AgentSkillsCN

System Health Check & Cleanup Skill

系统健康检查与清理技能

SKILL.md

System Health Check & Cleanup Skill

Version: 1.0.0 Category: System Administration / Performance Optimization Author: Claude Code Last Updated: November 22, 2025

Overview

Automated system health monitoring and cleanup workflow. Diagnoses performance issues, identifies resource bottlenecks, fixes orphaned services, kills stale processes, and cleans cache bloat. Designed for archimedes (32c/125GB) but works on any Linux system.

Features

1. System Diagnostics

  • •Memory usage (free, used, buffers, cache)
  • •Disk usage (mounted filesystems)
  • •CPU load average and core count
  • •Process analysis (top CPU/memory consumers)
  • •Service health (running, failed, crashed)
  • •Systemd journal errors

2. Issue Detection

  • •Crash loops: Services repeatedly failing to start
  • •Orphaned services: Missing directories/executables
  • •Process age: Long-running claude/python processes (>15 min)
  • •Cache bloat: Large development caches (pip, go-build, npm)
  • •Failed services: Systemd units in failed state
  • •Log spam: Error entries in journal (potential I/O bottlenecks)

3. Automated Cleanup

  • •Disable crashing services
  • •Kill old claude/python processes
  • •Clean development caches:
    • •pip cache (18GB typical)
    • •go-build cache (3.3GB typical)
    • •Mozilla/Firefox (1.5GB typical)
    • •Playwright browsers (126MB typical)
    • •Electron cache (442MB typical)
    • •Wave Terminal updater (150MB typical)
  • •Vacuum systemd journal
  • •Reload systemd configuration

4. Performance Metrics

  • •Before/after disk space comparison
  • •Memory efficiency improvements
  • •Load average reduction
  • •Process count normalization

Usage

Quick Health Check (No Changes)

bash
/system-health

Output: Comprehensive diagnosis showing:

  • •Memory/disk/CPU status
  • •Top resource consumers
  • •Failed services
  • •Cache sizes
  • •Recommendations

Full Cleanup (Automated Fix)

bash
/system-cleanup

Output:

  • •Disables crashing services
  • •Kills processes >15 min old
  • •Cleans all caches
  • •Optimizes journal
  • •Shows before/after metrics
  • •Confirms all fixes applied

Real-World Example

Scenario: System Glitching After Extended Uptime

Before Cleanup:

code
Load: 1.41 (with crash loop)
Memory: 6.4GB / 125GB used
Disk: 275GB / 915GB (32%)
Cache: 25GB
Processes: 80+ including stale claude instances
Failed Services: backend-api looping every 5 seconds
Journal Size: 1.2GB with error spam

Issues Detected:

  • •❌ backend-api.service fails every 5 seconds (directory missing)
  • •❌ 2 claude processes: 50% CPU, 21% CPU (14+ min old)
  • •❌ 23GB cache bloat (pip, go-build, mozilla)
  • •❌ Log spam from repeated service restarts

After Cleanup:

code
Load: 0.96 (smooth operation)
Memory: 6.4GB / 125GB used
Disk: 250GB / 915GB (29%) — 25GB freed
Cache: 2GB
Processes: 45 (lean system)
Failed Services: None
Journal Size: 100MB (optimized)

Results: ✅ Glitching eliminated, system responsive, 25GB freed

Skill Parameters

Scope

  • •diagnosis-only: Check health without changes
  • •safe-cleanup: Disable services + kill old processes
  • •full-cleanup: Includes cache purging (default)

Output Format

  • •summary: Table format (concise)
  • •detailed: Full command output
  • •json: Machine-readable metrics

Process Age Threshold

  • •Default: 15 minutes
  • •Configurable per run
  • •Only kills claude/python processes (safe)

Technical Details

Service Crash Detection

Monitors for:

  • •Restart=on-failure loops (5 second intervals)
  • •Missing WorkingDirectory or ExecStart paths
  • •systemctl list-units --failed status
  • •journalctl error spam from same PID

Cache Cleanup Strategy

Safe to clean (zero impact):

  • •pip cache (regenerated on first pip command)
  • •go-build cache (rebuilt on compilation)
  • •Mozilla/Firefox cache (recreated automatically)
  • •Playwright browsers (downloaded on first use)
  • •Electron updates (downloaded if needed)

Not touched:

  • •project dependencies (venv/)
  • •user data (Documents, Downloads, Projects)
  • •git repositories
  • •configuration files

Process Cleanup

Kills only:

  • •claude processes >15 min old
  • •Not in use by active terminal
  • •Confirmed via STAT field (Sl+ = active, Sl = background)

Preserves:

  • •System services (systemd, kernel)
  • •Active shells (pts, tty)
  • •Essential daemons (nginx, mysql, etc.)

