three-instanced-crowds
Pass
Audited by Gen Agent Trust Hub on Aug 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The audit script in EVAL.md ingests content from a local file (src/crowd.js) to perform verification checks. This creates an attack surface where instructions embedded in the audited file could potentially influence the auditing logic, though the script uses static regular expression matching. Ingestion point: src/crowd.js via fs.readFile. Boundary markers: Absent. Capability inventory: fs.readFile, page.evaluate. Sanitization: Absent.
- [DYNAMIC_EXECUTION]: The script uses page.evaluate to execute code within a browser context to inspect the Three.js renderer status. This is a functional requirement for validating draw calls and memory management.
- [EXTERNAL_DOWNLOADS]: The skill uses the Three.js library (three@0.169.0), which is a well-known and trusted library for 3D development.
Audit Metadata