generating-semantic-game-mutants
Pass
Audited by Gen Agent Trust Hub on Aug 18, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The test utility
assets/fixtures/check.mjsperforms dynamic loading of JavaScript modules usingimport()on file paths derived from command-line arguments. While this is the intended function of a test runner, it constitutes dynamic code execution from computed paths. - [PROMPT_INJECTION]: The skill's architecture presents an attack surface for indirect prompt injection by processing external data and code as part of its evaluation workflow.
- Ingestion points: Event traces (JSON) in
assets/fixtures/traces.jsonand game reducer logic (JS) inassets/fixtures/mutants/loaded byassets/fixtures/check.mjs. - Boundary markers: None present; the instructions mention isolation but the scripts do not implement delimiters or safety checks for the processed content.
- Capability inventory: The execution environment includes file system access via
readFileSyncand dynamic module execution viaimport()inassets/fixtures/check.mjs. - Sanitization: No validation, escaping, or integrity verification is performed on the traces or reducer logic before they are executed in the simulation.
Audit Metadata