differential-review
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted content (git diffs and pull requests) as its primary function. While this creates a surface for indirect prompt injection, it is the intended purpose of the tool and is handled safely within that context.\n
- Ingestion points: The skill accepts external diff data via the DIFF_SOURCE input (git diffs, PR URLs, or patch files).\n
- Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are defined in the workflow for the agent to use when processing the diff data.\n
- Capability inventory: The skill utilizes standard system tools like git, semgrep, and package audit utilities (npm-audit, pip-audit, govulncheck).\n
- Sanitization: The instructions do not specify explicit sanitization steps for the input data, but the agent is tasked with analysis rather than direct execution of the content.\n- [DYNAMIC_EXECUTION]: Static analysis hints regarding 'eval' and 'exec' usage are false positives. These keywords appear within the Markdown body as part of a reference list of insecure coding patterns for the agent to search for during a code review (e.g., in Python or JavaScript). The skill does not call these functions itself.\n- [EXTERNAL_DOWNLOADS]: The skill mentions downloading raw .diff or .patch files from pull request URLs. This is a standard operation for the intended code review workflow and does not involve downloading executable scripts from untrusted sources.
Audit Metadata