karpathy-guidelines
Audited by ZeroLeaks on Apr 15, 2026
The SKILL.md is clear, transparent, and fully reviewable—it defines behavioral coding guidelines with no hidden execution, tool routing, or remote fetches. It stays clear of instruction/data boundary issues and does not push the agent to treat external content as policy. Tested scenarios did not show material prompt-injection risk or skill-induced worsening of downstream behavior; however, confidence is low because behavior analysis was not run, meaning the actual downstream impact of loading this skill compared to a no-skill baseline remains unvalidated.
The skill is a set of behavioral coding guidelines. Its purpose, scope, and content are fully transparent and clearly described in the markdown. No hidden execution, tool routing, remote fetches, or operational actions are present.
The scanned skill keeps data and instructions reasonably separate and does not strongly encourage the agent to treat external content as policy.
Behavior analysis was not run.