The Agent Skills Directory

[SAFE]: The skill is composed of markdown-based instructions for the agent's reasoning process. It does not include any executable scripts, binaries, or configuration files that modify the host system or execute commands automatically.- [PROMPT_INJECTION]: The skill is designed to process untrusted external content (source code, plans, and commits). This creates a surface for indirect prompt injection where instructions could be hidden in code comments. However, the skill explicitly mitigates this by instructing the agent to maintain a "professionally adversarial" posture, treat all comments as suspect until verified, and assume the implementation is wrong until proven otherwise.- [DATA_EXFILTRATION]: No network-enabled tools or remote data transfer mechanisms are defined. The skill directs the agent to save its review findings as standalone files within a local reviews/ directory, keeping all data within the repository context.

review