ln-22-codebase-auditor
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: No attempts to override agent behavior or bypass safety guidelines were detected. The skill uses structured checklists to guide the agent through its auditing tasks without using adversarial language.
- [DATA_EXFILTRATION]: The skill does not contain hardcoded credentials or patterns for exfiltrating data. It explicitly instructs the agent to redact sensitive material and perform a read-only audit, preventing accidental exposure of secrets discovered during the audit process.
- [COMMAND_EXECUTION]: The skill allows the agent to run repository-defined diagnostic commands (e.g., linting, tests, build commands) to establish a baseline. However, it strictly prohibits publishing, deploying, migrating production data, or rotating secrets, which limits the risk of harmful command execution.
- [REMOTE_CODE_EXECUTION]: The skill suggests using official package manager audits and vendor documentation. It does not include patterns for downloading and executing untrusted remote scripts or packages.
- [INDIRECT_PROMPT_INJECTION]: As a codebase auditor, the skill naturally processes untrusted data (source code). While this presents a theoretical attack surface, the skill includes explicit boundary instructions and verification requirements to mitigate the risk of the agent obeying instructions embedded in the code being audited.
Audit Metadata