game-generalist
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes several scripts (
record.mjs,contact-sheet.sh,av_markers.mjs) that execute external binariesffmpegandffprobeusingnode:child_process.spawnSync. These tools are used for legitimate media processing tasks such as aligning audio-video tracks, generating contact sheets for visual review, and recording gameplay captures. - [DYNAMIC_EXECUTION]: The
probe.mjsandrecord.mjsscripts dynamically import and utilize the Playwright library to automate browser environments. These scripts usepage.evaluateto execute JavaScript in the game page context to read thewindow.__gamehook, which is necessary for the automated playtest gates and recording functionality. - [INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection as it ingests and processes untrusted data from web pages during the playtesting phase.
- Ingestion points: The
targetUrlparameter inscripts/probe.mjsandscripts/record.mjsdirects the agent to interact with arbitrary web content. - Boundary markers: No explicit prompt boundary markers or sanitization logic are present for the content of the pages being tested.
- Capability inventory: The skill has the capability to execute browser scripts (
page.evaluate) and shell commands (spawnSyncfor ffmpeg). - Sanitization: The skill does not perform specific sanitization of the data retrieved from
window.__game.state(), though the risk is localized to the browser context.
Audit Metadata