The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends a setup command (npx shadcn@latest add https://raw.githubusercontent.com/Aslam97/shadcn-minimal-tiptap/main/registry/block-registry.json) that fetches and installs configuration files from a GitHub repository belonging to an unverified third-party user (Aslam97). This establishes a dependency on an external, untrusted source.
[REMOTE_CODE_EXECUTION]: The use of npx shadcn add with a remote URL allows for the dynamic generation and addition of code files into a project based on remote JSON definitions. If the remote source is compromised, it can be used to inject malicious code directly into the developer's environment.
[COMMAND_EXECUTION]: The skill instructs users to execute shell commands (npx) that perform network requests and modify local file systems based on external data sources.
[PROMPT_INJECTION]: The skill is designed to process and render rich text and markdown content, creating a surface for indirect prompt injection (Category 8).
Ingestion points: Untrusted data enters the context through EditorContent, editor.commands.setContent(), and editor.commands.insertContent() in references/extensions.md and references/patterns.md.
Boundary markers: No explicit boundary markers or "ignore instructions" warnings are suggested when interpolating content into the editor.
Capability inventory: The skill facilitates subprocess calls via terminal commands and allows for data fetches to local API endpoints (e.g., /api/upload in references/image-upload.md).
Sanitization: The skill relies on Tiptap's default parsing but does not explicitly document custom sanitization or validation of the ingested HTML/Markdown content.

tiptap