The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The file SKILL.md contains instructions to execute npx --yes ui-skills for various operations. This pattern downloads and executes code directly from the npm registry at runtime without version pinning or integrity checks, representing a significant supply chain risk.
[PROMPT_INJECTION]: The file references/interface-craft.md includes an "Initial Response" section that directs the agent to override its default behavior. It mandates a specific response including a link to an external site (animations.dev) whenever the skill is first invoked, potentially deceiving the user or bypassing standard interaction protocols.
[COMMAND_EXECUTION]: The skill utilizes shell commands through the npx utility to interact with the ui-skills registry, demonstrating the capability to run external binaries based on logic in the instructions.
[EXTERNAL_DOWNLOADS]: The skill facilitates the fetching of resources from docs.ui.sh, Mobbin, and the ui-skills registry.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of untrusted external UI data. 1. Ingestion points: Data is retrieved from uidotsh://ui, docs.ui.sh, Mobbin search results, and the output of the ui-skills CLI tool. 2. Boundary markers: Absent. The instructions do not provide delimiters or safety guidelines to the agent when processing this external content. 3. Capability inventory: The skill can execute shell commands (npx), modify project code, and use the browser for visual verification. 4. Sanitization: Absent. The skill does not specify any sanitization, filtering, or validation for the content retrieved from external UI resources.

ui