The Agent Skills Directory

[SAFE]: The skill is a legitimate tool for code quality assurance. It does not perform network operations, exfiltrate data, or request elevated privileges.
[NO_CODE]: The skill contains no executable code (scripts or binaries); it consists entirely of markdown instructions and YAML configuration.
[PROMPT_INJECTION]: The skill processes untrusted external data (local code diffs and documentation files like AGENTS.md), which creates a surface for indirect prompt injection. However, the risk is negligible as the skill lacks access to high-risk tools.
Ingestion points: SKILL.md directs the agent to read staged/unstaged changes and repository-specific instruction files.
Boundary markers: The instructions do not require the agent to use secure delimiters or isolation markers when processing the code diffs.
Capability inventory: The skill is designed only to produce textual reports in the chat interface and does not include tools for network communication or file-system modification outside of the review scope.
Sanitization: No sanitization or filtering of the ingested diff content is prescribed.

review-pr