Use when the session feels sluggish, output quality drops, after adding several skills/agents/MCP servers, or to decide whether there is room to add more — audits what consumes the context window (agents, skills, MCP tools, rules, CLAUDE.md) and returns prioritized token-savings recommendations.
