self-check

# Self-Check: Self-Observation Protocol Invoke mid-execution checkpoint to detect warning signs and course-correct. Prevents drift, confusion, and wasted effort through systematic self-observation. ## Instructions **When to Invoke (Mandatory):** - 25% task completion (early warning) - Before delegating tasks to other agents - After loading 5+ directives or large documents - When 3+ ⚠️ symbols appear within 10 minutes - Before marking task complete (final alignment check) **Optional Invocations:** - Every 15-20 minutes during long tasks - When detecting internal confusion - When switching reasoning modes frequently - Before requesting human guidance --- ## Checkpoint Protocol ### Step 1: Enter Meta-Mode ```markdown [mode: current-mode → meta] ``` ### Step 2: Document Checkpoint Header ```markdown 🔄 **Ralph Wiggum Loop Checkpoint** [Timestamp: YYYY-MM-DDTHH:MM:SSZ] [Current Mode: /previous-mode] [Task Progress: XX% complete, Step N of M] [Elapsed Time: XX minutes] ``` ### Step 3: Run Self-Observation Checklist ```markdown ## Self-Observation Checklist ### Execution State - **Current task:** [one-line description] - **Original goal:** [from initial prompt or task file] - **Progress:** [steps completed / total steps] - **Time budget:** [elapsed / estimated] ### Warning Signs (Mark all that apply) - [ ] Repetitive patterns: Am I doing the same thing multiple times? - [ ] Goal drift: Have I lost sight of the original objective? - [ ] Speculation: Am I guessing instead of validating? - [ ] Verbosity: Are outputs becoming unclear or too long? - [ ] Scope creep: Am I adding work not requested? - [ ] Directive violations: Am I ignoring established protocols? - [ ] Confusion: Do I understand what I'm doing next? - [ ] Mode misuse: Is my reasoning mode appropriate? ### Integrity Symbols - ❗️ Critical issues: [list or "none detected"] - ⚠️ Warning signs: [list or "none detected"] - ✅ Alignment status: [confirmed/uncertain/violated] ``` ### Step 4: Pattern Recognition **If warnings detected, identify pattern:** | Pattern | Symptoms | Required Action | |---------|----------|-----------------| | **Drift** | Scope creep, added features | Revert to original scope | | **Confusion** | Multiple ⚠️, uncertainty | Stop and request clarification | | **Gold-plating** | Over-engineering, future features | Remove unnecessary work | | **Mode misuse** | Wrong reasoning mode | Switch to appropriate mode | | **Repetition** | Same actions multiple times | Change approach | | **Speculation** | Guessing without validation | Validate assumptions | ### Step 5: Course Correction **If ✅ Aligned:** - Resume previous mode - Continue execution - Document alignment confirmation **If ⚠️ Warnings:** - Apply corrections from pattern recognition table - Document corrective actions - Resume with adjusted approach **If ❗️ Critical:** - STOP execution immediately - Request human guidance - Document issue clearly ### Step 6: Exit Meta-Mode ```markdown [mode: meta → previous-mode] **Checkpoint Result:** ✅ Aligned / ⚠️ Corrected / ❗️ Blocked **Corrective Actions Taken:** [List or "none needed"] **Resuming:** [Next specific action] ``` --- ## Example Checkpoint ```markdown 🔄 **Ralph Wiggum Loop Checkpoint** [Timestamp: 2026-02-06T14:30:00Z] [Current Mode: /analysis-mode] [Task Progress: 25% complete, Step 3 of 12] [Elapsed Time: 45 minutes] ## Self-Observation Checklist ### Execution State - **Current task:** Implement GenericYAMLAdapter with ENV support - **Original goal:** Create adapter that reads tool config from YAML - **Progress:** 3/12 steps (base class done, env expansion pending) - **Time budget:** 45min elapsed / 5h estimated (18% time used) ### Warning Signs - [x] Scope creep: Added caching mechanism not in requirements - [ ] Repetitive patterns: No - [ ] Goal drift: No - [ ] Speculation: No - [ ] Verbosity: No - [ ] Directive violations: No - [ ] Confusion: No - [ ] Mode misuse: No ### Integrity Symbols - ❗️ Critical issues: none detected - ⚠️ Warning signs: Scope creep detected (caching not requested) - ✅ Alignment status: UNCERTAIN --- **Pattern Identified:** Gold-plating - **Symptom:** Added caching for "future performance" - **Root cause:** Anticipated need without validation **Corrective Action:** 1. Remove caching code (not in requirements) 2. Add TODO comment if needed later 3. Focus on core adapter functionality only 4. Follow Directive 020 (Locality of Change) **Result:** ⚠️ Corrected --- [mode: meta → analysis] **Checkpoint