Integration with Other Skills

With memory-keeper

code
System Health Check → Detects issues
                   → Reports to /update-memory
                   → Triggers cleanup
                   → Saves results to CLAUDE.md

With development workflows

code
Before running long jobs:
  /system-health          (check capacity)
    ↓
After long jobs:
  /system-cleanup         (reclaim resources)
    ↓
Continue development

Output Examples

Diagnostic Output

code
=== SYSTEM HEALTH REPORT ===

Memory:     6.4GB / 125GB (5% used) ✓
Disk:       250GB / 915GB (29%) ✓
CPU Cores:  32 (load avg: 0.96) ✓

TOP PROCESSES:
1. claude (pts/1)      46.7% CPU  - 10 min old  ⚠ OLD
2. claude (pts/3)      33.6% CPU  - 1 min old   ✓
3. firefox             12.7% CPU
4. syncthing           6.5% CPU
5. kitty               3.2% CPU

ISSUES DETECTED:
❌ backend-api.service crash loop (5s restart interval)
❌ claude process PID 3663 is 10+ minutes old
⚠ Cache at 25GB (pip: 18GB, go-build: 3.3GB)
✓ No failed systemd services

RECOMMENDATIONS:
1. Run /system-cleanup to fix all issues
2. Consider increasing cache cleanup frequency
3. Monitor CPU usage if claude processes >50%

Cleanup Summary

code
=== CLEANUP COMPLETE ===

✅ Disabled backend-api.service
✅ Killed 1 old claude process (PID 3663, age 15+ min)
✅ Cleaned pip cache (18GB freed)
✅ Cleaned go-build (3.3GB freed)
✅ Cleaned firefox cache (1.5GB freed)
✅ Cleaned playwright (126MB freed)
✅ Cleaned electron (442MB freed)
✅ Vacuumed journal to 100MB

BEFORE:
  Disk: 275GB (32%)
  Cache: 25GB
  Load: 1.41

AFTER:
  Disk: 250GB (29%) — 25GB freed ✅
  Cache: 2GB — 23GB freed ✅
  Load: 0.96 — smoother ✅

System responsive and optimized!

Troubleshooting

Issue: Permission Denied on Cache Cleanup

Solution: Skill uses askpass helper for sudo auth

  • •Automatically configured via SUDO_ASKPASS
  • •Falls back to interactive prompt if helper fails

Issue: Can't Kill Processes (EPERM)

Solution: Process is owned by root or system

  • •Skill only targets user-owned processes
  • •System services skipped for safety

Issue: Service Disabled But Still Running

Solution: Service may have been started manually

bash
systemctl --user stop backend-api.service
systemctl --user disable backend-api.service

Performance Impact

Runtime

  • •Diagnostic only: ~5 seconds
  • •Process cleanup: ~10 seconds
  • •Full cleanup with caches: ~30-60 seconds

System Impact

  • •Non-blocking (spawns background processes)
  • •I/O intensive during cache deletion (expected)
  • •Safe to run during active development

Parameters & Customization

Command: /system-health

  • •--verbose: Show all processes and services
  • •--format json: Output JSON metrics
  • •--threshold 30: Change process age to 30 minutes

Command: /system-cleanup

  • •--keep-caches: Skip cache cleanup
  • •--keep-processes: Skip process cleanup
  • •--dry-run: Show what would be deleted (no changes)

Known Limitations

  1. •Single machine: Works on archimedes; euclid requires euclid-specific cleanup
  2. •Development-focused: Assumes standard development environment layout
  3. •No backup verification: Deleted caches cannot be recovered (but regenerate automatically)
  4. •Sudo required: Needs passwordless sudo or askpass configured

Future Enhancements

  • • Real-time performance monitoring dashboard
  • • Automatic cleanup scheduling (systemd timer)
  • • Remote monitoring (euclid + vegetius)
  • • Custom cleanup rules (YAML config)
  • • Docker/container support
  • • Trend analysis (track cleanup frequency)

References

System Administration

Archive Tools

  • •rm(1) - Delete files/directories
  • •journalctl - Query systemd journal
  • •systemctl - Manage systemd units

Support

For issues:

  1. •Run /system-health --verbose to diagnose
  2. •Check ~/.local/log/system-health.log for full output
  3. •Verify sudo configuration: sudo -v
  4. •Consult system logs: journalctl -n 50 --no-pager

License: MIT (Part of astoreyai/claude-skills distribution) Repository: https://github.com/astoreyai/claude-skills/ Plugin: system-health-check