agentic-layer-audit（Agent 层审计）

Agentic Layer Audit Skill

Evaluate a codebase's agentic layer maturity and identify investment opportunities.

When to Use

• Assessing current agentic layer coverage
• Identifying gaps in automation
• Planning agentic layer investments
• Measuring progress toward 50%+ agentic time

Core Concept

"Am I working on the agentic layer or am I working on the application layer?"

This skill helps answer that question by auditing what exists.

Audit Checklist

1. Commands Directory

Check for .claude/commands/ or equivalent:

Look for:
- chore.md      # Chore planning template
- bug.md        # Bug fix template
- feature.md    # Feature planning template
- implement.md  # Implementation HOP
- test.md       # Test execution template
- review.md     # Review template

2. Specs Directory

Check for specs/ or equivalent:

Look for:
- Issue-based specs (issue-*.md)
- Generated plans (chore-*.md, feature-*.md)
- Deep specs (complex multi-file architectures)

3. ADW Directory

Check for adws/ or equivalent:

Look for:
- adw_modules/agent.py  # Core agent execution
- Gateway scripts (adw_prompt.py, adw_slash_command.py)
- Composed workflows (adw_*_*.py)
- Triggers (trigger_*.py)

4. Hooks Directory

Check for .claude/hooks/ or equivalent:

Look for:
- pre_tool_use hooks
- post_tool_use hooks
- user_prompt_submit hooks

5. Agent Output Directory

Check for agents/ or equivalent:

Look for:
- ADW ID directories
- State files (adw_state.json)
- Output files (cc_*.jsonl, cc_*.json)

6. Worktree Support

Check for trees/ or equivalent:

Look for:
- Git worktree setup
- Isolation configuration
- Port allocation patterns

Coverage Scoring

Total: 100 points

Key Memory References

• @agentic-layer-structure.md - What to look for
• @the-guiding-question.md - Why this matters
• @agentic-vs-application.md - Layer separation

Output Format

## Agentic Layer Audit Report

**Project:** {name}
**Audit Date:** {date}
**Coverage Score:** {score}/100

### Components Found
- [x] .claude/commands/ (5 templates)
- [x] specs/ (12 specs)
- [ ] adws/ (not found)
- [ ] hooks/ (not found)

### Maturity Level
{Level} - {Recommendation}

### Gaps Identified
1. No ADW scripts for workflow orchestration
2. No hooks for event-driven automation
3. No worktree isolation for parallelization

### Recommended Investments
1. Create adws/adw_modules/agent.py
2. Add gateway script (adw_prompt.py)
3. Create composed workflow for common tasks

### Time Investment Analysis
- Current: ~20% agentic layer
- Target: 50%+ agentic layer
- Gap: Need 30% more investment in agentic work

Anti-Patterns to Identify

• Commands exist but no specs (templates unused)
• Specs exist but no ADWs (manual execution)
• Many one-off scripts instead of composed workflows
• Application layer dominant (>70% of codebase)

Version History

• v1.0.0 (2025-12-26): Initial release

Last Updated

Date: 2025-12-26 Model: claude-opus-4-5-20251101