Writing Hooks

Writing Hooks

Overview

Hooks intercept events (prompts, permissions, tool calls) to inject behavior. Good hooks are fast, have graceful failures, and define bypass conditions.

When to Create

Create a hook when:

• Behavior should apply to ALL relevant events
• Pre-processing or validation is needed
• Automation can replace manual steps
• Skill activation should be automatic

Don't create for:

• Occasional interventions (use rules)
• Complex logic (use commands or skills)
• User-initiated workflows (use commands)

Structure Template

---
name: hook-name
type: [UserPromptSubmit | PermissionRequest | ToolExecution]
timeout_ms: [number]
---

# Hook Name

## Purpose
[What this hook accomplishes]

## Trigger Event
[Exact event type and conditions]

## Input Format
[JSON structure received]

## Processing Logic
[What the hook does]

## Output Format
[JSON structure returned or modification made]

## Timeout Handling
[What happens if hook exceeds timeout]

## Bypass Conditions
[When hook should be skipped]

## Error Handling
[Graceful failure behavior]

Quality Checklist

• Hook type correctly identified
• Timeout is reasonable (< 5000ms for prompt hooks)
• Input/output JSON formats defined
• Bypass conditions explicit
• Error handling prevents blocking
• Graceful degradation (fails open, not closed)
• No side effects on main context
• Cross-platform if using scripts

Testing Requirements

1. Trigger test - Hook fires on correct events
2. Timeout test - Behavior when timeout exceeded
3. Bypass test - Skipped when conditions met
4. Error test - Graceful failure on exceptions
5. Performance test - Stays within timeout
6. Platform test - Works on Windows/Mac/Linux

Examples

Good Hook (Skill Activation):

---
name: skill-activation
type: UserPromptSubmit
timeout_ms: 3000
---

## Purpose
Analyze prompts and recommend relevant skills.

## Trigger Event
Every UserPromptSubmit event.

## Input Format
{
  "session_id": "string",
  "prompt": "string"
}

## Processing Logic
1. Parse prompt for keywords
2. Match against skill-rules.json
3. Append skill recommendations to prompt

## Output Format
{
  "prompt": "original + skill recommendations"
}

## Timeout Handling
Return original prompt unchanged.

## Bypass Conditions
- Prompt starts with /command
- Prompt is < 5 characters

## Error Handling
Log error, return original prompt.

Good Hook (Permission):

---
name: safe-permissions
type: PermissionRequest
timeout_ms: 5000
---

## Purpose
Auto-approve safe operations, block dangerous ones.

## Processing Logic
Tier 1 (fast approve): Read, Write, Glob, Grep
Tier 2 (fast deny): rm -rf /, force push main
Tier 3 (analyze): Everything else via LLM

## Bypass Conditions
- User in --dangerous mode

Bad Hook:

---
name: helper-hook
---

Does helpful things when stuff happens.

(No type, no timeout, no error handling)

Common Mistakes