The Agent Skills Directory

[SAFE]: The skill serves a legitimate development purpose by providing code reviews based on industry standards and project-specific rules. It includes detailed instructions for scoring and formatting output to ensure consistency.
[COMMAND_EXECUTION]: The skill uses standard, safe development commands such as git diff and git status to analyze code changes. These commands are necessary for its primary function and pose no security threat.
[DATA_EXFILTRATION]: The skill utilizes the mcp-gitlab-usage tool to post feedback on GitLab issues. This behavior is restricted to the specific context of an "Issue review mode" and is documented as part of the intended workflow.
[CREDENTIALS_UNSAFE]: No hardcoded credentials or sensitive data handling issues were found. The skill proactively instructs the agent to check for security vulnerabilities like "SQL injection, XSS, sensitive logs" in the code it reviews, which promotes defensive security practices.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface because it processes untrusted code diffs and issue descriptions. However, it mitigates these risks through a structured output format and clear boundary constraints.
Ingestion points: Analyzes git diff output and external GitLab issues via ISSUE_URL.
Boundary markers: Enforces strict output headers (e.g., "Executive Summary") and mandatory final verdict lines to prevent model confusion.
Capability inventory: Limited to reading files/diffs and posting GitLab comments via specified tools.
Sanitization: Operates in a read-only mode regarding product files, preventing unauthorized modifications.

ns-reviewer