The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands (open on macOS or xdg-open on Linux) to display the generated HTML files. The file path is derived from a user-influenced filename, which is restricted by instruction to kebab-case to mitigate injection risks.
[EXTERNAL_DOWNLOADS]: The instructions direct the agent to include various libraries (e.g., Bootstrap, Chart.js, D3.js, Motion.js) from cdn.jsdelivr.net. These are well-known technology services and the references are standard for the skill's stated purpose of visualization.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes untrusted user content to generate complex HTML and SVG. It implements a defensive framework through mandatory quality gates and evaluation rubrics:
Ingestion points: User-provided text, ideas, or codebase content (SKILL.md).
Boundary markers: Instruction to use section markers like  (references/generation-protocol.md).
Capability inventory: Shell command execution (open/xdg-open) and network access via CDN links (SKILL.md, references/libraries.md).
Sanitization: Mandatory verification against quality gates (references/diagram-quality.md) and a scoring rubric (references/eval.md).

visualize