Retroactively audit what your AI coding agents already did. Recon surfaces every risky shell command, file write, and API call, then drafts a policy that prevents it going forward.