Hook Audit Skill | Agent Skills

Reference Files

Advanced hook patterns and best practices:

exit-codes.md - Exit code semantics (0=allow, 2=block) with examples
json-handling.md - Safe JSON stdin parsing patterns
error-patterns.md - Error handling and graceful degradation
performance.md - Timeout guidelines and optimization
examples.md - Good vs bad hook patterns with before/after examples

Hook Audit

Performs comprehensive security and quality audits of Claude Code hooks, ensuring correct JSON handling, exit code semantics, error handling, and performance.

Quick Start

Audit a single hook:

User: "Audit my validate-config.py hook"
Assistant: [Reads hook file, checks patterns, generates report]

Audit all hooks:

User: "Check all my hooks for best practices"
Assistant: [Finds all hooks, analyzes each, generates consolidated report]

Fix specific issues:

User: "My hook is blocking on errors"
Assistant: [Analyzes error handling, suggests fixes]

Hook Audit Checklist

Use this checklist to audit any Claude Code hook:

Critical Requirements

[ ] Shebang Line: Correct interpreter for hook type (see Shebang Standards below)
- Python hooks: #!/usr/bin/env python3
- Bash hooks: #!/bin/bash (NOT #!/usr/bin/env bash or #!/bin/sh)
[ ] JSON stdin Handling: Safe parsing with try/except and .get() methods
[ ] Exit Codes: Correct semantics (0=allow, 2=block, never 1)
[ ] Error Handling: Exit 0 on hook errors (never block on hook failures)
[ ] Settings.json Registration: Hook is registered with correct matcher and timeout

High Priority

[ ] Timeout: Configured appropriately for hook type (PreToolUse <500ms, PostToolUse <2s)
[ ] File Type Validation: Checks file type before processing
[ ] Error Messages: Clear messages to stderr
[ ] Performance: Executes within reasonable time

Medium Priority

[ ] Naming Convention: Uses kebab-case naming
[ ] Documentation: Header comments explain purpose and usage
[ ] Pattern Consistency: Follows existing hook patterns
[ ] Security: No security vulnerabilities or unsafe operations

Audit Workflow

Step 1: Identify Hook Type

Determine the hook type from settings.json registration:

PreToolUse: Runs before tool execution, can block operations
PostToolUse: Runs after successful tool execution
Notification: Runs on specific events (Idle, etc.)
SessionStart: Runs once at session start

Hook type determines performance requirements and exit code handling.

Step 2: Read Hook File and Check Syntax

Read the hook file and verify basic syntax:

# For Python hooks
python3 -m py_compile hook-file.py

# For Bash hooks
bash -n hook-file.sh

Check for:

Correct shebang line
Executable permissions
Valid syntax

Step 3: Verify JSON Handling Pattern

For hooks that receive JSON stdin (PreToolUse, PostToolUse), verify safe parsing:

Python Pattern:

try:
    data = json.load(sys.stdin)
    file_path = data.get("tool_input", {}).get("file_path", "")
    content = data.get("tool_input", {}).get("content", "")
except Exception as e:
    print(f"Error: {e}", file=sys.stderr)
    sys.exit(0)  # Don't block on parsing errors

Bash Pattern:

# Read stdin to variable
stdin_data=$(cat)

# Parse specific fields with jq
file_path=$(echo "$stdin_data" | jq -r '.tool_input.file_path // empty')

Critical: Use .get() with defaults, never direct key access.

Step 4: Check Exit Code Usage

Verify exit codes follow correct semantics:

Exit 0: Allow operation (or hook encountered error)
Exit 2: Block operation (validation failed)
Never Exit 1: Reserved, don't use

Pattern to Check:

# Good: Block on validation failure
if errors:
    print(f"Validation errors:", file=sys.stderr)
    sys.exit(2)

# Good: Allow on hook error
except Exception as e:
    print(f"Hook error: {e}", file=sys.stderr)
    sys.exit(0)  # Don't block user

# Bad: Exit 1 or non-zero on errors
sys.exit(1)  # ✗ Wrong, use 0 or 2

For details on exit code patterns, see exit-codes.md.

Step 5: Review Error Handling

Check that hooks degrade gracefully:

Dependency Check: Missing dependencies exit 0 (don't block)
Try/Except: All operations wrapped in try/except
Error Exit: Exceptions exit 0, not 1 or other codes
Clear Messages: Errors printed to stderr with context

Example Pattern:

try:
    import yaml
except ImportError:
    print("Warning: PyYAML not installed, skipping", file=sys.stderr)
    sys.exit(0)  # Don't block user

try:
    # Hook logic here
    ...
except Exception as e:
    print(f"Error in hook: {e}", file=sys.stderr)
    sys.exit(0)  # Don't block user

For error handling patterns, see error-patterns.md.

Step 6: Validate settings.json Registration

Check that the hook is properly registered:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          {
            "command": "python3 ~/.claude/hooks/validate-config.py",
            "timeout": 5000
          }
        ]
      }
    ]
  }
}

Verify:

Hook file exists at specified path
Matcher pattern is correct for intended triggers
Timeout is appropriate for hook type
Hook type (PreToolUse/PostToolUse/etc.) matches purpose

Step 7: Generate Report

Create a structured audit report following the output format below.

