method-figure-loop-codex
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: The skill programmatically generates p5.js JavaScript code based on user-provided methods and visual identities. This code is executed within a browser environment to produce visual outputs and vector exports.
- [COMMAND_EXECUTION]: Automated shell commands are used to invoke headless browser instances (Edge/Chrome) for rendering tasks, screenshot capture, and SVG generation. Evidence is found in Phase 3 where specific command-line arguments are provided for headless execution.
- [INDIRECT_PROMPT_INJECTION]: The skill acts as an ingestion point for external data, including method descriptions and composition source images or URLs. This data influences the generated code and the subsequent review by vision-capable sub-agents.
- Ingestion points: User-supplied method text and composition source references.
- Boundary markers: The skill includes specific scope limit blocks intended to constrain sub-agent behavior.
- Capability inventory: The skill possesses capabilities for file manipulation, shell command execution, network fetching, and sub-agent orchestration.
- Sanitization: The process relies on spawning fresh sub-agent instances for review to mitigate context contamination.
- [EXTERNAL_DOWNLOADS]: The skill fetches external dependencies, including the p5.js and p5.js-svg libraries and web-based fonts, to fulfill its rendering requirements. It also references the author's HERO-Anti-OverDefense documentation on GitHub.
Audit Metadata