claude-settings-audit

SUSPICIOUS. The core repository-audit behavior is mostly coherent and read-only, but the skill overreaches by recommending transitive skills and MCP setups. The Sentry MCP path appears same-org and plausible, yet still introduces a middleware trust boundary; the Linear MCP recommendation is materially inconsistent with Linear's official docs and forwards a sensitive API key to an npx-executed package, making the overall skill medium risk.