The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from code diffs and AGENTS.md files, which creates a potential surface for indirect prompt injection where malicious instructions in the reviewed code could attempt to influence the agent.
Ingestion points: AGENTS.md and the output of git diff commands (SKILL.md).
Boundary markers: The skill does not define specific delimiters to separate the reviewed code from the agent's instructions.
Capability inventory: The skill uses git merge-base and git diff for analysis. It is explicitly prohibited from modifying files, creating commits, or performing network operations (SKILL.md).
Sanitization: No sanitization or escaping of the ingested code content is specified.

review-agent