brain-dev 2.0.2 → 2.0.3

package/commands/brain/new-task.md

@@ -17,15 +17,33 @@ Unlike quick tasks, new-task applies the full pipeline. Unlike new-project, it w
 </objective>
 
 <context>
-Use new-task for features, refactors, or epics that need more ceremony than a quick fix but less than a new project.
-Examples: "add payment integration", "refactor auth to JWT", "add rate limiting to API"
+Use new-task for features, refactors, bug fixes, or epics that need structure.
+Examples: "add payment integration", "refactor auth to JWT", "fix ATS rescoring failure after AI improve"
 </context>
 
+<critical-rules>
+NEVER skip the pipeline. Even if the fix seems obvious:
+- DO NOT start debugging, coding, or investigating before running `npx brain-dev new-task "description"`
+- DO NOT "just fix it" and skip discuss/plan/verify steps
+- The pipeline exists to catch edge cases, ensure test coverage, and create audit trails
+- A "quick fix" that skips pipeline often misses root causes or introduces regressions
+
+If the user describes a BUG:
+1. First run `npx brain-dev new-task "fix: description of bug"` to create the task
+2. The discuss step IS the investigation — use it to understand root cause
+3. The plan step defines the fix + tests
+4. The execute step implements with TDD
+5. The verify step confirms the fix works
+
+If the task seems too small for full pipeline, suggest `/brain:quick` instead. But NEVER skip pipeline silently.
+</critical-rules>
+
 <process>
-1. Create: `npx brain-dev new-task "description"` [--research] [--light]
-2. Follow the step instructions output by the command
-3. After each step completes, run: `npx brain-dev new-task --continue`
-4. Pipeline: discuss → plan → execute → verify → complete
-5. On completion, optionally promote to roadmap: `npx brain-dev new-task --promote N`
-6. List tasks: `npx brain-dev new-task --list`
+1. Ask the user what task they want to work on using AskUserQuestion
+2. Run: `npx brain-dev new-task "description"` [--research] [--light]
+3. READ the output carefully — it contains step-by-step instructions
+4. Follow EACH step in order. After each step completes, run: `npx brain-dev new-task --continue`
+5. Pipeline: discuss → plan → execute → verify → complete
+6. On completion, optionally promote to roadmap: `npx brain-dev new-task --promote N`
+7. List tasks: `npx brain-dev new-task --list`
 </process>

package/hooks/bootstrap.sh

@@ -58,9 +58,11 @@ try {
     '',
     'Instructions for Claude:',
     '- When user types /brain:<command>, run: npx brain-dev <command> [args]',
-    '- Read output and present results. Follow any instructions in output.',
+    '- Read output carefully and FOLLOW the step instructions. Do not skip steps or improvise.',
     '- Use /brain:progress to check current state before suggesting next steps.',
-    '- NEVER run brain-dev init after an update. Updates are backward compatible. /brain:update handles migration automatically.',
+    '- NEVER run brain-dev init after an update. Updates are backward compatible.',
+    '- NEVER start coding/debugging before running the brain command. The pipeline IS the process.',
+    '- When user describes a bug or task, FIRST run the appropriate brain command (new-task, quick, story), THEN follow its output.',
     '--- End Brain Context ---'
   ];
   console.log(lines.join('\n'));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "brain-dev",
-  "version": "2.0.2",
+  "version": "2.0.3",
   "description": "AI-powered development workflow orchestrator",
   "author": "halilcosdu",
   "license": "MIT",