Agent Skills: Audit Hook Schema

Audit hook event schema for drift. Compares implementation against official docs.

UncategorizedID: melodic-software/claude-code-plugins/hook-schema-audit

Install this agent skill to your local

pnpm dlx add-skill https://github.com/melodic-software/claude-code-plugins/tree/HEAD/plugins/claude-code-observability/skills/hook-schema-audit

Skill Files

Browse the full folder contents for hook-schema-audit.

Download Skill

Loading file tree…

plugins/claude-code-observability/skills/hook-schema-audit/SKILL.md

Skill Metadata

Name
hook-schema-audit
Description
Audit hook event schema for drift. Compares implementation against official docs.

Audit Hook Schema

Detect drift between the hook dispatcher's EVENT_FIELD_REGISTRY and official Claude Code documentation.

Arguments

  • --fix: Generate updated registry code (output only, does not auto-apply)
  • --json: Machine-readable JSON output for CI/automation
  • --verbose: Show all fields, not just differences

Workflow

Step 1: Load Official Documentation

Load the hook-management skill and query for official hook event schemas:

Skill(hook-management)

Also spawn the claude-code-guide agent for live web documentation:

Task(claude-code-guide, "WebFetch https://code.claude.com/docs/en/claude_code_docs_map.md to find hook event documentation pages. Then WebFetch those pages. Extract the input fields for each hook event: PreToolUse, PostToolUse, PostToolUseFailure, PermissionRequest, Notification, UserPromptSubmit, Stop, SubagentStart, SubagentStop, PreCompact, SessionStart, SessionEnd, TeammateIdle, TaskCompleted. Return structured data with event name and field lists.")

Step 2: Parse Current Implementation

Read the EVENT_FIELD_REGISTRY from hook_dispatcher.py:

# Extract the registry definition
grep -A 30 "EVENT_FIELD_REGISTRY = {" plugins/claude-code-observability/hooks/hook_dispatcher.py

Parse the registry into a structured format for comparison.

Step 3: Compare and Report

For each hook event, compare:

  1. Documented fields (from official docs)
  2. Implemented fields (from EVENT_FIELD_REGISTRY)

Generate a report showing:

| Event | Documented | Implemented | Missing | Extra | |-------|------------|-------------|---------|-------| | pretooluse | 5 | 3 | tool_input | - | | notification | 2 | 0 | message, notification_type | - |

Step 4: Output Format

Default output:

Hook Schema Audit Report
========================
Schema Version: 1.7.0
Audit Date: 2026-02-15T15:30:00Z

Coverage Summary:
- Total events: 14
- Fully covered: 11 (79%)
- Partially covered: 2 (14%)
- Missing: 1 (7%)

Findings:
[WARNING] notification: Missing fields: message, notification_type
[WARNING] userpromptsubmit: Missing fields: prompt
[OK] pretooluse: All documented fields present
...

Recommendation: Update EVENT_FIELD_REGISTRY to include missing fields.

With --json:

{
  "schema_version": "1.7.0",
  "audit_timestamp": "2026-02-15T15:30:00Z",
  "coverage": {
    "total_events": 14,
    "fully_covered": 11,
    "partially_covered": 2,
    "missing": 1
  },
  "events": {
    "pretooluse": {
      "status": "ok",
      "documented": ["tool_name", "tool_input", "tool_use_id"],
      "implemented": ["tool_name", "tool_input", "tool_use_id"],
      "missing": [],
      "extra": []
    }
  }
}

With --fix:

Output updated Python code for EVENT_FIELD_REGISTRY that can be copy-pasted into hook_dispatcher.py.

Step 5: Recommendations

Based on findings, provide actionable recommendations:

  • If fields are missing: Suggest adding them to the registry
  • If extra fields found in logs: Investigate if they're new Claude Code fields
  • If schema version outdated: Suggest incrementing SCHEMA_VERSION

Usage Examples

# Basic audit
/claude-code-observability:audit-hook-schema

# CI integration (machine-readable)
/claude-code-observability:audit-hook-schema --json

# Generate fix
/claude-code-observability:audit-hook-schema --fix

# Detailed output
/claude-code-observability:audit-hook-schema --verbose

Exit Codes (for CI)

When using --json:

  • 0: All events fully covered
  • 1: Some fields missing (partial coverage)
  • 2: Critical drift detected (events missing entirely)