entropy-check

SUSPICIOUS. The core grading behavior fits the stated purpose, and data flows mostly go to GitHub/local repo state. However, the skill is not truly read-only: it edits files, pushes branches, opens PRs, creates labels, and files issue comments, which exceeds the description’s "never fix"/sensor framing. Main risk is transitive installation trust plus autonomous GitHub write actions, not overt malware or credential exfiltration.