code-review-ai-ai-review
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [PROMPT_INJECTION]: The skill takes code diffs via $ARGUMENTS and passes them to AI models for analysis. This creates a risk of indirect prompt injection where malicious code could manipulate the agent's behavior. * Ingestion points: The $ARGUMENTS variable in SKILL.md. * Boundary markers: There are no clear delimiters or instructions telling the AI to ignore instructions within the analyzed code. * Capability inventory: The skill can run shell commands and make network requests to AI services. * Sanitization: No sanitization is performed on the incoming code.
- [COMMAND_EXECUTION]: The skill utilizes several CLI tools like trufflehog, sonar-scanner, and semgrep through shell execution and Python subprocesses. If the input paths or parameters are influenced by $ARGUMENTS without validation, it could lead to command injection.
- [CREDENTIALS_UNSAFE]: The example Python implementation explicitly accesses sensitive secrets like GITHUB_TOKEN and ANTHROPIC_API_KEY from environment variables, which could be exposed if the execution environment is compromised.
Audit Metadata