game-asset-generator

Warn

Audited by Gen Agent Trust Hub on Jun 26, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/optimize-glb.mjs executes shell commands via execSync using command-line arguments (inputArg, outputArg) that are not sanitized. This presents a risk of command injection if the agent or user provides malicious filenames containing shell metacharacters.\n
  • Evidence: execSync(gltf-transform dedup "${inputPath}" "${tmpPath}", { stdio: 'inherit' }); in scripts/optimize-glb.mjs.\n- [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes user-provided prompts and transmits them to external APIs (Meshy AI, fal.ai, World Labs) and local scripts without sanitization or protective boundary markers.\n
  • Ingestion points: User requests for 3D models, textures, and sprites.\n
  • Boundary markers: Not implemented.\n
  • Capability inventory: Shell command execution via Bash and optimization scripts; network access for asset generation.\n
  • Sanitization: Not implemented.\n- [EXTERNAL_DOWNLOADS]: The skill downloads assets and configuration files from several well-known and trusted services including Meshy AI (api.meshy.ai), World Labs (api.worldlabs.ai), fal.ai (queue.fal.run), Sketchfab (api.sketchfab.com), Poly Haven (api.polyhaven.com), and Poly.pizza (api.poly.pizza). It also retrieves Draco decoders from Google's GStatic infrastructure.\n- [CREDENTIALS_UNSAFE]: The skill reads multiple API keys (such as MESHY_API_KEY, WLT_API_KEY, and FAL_KEY) from a local .env file. This is a standard and acceptable practice for managing environment-specific secrets in a local context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 26, 2026, 01:41 AM
Security Audit — agent-trust-hub — game-asset-generator