The Agent Skills Directory

[SAFE]: The skill implements a robust 'Phased Construction' pattern (Design, Build, Animate, Polish) that guides the AI agent through a structured development lifecycle. It includes automatic paradigm detection to ensure that code is generated using the correct architecture (Imperative, R3F, or WebGPU), which reduces logical errors and potential script failures.- [SAFE]: All external libraries and assets referenced in the templates are sourced from established and trusted services, including unpkg.com, jsdelivr.net, and Google's gstatic.com. These are standard practices for front-end development and do not involve untrusted remote code execution.- [SAFE]: The skill uses standard developer tools (Bash, Write, Edit, Task) to manage project files and execute local builds. The instructions emphasize building only what is requested and following repository-specific standards found in CLAUDE.md, demonstrating a focus on security and compliance.- [INDIRECT_PROMPT_INJECTION]: As a code generation tool, the skill is naturally exposed to user-provided instructions which could theoretically attempt to inject malicious code. However, the skill mitigates this risk through a highly structured scene-graph approach and modular code templates that enforce best practices. Ingestion points: User requests for 3D visualizations and scenes; Boundary markers: The 'Scene Plan' template and phased verification gates; Capability inventory: File system access (Write, Edit) and command execution (Bash); Sanitization: Relies on the agent's core safety guardrails as the skill itself does not process external untrusted data beyond the user prompt.

threejs-builder