The Agent Skills Directory

[SAFE]: The skill implements a robust mathematical expression parser in its pass1_template.html. This parser uses a recursive-descent approach to tokenize and evaluate user-provided equations, explicitly avoiding dangerous dynamic execution functions like eval() or new Function(). This demonstrates a strong security posture for handling user-defined logic.
[SAFE]: File operations are restricted to reading internal skill documentation and writing the final simulation to a user-specified path (model.html). No attempts were made to access sensitive system directories, hidden configuration files (e.g., .ssh, .aws), or environment variables.
[SAFE]: The skill's architecture relies on a multi-turn 'commissioning' process. It structures user input into a commissioning brief for a separate coding agent, maintaining a clear boundary between model design and infrastructure generation. This process includes mandatory validation steps (Readback and Extraction test) that provide human-in-the-loop oversight.
[SAFE]: External dependencies are limited to well-known technology services. The generated simulation deliverables load React and Recharts via esm.sh, a common and trusted CDN for modern web development. These references are documented neutrally and align with the skill's primary purpose.
[SAFE]: No obfuscation, prompt injection, or persistence mechanisms were found. The skill's instructions are transparent, and its behavior consistently matches its stated purpose of assisting with systems thinking simulations.

ai-stockflow-builder