The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill retrieves a component catalog from a remote URL to dynamically define its UI schema.
Evidence: SKILL.md specifies https://unpkg.com/@lynx-js/genui/a2ui/dist/catalog.json as the catalog source.
This download originates from a well-known service and provides official vendor-managed data structures.
[REMOTE_CODE_EXECUTION]: The skill implements safeguards to ensure the agent does not generate or execute arbitrary code.
Evidence: SKILL.md contains strict rules: 'There is no arbitrary code: never emit JavaScript, HTML, CSS, event handlers, scripts, or executable snippets.'
[PROMPT_INJECTION]: The skill processes external catalog data and user actions, which constitutes an indirect prompt injection surface.
Ingestion points: Fetches remote JSON from unpkg.com and processes data from A2UI_USER_ACTION strings in SKILL.md.
Boundary markers: Output is restricted to a strictly structured JSON array format.
Capability inventory: No file system access, subprocess execution, or unauthorized network transmission is performed by the skill.
Sanitization: Relies on the host's A2UI renderer to only instantiate trusted components defined in the catalog.

lynx-a2ui