hve-builder-tester
Pass
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- Sandboxed Execution of Artifacts: The skill exercises external artifacts (prompts, instructions, and agents) within a managed sandbox located at
.copilot-tracking/sandbox/. This design pattern is a security consideration that helps isolate the main environment from potential side effects during testing. - Indirect Prompt Injection Surface: As a tool that processes untrusted artifacts, the skill is exposed to indirect prompt injection. However, it implements a significant defense by explicitly instructing the agent to treat all targets and logs as data to be evaluated, rather than instructions to be obeyed.
- Native Execution Capabilities: The skill can execute artifacts natively if specific safety conditions are met, such as the target being read-only or an enforced sandbox being present. While native execution of external code is a potential risk, it is gated by caller approval and containment checks, which align with standard testing best practices.
- Human Oversight Requirement: The skill's success criteria include a requirement for a human reviewer to validate the results, ensuring that the AI agent does not have final authority over the test verdict.
Audit Metadata