Enhanced code review and frontend validation with intelligent triggers: Code Review Agent Enhancement: - Added Sequential Thinking MCP integration for complex issues - Triggers on 2+ rejections or 3+ critical issues - New escalation format with root cause analysis - Comprehensive solution strategies with trade-off evaluation - Educational feedback to break rejection cycles - Files: .claude/agents/code-review.md (+308 lines) - Docs: CODE_REVIEW_ST_ENHANCEMENT.md, CODE_REVIEW_ST_TESTING.md Frontend Design Skill Enhancement: - Automatic invocation for ANY UI change - Comprehensive validation checklist (200+ checkpoints) - 8 validation categories (visual, interactive, responsive, a11y, etc.) - 3 validation levels (quick, standard, comprehensive) - Integration with code review workflow - Files: .claude/skills/frontend-design/SKILL.md (+120 lines) - Docs: UI_VALIDATION_CHECKLIST.md (462 lines), AUTOMATIC_VALIDATION_ENHANCEMENT.md (587 lines) Settings Optimization: - Repaired .claude/settings.local.json (fixed m365 pattern) - Reduced permissions from 49 to 33 (33% reduction) - Removed duplicates, sorted alphabetically - Created SETTINGS_PERMISSIONS.md documentation Checkpoint Command Enhancement: - Dual checkpoint system (git + database) - Saves session context to API for cross-machine recall - Includes git metadata in database context - Files: .claude/commands/checkpoint.md (+139 lines) Decision Rationale: - Sequential Thinking MCP breaks rejection cycles by identifying root causes - Automatic frontend validation catches UI issues before code review - Dual checkpoints enable complete project memory across machines - Settings optimization improves maintainability Total: 1,200+ lines of documentation and enhancements Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2.3 KiB
Post-Reboot Testing Instructions
What Was Fixed
Commit: 359c2cf - Fix zombie process accumulation and broken context recall
5 Critical Fixes:
- Reduced periodic save from 1min to 5min (80% reduction)
- Added timeout=5 to all subprocess calls (prevents hangs)
- Removed background spawning (&) from hooks (eliminates orphans)
- Added mutex lock to prevent overlapping executions
- CRITICAL: Added UTF-8 encoding to log functions (enables context saves)
Expected Results:
- Before: 1,010 processes/hour, 3-7 GB RAM/hour
- After: ~151 processes/hour (85% reduction)
- Context recall: NOW WORKING (was completely broken)
Testing Commands
Step 1: Capture Baseline (Immediately After Reboot)
cd D:\ClaudeTools
powershell -ExecutionPolicy Bypass -File monitor_zombies.ps1
Note the TOTAL process count - this is your baseline.
Step 2: Work Normally for 30 Minutes
Just use Claude Code normally. The periodic save will run in the background every 5 minutes.
Step 3: Check Results After 30 Minutes
cd D:\ClaudeTools
powershell -ExecutionPolicy Bypass -File monitor_zombies.ps1
Compare TOTAL counts:
- Old behavior: ~505 new processes in 30min
- Fixed behavior: ~75 new processes in 30min (should see this)
Step 4: Verify Context Saves Are Working
Get-Content D:\ClaudeTools\.claude\periodic-save.log -Tail 20
What to look for:
- [OK] Context saved successfully (ID: ...)
- NO encoding errors (no "charmap" errors)
Step 5: Test Context Recall on Restart
- Close Claude Code window
- Reopen Claude Code in ClaudeTools directory
- Check if context is automatically injected at the start
Expected: You should see a "Previous Context" section automatically appear without needing to ask for it.
Quick Reference
Monitoring Script: monitor_zombies.ps1
Periodic Save Log: .claude\periodic-save.log
Results Log: zombie_test_results.txt (created by monitor script)
Project ID: 3c1bb5549a84735e551afb332ce04947
Success Criteria
✅ Process count increase <100 in 30 minutes (vs. ~505 before) ✅ No encoding errors in periodic-save.log ✅ Context auto-injected on Claude Code restart ✅ Memory usage stable (not growing rapidly)
DELETE THIS FILE after successful testing