Hook Type Reference

PreToolUse Hooks

Purpose: Validate operations before execution Can Block: Yes (exit 2 to block) Performance Target: <500ms (blocks user interaction) JSON stdin: Tool name and input parameters

Example Matchers:

Edit|Write - Validate before file modifications
Bash - Validate before bash commands
.* - All tools (use cautiously)

Common Uses:

File content validation (YAML, config files)
Security checks (prevent writing secrets)
Policy enforcement

PostToolUse Hooks

Purpose: Process results after successful execution Can Block: No (exit code ignored, operation already completed) Performance Target: <2s (runs after operation) JSON stdin: Tool name, input, and output

Common Uses:

Auto-formatting (gofmt, prettier)
Logging operations
Side effects (notifications, metrics)

Notification Hooks

Purpose: React to events Can Block: No Performance Target: <100ms (quick notifications) JSON stdin: Event-specific data

Common Uses:

Desktop notifications (Idle event)
Metrics tracking
Alert systems

SessionStart Hooks

Purpose: Initialize session state Can Block: No Performance Target: <5s (one-time initialization) JSON stdin: Session metadata

Common Uses:

Load git repository context
Set environment variables
Initialize session state

Common Issues

Missing try/except

Problem: Hook crashes on errors, blocks user

# ✗ Bad: No error handling
data = json.load(sys.stdin)
file_path = data["tool_input"]["file_path"]

Fix: Wrap in try/except, exit 0 on errors

# ✓ Good: Safe error handling
try:
    data = json.load(sys.stdin)
    file_path = data.get("tool_input", {}).get("file_path", "")
except Exception as e:
    print(f"Error: {e}", file=sys.stderr)
    sys.exit(0)

Wrong Exit Codes

Problem: Using exit 1 instead of 0 or 2

# ✗ Bad: Exit 1
if error:
    sys.exit(1)

Fix: Use exit 2 to block, exit 0 to allow

# ✓ Good: Correct exit codes
if validation_failed:
    sys.exit(2)  # Block operation

if hook_error:
    sys.exit(0)  # Don't block user

Blocking on Hook Errors

Problem: Hook exits non-zero on internal errors, blocks user

# ✗ Bad: Blocks user on missing dependency
import yaml  # Crashes if not installed

Fix: Check dependencies, exit 0 on errors

# ✓ Good: Graceful degradation
try:
    import yaml
except ImportError:
    print("Warning: PyYAML not installed", file=sys.stderr)
    sys.exit(0)

Slow Performance

Problem: PreToolUse hook takes >1s, blocks user

Fix: Optimize or move to PostToolUse

Cache expensive operations
Use faster parsing (regex instead of full AST)
Move non-critical checks to PostToolUse

For performance optimization, see performance.md.

Shebang Standards

All hooks MUST use the correct shebang line for their language. Different shebangs are required for different hook types.

Bash Hooks: `#!/bin/bash`

All bash hooks MUST use:

#!/bin/bash

NOT:

#!/bin/sh              # Too restrictive - lacks bash features
#!/usr/bin/env bash    # Unnecessary indirection for hooks

Why #!/bin/bash for hooks:

Bash Features: Hooks may use bash-specific features:
- Arrays for processing multiple items
- [[ conditional expressions for safer string comparisons
- jq integration patterns with process substitution
- Extended pattern matching
Consistency: All hooks should behave identically across environments
Performance: Direct path avoids env lookup overhead
- Matters for PreToolUse hooks on critical path
- Faster startup time (<1ms vs ~10ms)
Reliability: macOS and Linux both guarantee bash at /bin/bash
- Universal location across platforms Claude Code supports
- No PATH dependency issues
POSIX sh Limitations: /bin/sh might be dash, ash, or other minimal shells
- Missing bash arrays, [[ tests, process substitution
- Different behavior across systems

Current hooks using #!/bin/bash (all correct):

auto-format.sh
load-session-context.sh
log-git-commands.sh
notify-idle.sh

Python Hooks: `#!/usr/bin/env python3`

All Python hooks MUST use:

#!/usr/bin/env python3

Why #!/usr/bin/env python3 for Python:

Python Location Varies: Python may be in /usr/bin/, /usr/local/bin/, virtualenv, etc.
Virtual Environments: env respects activated virtualenvs
Version Clarity: python3 is explicit (not python which might be Python 2)
Cross-Platform: Works on systems with non-standard Python installations

Current hooks using #!/usr/bin/env python3 (all correct):

validate-bash-commands.py
validate-config.py
validate-markdown.py

Validation

To verify hooks comply with shebang standards:

Check all bash hooks:

for hook in ~/.claude/hooks/*.sh; do
  shebang=$(head -1 "$hook")
  if [[ "$shebang" != "#!/bin/bash" ]]; then
    echo "ERROR: $hook uses wrong shebang: $shebang"
    echo "  Expected: #!/bin/bash"
  fi
done

