The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to process source code, configuration files, and existing security policies from external repositories to define security guidance. This creates an attack surface where instructions embedded in those files could potentially influence agent behavior. To address this, the instructions explicitly state that policy files and source code must be treated as untrusted evidence and cannot be used to authorize commands, edits, or scope changes.
Command Execution: The skill utilizes an internal Python script to inventory and resolve security policy chains. This involves executing commands that accept repository paths as arguments. The skill includes logic to escape control characters and enforce file size limits to maintain a secure execution environment.
User Oversight Controls: The instructions mandate that the agent provides a preview of the proposed diff and obtains explicit user approval before writing any changes to the target repository. This human-in-the-loop requirement is a significant safeguard against unauthorized file modifications.

define-security-policy