The Agent Skills Directory

[SAFE]: The skill instructions promote security best practices, such as identifying hardcoded credentials and PII in the code being reviewed. No malicious behavior or external dependencies are present.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes user-supplied code, which is a standard surface for indirect prompt injection attacks. 1. Ingestion points: Analyzes notebook and script content provided by the user as part of the review process. 2. Boundary markers: No specific delimiters or safety instructions (e.g., 'treat the following as data only') are defined in the skill for processing the untrusted code. 3. Capability inventory: The skill does not perform file writes, network requests, or command execution; it is restricted to providing a structured natural language review. 4. Sanitization: No input sanitization or filtering of the code content before processing is mentioned.

exp-review