The Agent Skills Directory

[PROMPT_INJECTION]: The 'Progressive Disclosure System' and 'Keyword-Based Activation' patterns described in reference.md enable skill loading based on conversation context. This architecture creates a surface for indirect prompt injection where untrusted data from a conversation can trigger the activation of specific internal skills.
Ingestion points: User conversation history and specific keywords (e.g., 'authentication', 'user management') as specified in the activation logic within reference.md.
Boundary markers: The documentation does not define delimiters or instructions to ignore embedded commands within the conversation context used for activation.
Capability inventory: The skill metadata in SKILL.md defines requirements for Read, Bash, and Glob tools.
Sanitization: No mechanisms for sanitizing or validating conversation input are described before it triggers skill activation.
[EXTERNAL_DOWNLOADS]: The reference.md file contains references to official documentation and guidelines from well-known technology and academic organizations, including the Google Developer Documentation Style Guide, the OpenAPI Specification, and Stanford University's Knowledge Systems Lab.

moai-cc-skills