The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple bash scripts located within the project's script directories or plugin paths to resolve backends and model routing. Specifically, it calls cursor-companion.sh to handle the interaction with the review tool.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) by ingesting untrusted code changes from git diff directly into the AI prompt. The ingestion point is in SKILL.md where the PROMPT variable is constructed. Boundary markers and sanitization are absent for this content. The risk is mitigated by the skill's logic, which explicitly instructs the agent to treat this input as advisory and to provide the final verdict independently.

cursor-review