineer can implement without further exploration.",
+      "expected_output": "What 'done' looks like — artifact produced, PR shape, file format, or observable behavior.",
+      "scope_boundaries": ["What is explicitly OUT of scope for this sub-task", "Another exclusion"],
       "estimated_complexity": "small|medium",
       "depends_on": [],
-      "acceptance_criteria": ["Criterion 1", "Criterion 2"],
-      "expected_output": "What done looks like — the artifact produced (e.g., a PR adding X, a new file at Y, a config change).",
-      "scope_boundaries": ["Out-of-scope item 1", "Out-of-scope item 2"]
+      "acceptance_criteria": ["Verifiable criterion 1", "Verifiable criterion 2"]
     }
   ]
 }

package/playbooks/fix.md

@@ -29,7 +29,9 @@ Use subagents only for genuinely parallel, independent tasks. For sequential wor
 
 1. You are already in the correct worktree on branch `{{pr_branch}}`. Do NOT create additional worktrees.
 
-2. Fix each issue listed above
+2. For each issue listed above, use your judgment:
+   - **Fix it** if the feedback is valid and improves the code
+   - **Explain your rationale** if you believe the current approach is correct — reply on the review thread explaining why, with specific reasoning (e.g., performance, consistency with codebase patterns, intentional design choice). Do NOT silently ignore feedback — always respond.
 
 3. Handle merge conflicts if any:
    - If `git pull` or the PR shows conflicts, resolve them in the worktree
@@ -64,11 +66,11 @@ Do NOT remove the worktree — the engine handles cleanup automatically.
 
 ## Resolve Review Comments
 
-After pushing, resolve each review comment/thread that you've addressed:
-- **GitHub**: Reply to each review comment confirming the fix, then resolve the conversation if possible
-- **ADO**: Reply to each thread with what was fixed, then set the thread status to `fixed` or `closed`
-
-Do NOT leave review threads open if you've addressed the finding — unresolved threads block auto-merge on some repos and create noise for human reviewers.
+After pushing, respond to each review comment/thread:
+- **If you fixed it**: Reply confirming the fix, then resolve the thread
+- **If you chose not to fix it**: Reply with your rationale explaining why the current approach is preferred — leave the thread open for the reviewer to decide
+- **GitHub**: Reply to each review comment, resolve conversations you've fixed
+- **ADO**: Reply to each thread, set status to `fixed` or `closed` for fixes; leave `active` for rationale replies
 
 ## When to Stop