gmgn-cooking

The SKILL.md raises transparency concerns due to obfuscated or encoded execution paths and dense payload-like content, which meaningfully weaken a reviewer's ability to assess what the skill actually does before loading it. On prompt injection, the skill stays reasonably clear of blurring instruction and data boundaries. However, because behavior analysis was not run and the transparency issues limit reviewability, confidence is low—the scan passed on injection separation, but the obfuscation findings mean I cannot confidently rule out that loading this skill materially changes downstream behavior in ways not apparent from surface review.