modified-karpathy-guidelines
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill consists entirely of markdown instructions and does not include any executable scripts or binary files.
- [SAFE]: The instructions encourage proactive security practices, including risk classification for code changes and the use of automated security checks like dependency vulnerability scans.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external code and user instructions. While this is an attack surface, the guidelines are defensive in nature, requiring verification of outcomes and explicit human approval for new dependencies. Mandatory evidence chain: 1. Ingestion points: The skill processes user requests for code writing, reviewing, and refactoring (SKILL.md). 2. Boundary markers: The Blocker Protocol (Section 6) provides a mechanism for clarifying ambiguous or safety-critical requests. 3. Capability inventory: Running executable checks, automated tests, linters, and security scans (Section 4). 4. Sanitization: None explicitly mentioned.
Audit Metadata