security-diff-scan

Pass

Audited by Gen Agent Trust Hub on Aug 23, 2026

Risk Level: SAFE
Full Analysis
  • [Untrusted Data Handling]: The skill instructs the agent to treat external material and user context as untrusted. This is an essential practice for code review tools to prevent prompt injection from within the analyzed code.
  • [Controlled Resource Access]: Instructions restrict URL access to authorized, one-time reads and prevent recursive fetching. This helps protect the agent from external data exfiltration or SSRF attempts.
  • [Integrated Scanning Workflow]: The skill utilizes scripts located within its plugin directory to prepare file lists and process scan results. This local command execution is part of the core functionality for performing diff-based security analysis.
  • [Verification Procedures]: The requirement for preflight checks and adherence to SECURITY.md guidelines ensures that scans are performed within a validated framework.
  • [Indirect Prompt Injection Surface]:
  • Ingestion points: The skill ingests untrusted data from Git diffs, pull requests, patches, and user-provided context.
  • Boundary markers: Explicit instructions are present to treat external material as untrusted and to limit URL interaction.
  • Capability inventory: The skill uses tools for candidate discovery, scan progress updates, and scan finalization, along with shell execution of local Python scripts.
  • Sanitization: The skill relies on prompt-level instructions to handle untrusted content safely.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 23, 2026, 11:49 PM
Security Audit — agent-trust-hub — security-diff-scan