CopilotKit

Use this skill to implement CopilotKit-based agent UX with correct provider wiring, runtime selection, shared state, frontend actions, and interrupt handling.

Quick Triage

• Use this skill for CopilotKit provider setup, runtime endpoints, hooks, shared-state rendering, and human-in-the-loop UX.
• Switch to $langgraph when backend graph behavior is the main issue.
• Switch to $ag-ui when protocol event compliance is the main issue.
• Switch to $langchain when core model/tool orchestration logic is the main issue.

Workflow

1. Choose runtime topology. Decide LangGraph platform endpoint vs remote backend runtime before UI edits.
2. Configure provider boundary. Set <CopilotKit> provider and runtime endpoint settings first.
3. Select hooks by responsibility. Use useCoAgent for state coupling, useCopilotAction for frontend actions, and interrupt hooks for HITL flow.
4. Define shared state contract. Separate input/output/internal state and keep serialization stable.
5. Implement HITL path. Map interrupt payloads to explicit frontend action handlers.
6. Implement persistence behavior. Choose thread/message persistence strategy and ensure resume consistency.
7. Validate end-to-end UX. Test streaming, tool visibility, shared-state render, and resume flow.

Default Patterns

• Keep runtime and UI concerns separated.
• Keep shared-state schema explicit and minimal.
• Keep action handlers idempotent.
• Treat interrupts as typed control flow, not free-form messages.

Failure Modes

• Endpoint mismatch (wrong runtime URL/type).
• Hooks bound to wrong agent name or provider scope.
• Shared-state updates not reflected due schema mismatch.
• Interrupt action handlers not wired to expected payload type.

Reference Map

Load only what is needed for the current subtask.

• Runtime and provider setup: references/runtime-and-provider.md
• Hooks and actions: references/hooks-and-actions.md
• CoAgents and shared state: references/coagents-shared-state.md
• HITL and troubleshooting: references/hitl-and-troubleshooting.md