The Agent Skills Directory

[OBFUSCATION]: Detected URL-encoded segments embedded within image URLs in references/reference.md. The URLs use a nested protocol pattern (e.g., http://https:%2F%2F...) which obscures the target domain context7.com and the file llms.txt from simple pattern matchers.
Evidence: http://https:%2F%2Fcontext7.com%2Fwebsites%2Fshadcn_io%2Fllms.txt/thumbnail.jpg and http://https:%2F%2Fcontext7.com%2Fwebsites%2Fshadcn_io%2Fllms.txt/full-size.jpg in references/reference.md.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it facilitates reading and writing local code files using broad capabilities.
Ingestion points: Uses Read and Glob tools to ingest local project files and component code.
Capability inventory: Possesses Write, Edit, and Bash capabilities which allow for modification of the file system and execution of shell commands.
Sanitization: The instructions do not specify sanitization or validation logic for the content being processed from the local environment.
Boundary markers: No specific delimiters or safety warnings are provided in the implementation examples to prevent the agent from obeying instructions embedded in the processed data.
[COMMAND_EXECUTION]: The skill frequently instructs the agent to execute shell commands for project initialization and component installation (e.g., npx shadcn@latest init, npm install). While these are standard operations for the shadcn/ui library, they involve fetching and executing remote code from package registries.

shadcn-ui