pixi-vn-testing
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill creates a surface where an AI agent can ingest untrusted data from the game environment.
- Ingestion points: The
window.pixiVN.getState()function returns game dialogue, character information, and player choices directly into the agent's context (SKILL.md §3). - Boundary markers: The skill documentation does not provide instructions for setting up delimiters or warnings to prevent the agent from obeying instructions embedded in game dialogue.
- Capability inventory: The agent is granted significant capabilities including game navigation (
continue,jump,start), storage manipulation (storage.set,storage.clear), and game state resets (Game.clear). - Sanitization: There is no mention of sanitizing or escaping the dialogue content before the agent processes it.
- [COMMAND_EXECUTION]: The skill facilitates the execution of JavaScript commands within a browser environment.
- Evidence: Instructions describe using the Chrome DevTools MCP server's
evaluate_scriptfunction to execute logic against thewindow.pixiVNbridge. - Context: While this is the intended functionality for play-testing and automation, it grants the agent lateral control over the application's runtime environment.
Audit Metadata