Historique des audits

Eight findings are false positives involving Markdown formatting, read-only history inspection, citation rules, or explicit secret protections. One finding is confirmed because the skill directs execution of unfamiliar repository tests and lint commands without requiring review or isolation. No evidence found of data exfiltration, destructive intent, or prompt injection.

Eight detections are benign Markdown, read-only history inspection, or protective rules. One low-severity risk remains: repository-defined tests and lint may execute untrusted project code without a trust check. No evidence found of prompt injection, secret exfiltration, or malicious intent.