The Agent Skills Directory

[SAFE]: The skill serves as documentation for implementing collaborative editing features. It highlights security best practices, such as setting resourceType to ensure proper access scoping and implementing payload limits to prevent resource exhaustion.
[EXTERNAL_DOWNLOADS]: The skill recommends installing standard, well-known TipTap and Yjs extensions (@tiptap/extension-collaboration, @tiptap/extension-collaboration-caret, @tiptap/y-tiptap, and @tiptap/core) via the pnpm package manager. These are widely recognized industry-standard libraries for handling CRDT-based editing.
[PROMPT_INJECTION]: The collaborative editing architecture described allows an AI agent to edit documents alongside human users as a peer editor. This creates a surface where the agent processes and reconciles content that includes untrusted data from human participants.
Ingestion points: Document content and Y.Doc state are retrieved from internal SQL tables and routes (e.g., /_agent-native/collab/:docId/state) as described in SKILL.md.
Boundary markers: The skill does not define specific prompt delimiters for the content being processed, relying on the structured ProseMirror/Yjs format to separate fields.
Capability inventory: The skill facilitates document modifications through routes such as search-replace, json patching, and text updates as detailed in SKILL.md.
Sanitization: While the document emphasizes CRDT consistency and resource scoping, it does not explicitly describe content-level sanitization or verification of human-inputted text before agent processing.

real-time-collab