#Hook Design Pattern Reference

#Hook Event Types

#Exit Code Conventions

exit 0   → Success, continue (completely silent)
exit 2   → Failure, feedback error to Agent, Agent continues to fix
exit other → Failure, no feedback to Agent (non-blocking, silent failure)

Key Principle: Success is always silent. 4000 lines of pass logs cause Agent to lose task focus and start discussing test files instead of completing the task.

#4 Handler Types

Selection Decision Tree:

Need "semantic understanding"?
  → Yes: Use Prompt or Agent type
  → No: Use Command type (faster and cheaper)

Need to read multiple files or execute multiple steps?
  → Yes: Use Agent type
  → No: Use Prompt type

Need to trigger external systems?
  → Yes: Use HTTP type

#Choosing Hooks by Failure Type

#settings.json Configuration Example

{
  "hooks": {
    "Stop": [{
      "matcher": "",
      "hooks": [
        { "type": "command", "command": ".claude/hooks/stop-typecheck.sh" }
      ]
    }],
    "PreToolUse": [{
      "matcher": "Bash|Edit|Write",
      "hooks": [
        { "type": "command", "command": ".claude/hooks/pre-protect-env.sh" }
      ]
    }],
    "PostToolUse": [{
      "matcher": "Edit|Write",
      "hooks": [
        { "type": "command", "command": ".claude/hooks/post-format.sh" }
      ]
    }]
  }
}

#Minimum Viable Hook Set (Day 1)

Top three highest ROI Hooks:

1. Stop Hook — Type Checking: Ensure code is type-correct every time Agent completes
2. PostToolUse Hook — Auto Formatting: Ensure consistent code style, completely silent on success
3. PreToolUse Hook — Protect Sensitive Files: Prevent access to .env, secret, etc.

These three Hooks cover 80% of common quality issues and are the fastest way to establish a Harness.