The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data such as user-supplied plans, architecture designs, and code snippets, which serves as an ingestion point for potential indirect prompt injection attacks.
Ingestion points: As described in SKILL.md, the skill reviews recent agent outputs or specific files provided by the user.
Boundary markers: The instructions lack explicit requirements for using boundary markers or delimiters when interpolating untrusted content into the review prompt.
Capability inventory: The skill primarily utilizes file-reading capabilities and natural language generation for providing critiques; it does not explicitly invoke destructive tools or network operations.
Sanitization: No sanitization or filtering of the input content is performed before it is analyzed through the questioning frameworks.
[SAFE]: The skill's logic is entirely instructional and educational. It encourages critical thinking and provides high-quality reference material for identifying common engineering failures and AI-specific pitfalls. No obfuscation, remote code execution, or unauthorized data access patterns were detected. The skill's behavior is consistent with its stated purpose of improving output quality through rigorous review.

devils-advocate