What problem does it solve? AI coding agents accumulate risky settings over time: permission-bypass flags, wildcard allow rules, risky hooks, unvetted MCP servers, and secrets pasted into session transcripts. This Skill scans that local setup, tells you which findings actually mattered, and applies fixes. ## Core Features & Use Cases - Deterministic security scan: Audits settings, MCP configs, and session transcripts for permission bypasses, wildcard rules, secret-shaped text, and risky shell commands, with a verdict per finding (needs-decision, unsure, test-data, handled). - Session replay and signals: Replays the redacted turns around any flagged line so you can see whether a command actually ran or was just written into a file. - One-click fixes: Applies guard hooks, settings edits, PRs against the session repository, key rotation guidance, or dismissals, and tracks posture changes across scans over time. - Use Case: Ask how safe your Claude Code setup is; the scan finds a wildcard Bash allow rule and an API key pasted in a transcript, replays the context, and offers a guard hook plus a rotation link. ## Quick Start Ask the assistant to run an agent security scan of my local AI-agent setup and show me any findings that need a decision.
