The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes external data such as pull requests, git diffs, and session edits (ingestion points). These contents serve as the primary source for model review. While the skill does not explicitly define unique boundary markers for the analyzed code, it employs a multi-model consensus strategy to ensure that findings are independently verified. The agent has the capability to run tests and modify project files (capability inventory), and no specific sanitization of code content is performed prior to analysis.
Command Execution Capability: During the review process, subagents are prompted to 'run focused tests' and 'check diagnostics.' this capability allows the agent to execute shell commands to verify findings and check for regressions in the environment.
Active Code Modification: The skill includes a 'review-and-fix' mode as its default behavior, enabling it to write changes directly to the codebase to address consensus issues. This functionality is intended to streamline the development workflow by resolving identified bugs automatically.

council-review