The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest and analyze untrusted source code and diffs provided by users, creating an indirect prompt injection surface. A malicious actor could place adversarial instructions inside code comments or string literals within the files being reviewed to influence the agent's behavior. Ingestion points: User-provided code files and diffs (SKILL.md). Boundary markers: The instructions do not specify delimiters or use explicit warnings to ignore natural language instructions found within the code being analyzed. Capability inventory: The skill facilitates file system access and shell execution through tools like Bash, Task, Read, and Grep (SKILL.md). Sanitization: No filtering or sanitization steps are defined for the input data before processing.- [DYNAMIC_EXECUTION]: The skill mentions dangerous patterns such as eval() and exec() in its "Quick Patterns Reference" (SKILL.md). These are provided as examples for the agent to look for when reviewing third-party code and are not intended for execution by the agent itself.

security-review