Check all Python hooks:

for hook in ~/.claude/hooks/*.py; do
  shebang=$(head -1 "$hook")
  if [[ "$shebang" != "#!/usr/bin/env python3" ]]; then
    echo "ERROR: $hook uses wrong shebang: $shebang"
    echo "  Expected: #!/usr/bin/env python3"
  fi
done

Note on Example Scripts vs Hooks

Example scripts (in skills, not hooks) may use #!/usr/bin/env bash because:

They're intended for distribution across different systems
Users might install them in non-standard locations
Portability is more important than microsecond performance

Hooks are different because:

They live in ~/.claude/hooks/ (fixed location)
They run on every tool invocation (performance matters)
They need bash features (reliability matters)
They don't need portability (only run locally)

Report Format

Generate audit reports in this standardized structure:

# Hook Audit Report: {hook-name}

**Hook**: {name}
**Type**: PreToolUse | PostToolUse | Notification | SessionStart
**File**: {path}
**Audited**: {YYYY-MM-DD HH:MM}

## Summary

{1-2 sentence overview of hook and audit results}

## Compliance Status

**Overall**: PASS | NEEDS WORK | FAIL

- **Shebang Line**: ✓/✗
- **JSON Handling**: ✓/✗
- **Exit Codes**: ✓/✗
- **Error Handling**: ✓/✗
- **Performance**: ✓/✗
- **Registration**: ✓/✗

## Critical Issues

{List of critical failures that must be fixed}

### {Issue Title}

- **Severity**: CRITICAL
- **Location**: {file}:{line}
- **Issue**: {description}
- **Fix**: {specific remediation}

## High Priority Issues

{List of important improvements}

### {Issue Title}

- **Severity**: HIGH
- **Location**: {file}:{line}
- **Issue**: {description}
- **Fix**: {specific remediation}

## Medium Priority Issues

{List of best practice violations}

### {Issue Title}

- **Severity**: MEDIUM
- **Location**: {file}:{line}
- **Issue**: {description}
- **Fix**: {specific remediation}

## Recommendations

1. **Critical**: {must-fix items for safety/correctness}
2. **Important**: {should-fix items for reliability}
3. **Nice-to-Have**: {polish items for maintainability}

## Next Steps

{Specific actions to improve hook quality and safety}

Integration

With Other Auditors

evaluator: General correctness and clarity
test-runner: Functional testing
audit-coordinator: Orchestrates multiple auditors

With Hook Validation

The validate-config.py hook validates YAML frontmatter in agents/skills. The audit-hook validates hooks themselves - a meta-validation.

Examples

For detailed examples of good and bad hook patterns, including before/after refactoring, see examples.md.

Quick Examples:

Good Hook (validate-config.py):

✓ Safe JSON parsing with try/except
✓ Correct exit codes (0 on error, 2 to block)
✓ Dependency checking (PyYAML)
✓ Clear error messages
✓ File type validation

Simple Hook (log-git-commands.sh):

✓ Basic bash pattern
✓ Safe jq parsing
✓ Always exits 0 (informational)
✓ Clear purpose

For complete examples and anti-patterns, see the references above.

Agent Skills: Hook Audit

Skill Files

Reference Files

Hook Audit

Quick Start

Hook Audit Checklist

Critical Requirements

High Priority

Medium Priority

Audit Workflow

Step 1: Identify Hook Type

Step 2: Read Hook File and Check Syntax

Step 3: Verify JSON Handling Pattern

Step 4: Check Exit Code Usage

Step 5: Review Error Handling

Step 6: Validate settings.json Registration

Step 7: Generate Report

Hook Type Reference

PreToolUse Hooks

PostToolUse Hooks

Notification Hooks

SessionStart Hooks

Common Issues

Missing try/except

Wrong Exit Codes

Blocking on Hook Errors

Slow Performance

Shebang Standards

Bash Hooks: `#!/bin/bash`

Python Hooks: `#!/usr/bin/env python3`

Validation

Note on Example Scripts vs Hooks

Report Format

Integration

With Other Auditors

With Hook Validation

Examples

Agent Skills: Hook Audit

Skill Files

Reference Files

Hook Audit

Quick Start

Hook Audit Checklist

Critical Requirements

High Priority

Medium Priority

Audit Workflow

Step 1: Identify Hook Type

Step 2: Read Hook File and Check Syntax

Step 3: Verify JSON Handling Pattern

Step 4: Check Exit Code Usage

Step 5: Review Error Handling

Step 6: Validate settings.json Registration

Step 7: Generate Report

Hook Type Reference

PreToolUse Hooks

PostToolUse Hooks

Notification Hooks

SessionStart Hooks

Common Issues

Missing try/except

Wrong Exit Codes

Blocking on Hook Errors

Slow Performance

Shebang Standards

Bash Hooks: #!/bin/bash

Python Hooks: #!/usr/bin/env python3

Validation

Note on Example Scripts vs Hooks

Report Format

Integration

With Other Auditors

With Hook Validation

Examples

Bash Hooks: `#!/bin/bash`

Python Hooks: `#!/usr/bin/env python3`