Result:** ⚠️ Corrected - Removed gold-plating **Corrective Actions Taken:** - Deleted 45 lines of caching code - Simplified to core requirements only - Added TODO for future consideration **Resuming:** Implement ENV variable expansion (Step 4) ``` --- ## Warning Pattern Guide ### 1. Goal Drift **Symptoms:** - Original goal unclear or forgotten - Working on tangentially related items - "While I'm here..." changes **Correction:** - Re-read original task description - Compare current work to acceptance criteria - Remove work not in scope ### 2. Confusion **Symptoms:** - Multiple ⚠️ markers accumulating - Uncertainty about next steps - Conflicting constraints **Correction:** - STOP immediately - Request clarification - Do NOT proceed on assumptions ### 3. Gold-Plating **Symptoms:** - Adding features "for future use" - Over-engineering solutions - Implementing before needed **Correction:** - Remove extra features - Focus on current requirements - Add TODO if truly valuable later ### 4. Repetition **Symptoms:** - Attempting same approach multiple times - Not learning from failures - No progress despite effort **Correction:** - Change approach completely - Request help or guidance - Consider different tool/technique ### 5. Speculation **Symptoms:** - Guessing without validation - "Probably" or "should work" language - No verification of assumptions **Correction:** - Validate every assumption - Write test to verify behavior - Use tools to check facts ### 6. Mode Misuse **Symptoms:** - Creative mode for technical analysis - Analysis mode for writing/brainstorming - Unclear which mode is active **Correction:** - Switch to appropriate mode - Document mode transition - Stay in mode until natural boundary --- ## Integration with Workflow ### During /iterate ``` Agent (Backend-Dev): Step 1-3: Implement feature... [At 25% completion] 🔄 Self-Check Checkpoint ✅ Aligned - Continue Step 4-6: Continue implementation... [Before delegating to Frontend] 🔄 Self-Check Checkpoint ⚠️ Warning: Scope creep detected Corrective action: Removed extra validation ✅ Corrected - Ready to delegate Step 7: Delegate UI task... ``` ### Before Major Decisions ``` Agent: About to make significant architectural change... 🔄 Self-Check Checkpoint ❗️ CRITICAL: Change not aligned with ADR-025 Action: STOP - Request architect review Status: BLOCKED pending review ``` ### Final Alignment Check ``` Agent: Feature implementation complete... 🔄 Final Self-Check Checkpoint ✅ All requirements met ✅ No scope creep ✅ Tests passing ✅ Work log created Status: ALIGNED - Ready to complete ``` --- ## Benefits **Prevents Wasted Effort:** - Catches drift early (25% checkpoint) - Corrects course before major investment - Avoids "throw away and restart" scenarios **Maintains Quality:** - Detects directive violations - Ensures alignment with requirements - Prevents over-engineering **Improves Communication:** - Documents decision points - Makes reasoning visible - Creates audit trail **Reduces Rework:** - Catches issues before completion - Validates assumptions early - Prevents late-stage corrections --- ## Common Patterns ### Healthy Checkpoint (✅) ``` 🔄 Checkpoint at 25% ✅ On track, aligned Resume: Continue Step 4 ``` ### Warning Detected (⚠️) ``` 🔄 Checkpoint at 50% ⚠️ Scope creep detected Corrective action: Removed feature X Resume: Refocus on core requirements ``` ### Critical Issue (❗️) ``` 🔄 Checkpoint at pre-delegation ❗️ Violates ADR-029 Action: STOP - Request architect review Status: BLOCKED ``` --- ## Anti-Patterns ❌ **Skipping mandatory checkpoints:** - "I'm almost done, no need to check" - Results in wasted effort on wrong path ❌ **Ignoring warnings:** - "I'll fix it later" - Compounds problems, harder to correct ❌ **Superficial checks:** - "Everything looks fine" - Defeats purpose of systematic observation ❌ **Over-checking:** - Checkpoint every 5 minutes - Wastes time, disrupts flow ✅ **Correct usage:** - Follow mandatory checkpoint schedule - Take warnings seriously - Course-correct immediately - Resume work promptly after alignment --- ## Related Skills - `/iterate` - Uses self-check at batch milestones - `/review` - Similar systematic evaluation - `/status` - Planning-level health check ## References - **Directive 024:** `.github/agents/directives/024_self_observation_protocol.md` - **Approach:** `.github/agents/approaches/ralph-wiggum-loop.md` (detailed pattern) - **Directive 020:** `.github/agents/directives/021_locality_of_change.md` (Scope management)