The Agent Skills Directory

[PROMPT_INJECTION] (LOW): Indirect Prompt Injection Surface. The skill is designed to process external code review comments from potentially untrusted sources.
Ingestion points: Code review comments from external/untrusted reviewers (triggered by 'Receiving code review comments from any source').
Boundary markers: None explicitly defined in the provided markdown file to separate untrusted feedback from system instructions.
Capability inventory: Access to git commands and the ability to dispatch subagents via a Task tool.
Sanitization: The protocol instructs the agent to 'Verify technically correct' and 'push back if wrong', which serves as a logical check but not a structural sandbox for the prompt content.
[COMMAND_EXECUTION] (LOW): The skill mandates running shell commands for verification (e.g., 'RUN full command', 'git rev-parse').
Context: This is the primary intended purpose of the skill (Verification Gates), designed to ensure code quality through tests and builds.
Risk: While high-privilege, the severity is downgraded to LOW as it is essential for the stated developer-tool functionality and the instructions focus on evidence-based validation rather than arbitrary execution.

code-review