System Medic

System Medic Skill

Diagnóstico y reparación automática del entorno Claudebot. Detecta problemas comunes y los resuelve autonomously, aprende de fallos pasados via episodic memory.

When to Use This Skill

Trigger when user:

• Menciona "system medic", "diagnóstico", "reparar entorno", "check salud"
• Dice "ollama caído", "playwright no funciona", "faltan dependencias"
• Reporta errores de infraestructura: ConnectionError, ModuleNotFoundError, playwright not found
• Pide "diagnosticar", "repair", "fix environment", "health check"

Core Workflow

1. Ollama Health Check & Restart

# Check if Ollama is running
python scripts/medic.py check ollama

# If not running, restart it
python scripts/medic.py fix ollama

Detection Logic:

• Try curl http://localhost:11434/api/tags
• If fails → Ollama not responding
• Attempt restart via ollama serve or service restart

Restart Method:

# Windows: Start Ollama service
Start-Process -FilePath "ollama" -ArgumentList "serve" -WindowStyle Hidden

# Or via sc query (service mode)
sc query Ollama

2. Playwright Chromium Installation

# Check if chromium is installed
python scripts/medic.py check playwright

# If missing, install chromium
python scripts/medic.py fix playwright

Detection Logic:

# Try to launch browser
python -c "from playwright.sync_api import sync_playwright; p.sync_playwright().start().chromium.launch()"

# If fails → chromium not installed

Fix:

# Install chromium via patchright (recommended)
python -m patchright install chromium

# Or via playwright
python -m playwright install chromium

3. .env Configuration Validation

# Check if .env files are present
python scripts/medic.py check env

# If missing, report which env files are needed
python scripts/medic.py fix env

Files to Check:

• .env (root)
• .env.local (root)
• credentials.json (if required)

Validation:

import os
from pathlib import Path

required_files = [
    ".env",
    "credentials.json"
]

missing = []
for f in required_files:
    if not Path(f).exists():
        missing.append(f)

Script Reference

Medic Script (medic.py)

# Run all health checks
python scripts/medic.py check all

# Check specific component
python scripts/medic.py check ollama
python scripts/medic.py check playwright
python scripts/medic.py check env

# Fix specific component
python scripts/medic.py fix ollama
python scripts/medic.py fix playwright
python scripts/medic.py fix env

# Auto-fix everything
python scripts/medic.py repair

# Show status report
python scripts/medic.py status

Output Format

{
  "ollama": {
    "status": "running|stopped|error",
    "version": "0.1.2",
    "last_check": "2026-04-19T12:00:00Z"
  },
  "playwright": {
    "chromium": "installed|missing",
    "last_install": "2026-04-18T10:00:00Z"
  },
  "env": {
    ".env": "present|missing",
    "credentials.json": "present|missing"
  }
}

Decision Flow

User invokes System Medic
    ↓
Run medic.py check all
    ↓
┌─────────────────────────────────────────┐
│ Detect issues:                          │
│ - Ollama stopped → fix ollama            │
│ - Chromium missing → fix playwright     │
│ - .env missing → report missing files  │
└─────────────────────────────────────────┘
    ↓
Log findings to episodic_memory
    ↓
Report to user with status
    ↓
If user approves → auto-repair

Learning from Failures

System Medic writes to episodic memory:

• memory/logs/tasks_YYYY-MM-DD.jsonl

Logged Events:

{
  "timestamp": "2026-04-19T12:00:00Z",
  "event": "ollama_restart",
  "success": true,
  "attempt": 2,
  "method": "service_restart"
}

Memory Pattern Recognition:

• Track failures across sessions
• Use preferred fix method per component
• Wait longer before retry if multiple failures

Troubleshooting

Best Practices

1. Always check first - python scripts/medic.py check all
2. Log to memory - Track all repair attempts
3. One fix at a time - Verify after each fix
4. Report status - Show before/after state
5. Ask before repair - Confirm before applying fixes

Limitations

• Windows-only support (PowerShell commands)
• Assumes Ollama at localhost:11434
• No network diagnosis (internet connectivity)
• No database health checks
• No Git repository validation

Resources

• scripts/medic.py - Main diagnostic script
• memory/logs/ - Event logs for learning