The Agent Skills Directory

[SAFE]: The skill provides structured instructions for performing automated code reviews using standard tools like git and gh (GitHub CLI). It includes comprehensive checklists for security vulnerabilities such as SQL injection, XSS, and hardcoded secrets.
[COMMAND_EXECUTION]: Uses common development tools (git diff, gh pr diff, cat, read) to retrieve code for analysis. These are read-only operations necessary for the skill's primary function and originate from a trusted environment.
[DATA_EXFILTRATION]: While the skill reads source code and potentially sensitive files for review, it does not contain instructions to send this data to external or untrusted domains. The intended use case involves reporting findings back to the user or via standard GitHub pull request mechanisms.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests untrusted code from external sources (pull requests and files) for analysis. Ingestion points: gh pr diff, git diff, cat, read (SKILL.md). Boundary markers: Absent. Capability inventory: No dangerous capabilities such as exec, eval, or unauthorized network write-operations are present in the skill's instructions. Sanitization: Absent. The risk is mitigated by the skill's focus on static analysis rather than execution of the code being reviewed.

code-review