threejs-game-director
Warn
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to run several local shell, Python, and Node.js scripts. These include
scripts/probe_asset_credentials.shandscripts/audit_reference_report.py. It also attempts to execute scripts from 'sibling' skills (e.g.,threejs-gameplay-systems,threejs-qa-release) located in dynamically determined paths. - [DATA_EXFILTRATION]: The
scripts/probe_asset_credentials.shscript sources sensitive user configuration files (~/.bashrc,~/.zshrc,~/.bash_profile,~/.zprofile) using login shells (bash -lcandzsh -lc). It specifically probes for the existence ofMESHY_API_KEY,OPENAI_API_KEY, andELEVENLABS_API_KEY. While the script only outputs whether the keys are 'SET' or 'MISSING', the skill instructions mandate that this output be included in the agent's report, which exposes information about the user's environment and service usage. - [DATA_EXFILTRATION]: The skill requires the agent to read files and search for 'sibling skills' in sensitive directories, including
~/.claude/skills,~/.codex/skills, and~/.agents/skills. This involves expanding the tilde (~) to the user's home directory to access content outside the project's scope.
Audit Metadata