What does this repo hand its agents?

Every MCP server, skill, plugin, hook and agent instruction file a repository configures (.mcp.json, .claude/, .cursor/, CLAUDE.md and friends), each matched to its verdict and ranked by blast radius. Hidden instructions in CLAUDE.md-type files, installer lures (a README that sends you to a committed ZIP) and credential values pasted into config files are flagged. Hosted MCP servers are probed read-only (initialize and tools/list, never a tool call) so you see whether the endpoint answers, how many tools it exposes, and whether any tool description carries hidden instructions. Names only; no code and no secret values are stored.