The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs git commands, including git log and git diff, to inspect the codebase and identify changes. These operations are standard for a code review tool and are performed on the local repository.
[PROMPT_INJECTION]: As the skill is designed to process untrusted data (git diffs and task descriptions), it possesses an inherent attack surface for indirect prompt injection. Malicious code being reviewed could attempt to influence the agent's verdict. However, the instructions provide a structured review protocol that encourages objective analysis, and no malicious overrides were found in the skill's own logic.
[SAFE]: All operations, including repository inspection and task state management through internal APIs (find_tasks, find_documents, manage_task), are consistent with the skill's stated purpose. The skill correctly treats security concerns like hardcoded secrets and SQL injection as items to be detected in the code under review, rather than exhibiting these vulnerabilities itself.

harness-reviewer