Optimizing Claude Code

Default: Audit-only. Present findings and await approval before modifying files.

Workflow

1. Run Audit

Execute from repo root:

bash
python scripts/audit_repo.py --root . --include-user-scope

Analyzes:

• Repository scale and directory hotspots
• CLAUDE.md files (discovery, quality, @imports)
• Settings (user/project/local scopes)
• MCP configurations
• Skills and subagents

2. Generate Report

Structure output as:

Executive Summary (5 bullets max)

• Readiness assessment
• Key strengths
• Critical gaps
• Priority actions

Repository Profile

• File count and scale (Small/Medium/Large)
• Primary language/framework

Memory Files Assessment

• Per-file metrics
• @import validation
• Content quality
• Staleness indicators

Settings & Configuration

• Permissions analysis
• MCP server inventory
• Skills/subagents discovered

Issues by Priority

• P0: Blocks agentic work
• P1: High impact
• P2: Nice-to-have

Recommendations

• Specific suggested edits
• Reference best practices
• Do NOT apply without approval

3. Apply Changes (If Requested)

When asked to implement:

1. Confirm which changes
2. Edit incrementally
3. Show diffs
4. Re-audit to verify

Large Repository Guidance

When file count exceeds 3000, include scaling recommendations:

• Vanilla context becomes unreliable at scale
• Consider context engines or retrieval layers
• Strategic CLAUDE.md placement becomes critical
• Focus on navigation aids over exhaustive documentation

Reference Files

• best-practices.md: Environment setup, CLAUDE.md structure, prompting patterns, workflows, safety
• rubric.md: CLAUDE.md scoring criteria (0-16 scale), anti-patterns, templates
• checklist.md: Step-by-step audit workflow, prioritization guide