hrt-apply-code-review
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill orchestrates the execution of local repository tools to verify code quality and security.
- Evidence includes instructions to run full test suites, linters, and repository-specific security scanners (e.g., Gitleaks, OSV-Scanner, Semgrep).
- Access is limited to tools already configured in the environment (e.g., via scripts in package.json, Makefile, or pre-commit configuration).
- Explicit safety guardrails prevent the downloading or installation of any new software during the review process.
- [INDIRECT_PROMPT_INJECTION]: The agent processes various forms of untrusted input that could potentially contain malicious instructions intended to subvert the review process.
- Ingestion points include code diffs, design documents, task lists, and the text output from automated scanners.
- The skill mitigates this surface by instructing the agent to treat all tool output strictly as data and specifically forbidding the execution of commands recommended by external tools.
- The multi-pass review structure and the 'weak-oracle gate' requirement enforce a critical evaluation of data rather than blind obedience to the input content.
- [SAFE]: The skill demonstrates best practices for agent-led security reviews by integrating specialized security logic (hrt-security-pass) and ensuring clear, human-vetted communication for significant behavioral changes.
Audit Metadata