andrej-karpathy-perspective

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill consists entirely of Markdown documents and a YAML frontmatter definition. No scripts, binaries, or automated shell commands were found in the package.
  • [SAFE]: All remote references target well-known and trusted sources including karpathy.ai, karpathy.github.io, X.com, and the official YouTube channels/podcasts of Andrej Karpathy, Lex Fridman, and Dwarkesh Patel.
  • [PROMPT_INJECTION]: The SKILL.md file contains instructions to override the agent's default persona to adopt the Karpathy perspective. However, these are standard 'perspective' type skill instructions intended for role-play and do not attempt to bypass safety filters or extract sensitive system information. It includes explicit exit triggers ('退出', '不用扮演了') for safety.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest technical queries from users and process them through the 'Karpathy' framework. While it lacks strict input sanitization markers, its low-privilege nature (it does not have file-write or network-send capabilities) makes the risk of indirect injection exploitation negligible.
  • [EXTERNAL_DOWNLOADS]: The README mentions a recommended installation method using npx skills add, which fetches content from a known vendor repository (alchaincyf/karpathy-skill). This aligns with standard package management practices for the platform and targets the author's own namespace.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 01:39 PM
Security Audit — agent-trust-hub — andrej-karpathy-perspective