threejs-qa-release
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill uses a headless browser to inspect a target URL (local or remote), which could lead to indirect prompt injection if the page content, console logs, or global variables contain instructions designed to manipulate the agent's release decision.
- Ingestion points: The
scripts/inspect-threejs-canvas.mjsscript navigates to a URL and capturesconsolemessages,pageerrorevents, and thewindow.__THREE_GAME_DIAGNOSTICS__object from the browser context. - Boundary markers: The skill does not use specific delimiters or instructions to warn the agent about potentially malicious instructions embedded in the browser output.
- Capability inventory: The agent has the capability to execute build scripts, manage local files, and document release risks based on the QA results.
- Sanitization: There is no evidence of filtering or sanitization performed on the console logs or diagnostic data retrieved from the target page.
- [COMMAND_EXECUTION]: The skill includes a Node.js utility,
scripts/inspect-threejs-canvas.mjs, which launches a Chromium browser instance using the Playwright library. This is a standard and legitimate behavior for automated QA and visual verification tools.
Audit Metadata