leverage-point-audit

Leverage Point Audit

Audit a codebase against the 12 leverage points framework to identify gaps and improve agentic coding success.

When to Use

• Before starting a new agentic coding project
• When agents are failing or requiring many attempts
• When KPIs (Size, Attempts, Streak, Presence) are not improving
• For periodic health checks of agentic capability

The 12 Leverage Points

In-Agent (Core Four)

1. Context - CLAUDE.md, README, project docs
2. Model - Appropriate model selection
3. Prompt - Clear instructions and templates
4. Tools - Required capabilities available

Through-Agent (External)

1. Standard Out - Logging for visibility
2. Types - Information Dense Keywords (IDKs)
3. Documentation - Agent-specific context
4. Tests - Self-correction capability (HIGHEST LEVERAGE)
5. Architecture - Navigable codebase structure
6. Plans - Meta-work communication
7. Templates - Reusable prompts (slash commands)
8. ADWs - Autonomous workflows

Audit Workflow

Step 1: Check Context (Leverage Points 1-4)

CLAUDE.md presence:

Search for: CLAUDE.md, .claude/CLAUDE.md
Check: Does it explain the project? Conventions? Common commands?

README.md quality:

Search for: README.md
Check: Does it explain structure? How to run? How to test?

Permissions configuration:

Search for: .claude/settings.json
Check: Are required tools allowed?

Step 2: Check Visibility (Leverage Point 5)

Standard out patterns:

Search for: print(, console.log(, logger., logging.
Check: Are success AND error cases logged?
Check: Can agent see what's happening?

Anti-pattern detection:

Look for: Silent returns, bare except blocks, empty catch blocks
These prevent agent visibility.

Step 3: Check Searchability (Leverage Point 6)

Type definitions:

Search for: interface, type, class, BaseModel, dataclass
Check: Are names information-dense? (Good: UserAuthToken, Bad: Data)

Step 4: Check Documentation (Leverage Point 7)

Internal docs:

Search for: *.md files, docstrings, comments
Check: Do they explain WHY, not just WHAT?

Step 5: Check Validation (Leverage Point 8) - HIGHEST PRIORITY

Test presence:

Search for: test_*.py, *.test.ts, *.spec.ts, *_test.go
Check: Do tests exist? Are they comprehensive?

Test commands:

Check: Is there a simple test command? (npm test, pytest, etc.)
Check: Do tests run quickly?

Step 6: Check Architecture (Leverage Point 9)

Entry points:

Check: Are entry points obvious? (main.py, index.ts, server.py)

File organization:

Check: Consistent structure? Related files grouped?
Check: File sizes reasonable? (< 1000 lines)

Step 7: Check Templates (Leverage Point 11)

Slash commands:

Search for: .claude/commands/
Check: Are common workflows automated?

Step 8: Check ADWs (Leverage Point 12)

Automation:

Search for: GitHub Actions, hooks, triggers
Check: Are workflows automated?

Output Format

After audit, provide:

Summary Table

Leverage Point	Status	Priority	Recommendation
Context	Good/Fair/Poor	High/Med/Low	Specific action
...	...	...	...

Priority Actions

List top 3-5 improvements in order of impact:

1. [Highest Impact] - Specific recommendation
2. [High Impact] - Specific recommendation
3. [Medium Impact] - Specific recommendation

Detailed Findings

For each leverage point:

• Current state
• Specific gaps found
• Recommended improvements
• Example of what good looks like

Example Audit Output

## Leverage Point Audit Results

### Summary
- Tests: POOR (no test files found) - HIGHEST PRIORITY
- Standard Out: FAIR (some logging, missing error cases)
- Architecture: GOOD (clear structure, reasonable file sizes)

### Priority Actions
1. Add test suite - enables self-correction
2. Add error logging to API endpoints - enables visibility
3. Create /prime command - enables quick context

### Detailed Findings
[... specific recommendations ...]

Related Memory Files

• @12-leverage-points.md - Complete framework reference
• @agentic-kpis.md - How to measure improvement
• @agent-perspective-checklist.md - Quick pre-task checklist

Version History

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

---

Last Updated

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