context-audit

# Context Audit Skill

Audit a codebase's context engineering health and identify optimization opportunities.

## Purpose

A focused agent is a performant agent. This skill helps you understand what's consuming your context window and where to apply the R&D framework.

## When to Use

- Starting work on a new codebase
- Agent performance feels sluggish
- Context warnings appearing
- Before optimizing context strategy
- Periodic context health checks

## Audit Process

### 1. Memory File Analysis

Scan for CLAUDE.md and related memory files:

```text
Check:
- Root CLAUDE.md size (target: <2KB)
- Number of imports
- Per-directory CLAUDE.md files
- Total memory file tokens
```

Score memory health:

| Size | Score | Assessment |
| --- | --- | --- |
| <1KB | Excellent | Minimal and focused |
| 1-2KB | Good | Within target range |
| 2-5KB | Needs Review | Growing, audit content |
| >5KB | Action Required | Bloated, needs R&D |

### 2. MCP Server Analysis

Check MCP configurations:

```text
Check:
- .mcp.json existence
- Number of MCP servers configured
- Per-server token estimate (2-5% each)
- Active vs unused servers
```

Score MCP health:

| Servers | Score | Assessment |
| --- | --- | --- |
| 0 | Excellent | No MCP bloat |
| 1-2 | Good | Targeted usage |
| 3-5 | Review | May be over-provisioned |
| >5 | Action Required | Likely consuming 15%+ |

### 3. Commands Analysis

Review .claude/commands/:

```text
Check:
- Number of commands
- Command complexity (simple vs complex)
- Priming commands present?
- Task-type coverage
```

Score command health:

| Commands | Score | Assessment |
| --- | --- | --- |
| Has priming | Excellent | Dynamic context loading |
| No priming | Needs Attention | Relying on static memory |

### 4. Hooks Analysis

Check for context-consuming hooks:

```text
Check:
- Number of hooks
- Hook event types
- Potential context injection
```

### 5. Overall Context Score

Calculate overall context engineering score:

| Component | Weight | Max Points |