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:**

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

**README.md quality:**

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

**Permissions configuration:**

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

### Step 2: Check Visibility (Leverage Point 5)

**Standard out patterns:**

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

**Anti-pattern detection:**

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

### Step 3: Check Searchability (Leverage Point 6)

**Type definitions:**

```yaml
Search for: interface, type, class, BaseModel, dataclass
Check: Are names information-de