state-machine-simulation
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns or security vulnerabilities were identified in the skill. The instructions and scripts are focused on legitimate visualization and modeling tasks.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill processes user-supplied behavioral models to generate local visualization files. It does not interact with sensitive system files (e.g., SSH keys, credentials) or perform network operations.
- [REMOTE_CODE_EXECUTION]: The workflow does not involve downloading or executing external scripts. Generated HTML artifacts are self-contained and explicitly prohibited from requiring external fonts, assets, or network access.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input (target behavior). 1. Ingestion points: Target behavior provided as prose or code. 2. Boundary markers: The completion report distinguishes source-grounded behavior from scenario choices. 3. Capability inventory: Local file read/write access for artifacts. 4. Sanitization: The specification mandates using safe DOM APIs (e.g.,
textContent) and strictly forbidseval()or script injection for target-derived data. - [DYNAMIC_EXECUTION]: While the skill produces interactive apps, these use a restricted declarative interpreter for the simulation. The instructions prevent the execution of target code by forbidding dynamic execution sinks like
eval(),Function(), or inline event-handler attributes.
Audit Metadata