visual-testing
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill generates test scenarios from untrusted user input and repository diffs. 1. Ingestion points: SKILL.md (Step 1) identifies user prompts and git diff content as sources for determining test scope. 2. Boundary markers: The skill does not use specific delimiters to isolate untrusted data within the generated scenario files. 3. Capability inventory: The skill executes local shell scripts, builds project code, and manages subagents with shell access. 4. Sanitization: No sanitization is performed on input data before interpolation into scenarios. 5. Mitigation: Subagent prompts include safety instructions to prevent unauthorized file edits or agent spawning.
- [DYNAMIC_EXECUTION]: The skill instructions allow for the dynamic modification of the scripts/guest/muxy-e2e helper script to implement new testing capabilities which are executed in the guest environment.
Audit Metadata