The Agent Skills Directory

[SAFE]: No malicious patterns were identified. The skill performs file system operations (reading and writing) on project configuration files (AGENTS.md, CLAUDE.md) to manage behavioral guidelines as explicitly described in its documentation.
[PROMPT_INJECTION]: The skill is designed to install persistent behavioral instructions into a project. By design, this modifies the agent's core operating constraints for future sessions within that repository. These instructions are sourced from local files provided with the skill, and the skill includes logic to ensure they are integrated correctly without duplicating existing headers.
[EXTERNAL_DOWNLOADS]: The file references/karpathy.md contains a provenance link to a public GitHub repository (https://raw.githubusercontent.com/multica-ai/andrej-karpathy-skills/refs/heads/main/CLAUDE.md) inside an HTML comment. The skill's instructions specifically direct the agent to strip these comments before writing content to the project files, and no actual network requests or downloads are initiated by the skill during its workflow.

project-behavior