What problem does it solve? Slow or expensive Claude Code sessions are often caused by machine-level global configuration — user-scope MCP servers, global hooks, pinned reasoning effort — rather than by project code or workflow quality. This Skill systematically audits that global config so cost/latency issues are not misdiagnosed as project problems. ## Core Features & Use Cases - Evidence Collection: Runs claude mcp list/get, reads ~/.claude/settings.json, global CLAUDE.md, and global skills to gather raw diagnostic data. - Red Flag Detection: Identifies user-scoped MCP servers unused by the current project, broad PreToolUse/PostToolUse hooks, duplicate hook matchers, globally pinned high effort levels, and failed MCP connections. - Cost Quantification: Analyzes whether cache read/write dominates session cost, pointing to oversized or repeatedly re-cached base context. - Use Case: After an unusually expensive session, run this audit to discover a user-scope MCP server loading its full tool schema into every project, then demote it to project scope with explicit confirmation. ## Quick Start Ask the agent to run the machine-audit skill to diagnose why recent Claude Code sessions have been slow or expensive.
