The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted source code, which provides a surface for indirect prompt injection. This is an inherent risk of the auditing task and does not indicate malicious intent within the skill itself. \n- Ingestion points: Codebase files are accessed through the Read, Grep, and Glob tools. \n- Boundary markers: The skill does not provide delimiters or instructions to treat analyzed code as data distinct from instructions. \n- Capability inventory: The agent possesses the Bash tool, allowing for command execution and code modification (e.g., adding WARNING comments). \n- Sanitization: There is no process for validating or sanitizing the content of audited files before they are processed by the agent.

code-security