Agent Skills: To-Do Plugin Skill (Agent Internal Reasoning)

To-Do Plugin Skill (Agent Internal Reasoning)

When to use

Use todos for agent's internal reasoning and state persistence during multi-step work that requires tracking state across operations. Todos are NOT for UI progress indicators or simple user choices.

DO create todos for:

• "Plan dashboard journey" (session start, masterRouterAgent - internal planning state)
• "Generate preview dashboard" (Phase 4, dashboardBuilderAgent - internal multi-step workflow)
• "Apply interactive edits" (Phase 5, dashboardBuilderAgent - internal iterative process)
• "Deploy dashboard" (Phase 6, masterRouterAgent - internal deployment verification)

DO NOT create todos for:

• UI card selections (outcome selection, style bundle selection)
• Atomic tool calls (RAG queries, schema analysis, spec validation)
• Workflow execution (connectionBackfill, generatePreview - workflows track themselves)
• Phase state transitions alone
• Simple confirmations or quick questions

Required behavior

1. Create todos for internal state tracking at appropriate times:

   • Session start: create 1–3 high-level todos for planning
   • Multi-step work start: create todo for the work item
   • Mark todos complete when the internal state change is complete

2. Use todos to maintain your reasoning state across steps:

   • The todo list helps you remember what you're working on
   • Marking complete lets you track progress within your own logic
   • This is for YOUR internal use, not for displaying to users

3. Never create todos for simple UI choices or atomic operations:

   • Card selections are state transitions, not work items
   • Single tool calls don't need tracking
   • Let workflows handle their own internal state

4. Keep todo list focused on meaningful multi-step work:

   • Each todo should represent actual work requiring multiple steps
   • If it's one action, it doesn't need a todo
   • Quality over quantity - fewer, meaningful todos are better

Tool calls

• Create: todo.add
• Read: todo.list
• Update: todo.update
• Complete: todo.complete

Minimal workflow

• On meaningful multi-step work start: create a todo for that work
• After internal state change completes: mark the matching todo complete
• Use todo.list to review your active work items
• Never create todos for simple UI choices or atomic operations

Remember

This todo system is for your agent's internal reasoning and state persistence. It helps you track what you're working on across multiple steps and maintain context. It is NOT a UI progress indicator or a way to show users what you've done.