context-audit（上下文审计）

Context Audit

Analyzes what's consuming your context window and recommends optimizations. Three audit modes can run independently or together.

Quick Start

• /context-audit or "audit my context" — runs all three audits
• "static audit" or "context inventory" — file inventory only
• "session analysis" — JSONL token parsing only
• "context score" — scoring and recommendations only

Audit Modes

1. Static Inventory

Run scripts/audit-context to automate the static inventory. Supports --json for structured output, --flagged for problems only, --top N for largest items.

The script scans all context-contributing sources:

• Skills (SKILL.md, rules/, references/)
• CLAUDE.md files (global + project + subdirectories)
• Auto-memory files (~/.claude/projects/*/memory/*.md)
• Plugins with per-plugin tool count estimates
• MCP servers

Thresholds: Flag SKILL.md > 500 words, any rules/ directory, CLAUDE.md > 2KB, 5+ MCP servers, plugins with 10+ tools.

2. Live Context Window (/context)

After running the static inventory, tell the user about the built-in /context command:

• It shows real-time token usage: current tokens, max capacity, and percentage used
• It breaks down what's in the context window right now (system prompt, conversation, tool results)
• Recommend the user run /context themselves for live token data — it complements the static inventory
• If the user shares /context output, incorporate it into the scoring (Session Efficiency component)

3. Session Token Analysis

Parse the current session's JSONL to track context growth:

1. Find the active session JSONL in ~/.claude/projects/
2. Extract usage.input_tokens and usage.cache_read_input_tokens per turn
3. Identify the 5 largest token jumps between consecutive turns
4. Correlate jumps with tool calls from preceding turns
5. Report what triggered each spike

Read references/audit-procedures.md for the full JSONL parsing procedure.

4. Recommendations & Scoring

Generate actionable recommendations and a letter grade (A-F, 0-100).

Scoring weights:

Output Format

Produce a single report with sections:

1. Static Inventory table (from audit-context script output)
2. /context note — remind the user to run /context for live token breakdown
3. Session Analysis (if JSONL available)
4. Top Recommendations
5. Score

Read references/audit-procedures.md for detailed procedures, scoring rubric, and recommendation rules.