adversarial-branch-review

Pass

Audited by Gen Agent Trust Hub on Jul 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • Command Execution: The skill utilizes standard git commands to identify differences between code branches. This is a routine and expected operation for performing a technical branch review.
  • Indirect Prompt Injection Surface: As a code review utility, the skill processes C and C++ source files which are external to the agent's instructions. In theory, adversarial content within these files could attempt to influence the agent's behavior. However, the skill includes rigorous behavioral constraints and a mandatory coverage ledger to ensure the agent remains focused on technical defects rather than embedded instructions.
  • Local Data Access: The instructions require the agent to read file contents and branch metadata. This access is scoped to the project environment and is necessary to perform the requested senior-level systems engineering analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 10, 2026, 02:59 PM
Security Audit — agent-trust-hub — adversarial-branch-review