POC: Native Task Integration Evaluation

POC: Native Task Integration Evaluation

Status: ✅ Complete Created: 2025-01-23 Related: TASK-34

1. Feature Description

Evaluate Claude Code's native task management tools (TaskCreate, TaskUpdate, TaskGet, TaskList) for potential integration with Navigator's workflow systems.

Current State:

• Claude Code v2.1.16+ provides native task tools
• Native tasks are session-only (ephemeral)
• Navigator uses .agent/tasks/ for persistent documentation
• TodoWrite tool provides in-conversation task lists

Hypothesis: Native tasks could provide better UX for in-session tracking (Loop Mode iterations, Task Mode phases) while .agent/ continues to provide persistence.

2. Implementation Steps

Phase 1: Research & Discovery (1-2 hours)

1. Document native task API

   • Test TaskCreate with sample tasks
   • Test TaskUpdate status transitions
   • Test TaskGet/TaskList retrieval
   • Document dependency tracking (blockedBy/blocks)
   • Identify any UI integration or visibility

2. Compare with TodoWrite

   • What does TodoWrite provide?
   • How do native tasks differ?
   • Can both coexist?

3. Test session persistence

   • Create tasks, compact context, check survival
   • Create tasks, end session, check survival
   • Confirm ephemeral nature

Phase 2: Integration Design (1-2 hours)

4. Map to Navigator workflow

   Loop Mode:
     INIT → TaskCreate("Loop: {goal}", pending)
     Each iteration → TaskUpdate(id, {phase, progress})
     EXIT_SIGNAL → TaskUpdate(id, completed)
   
   Task Mode:
     Detection → TaskCreate("Task: {summary}", pending)
     Phase transition → TaskUpdate(id, {phase: PLAN|IMPL|VERIFY})
     Completion → TaskUpdate(id, completed)
   

5. Design skill integration points

   • nav-loop: Use native tasks for iteration tracking
   • nav-task-mode: Use for phase progression
   • Keep .agent/tasks/ for documentation (unchanged)

Phase 3: Prototype (2-3 hours)

6. Create test skill wrapper

   • Add native task calls to nav-loop
   • Test with real workflow
   • Measure UX improvement

7. Evaluate results

   • Does native task UI add value?
   • Is complexity justified?
   • Any conflicts with WORKFLOW CHECK?

Phase 4: Decision (30 min)

8. Write recommendation
   • Integrate vs Don't integrate
   • If integrate: Full integration plan
   • If don't: Document reasons, close TASK-34

3. Files to Modify (If Integration Proceeds)

4. Expected Outcome

Deliverable: Evaluation report answering:

1. Native Task Capabilities

   • What tools exist (confirmed API)
   • Session vs persistent behavior
   • UI/visibility features

2. TodoWrite Comparison

   • Overlap assessment
   • Use case differentiation

3. Navigator Integration Assessment

   • Value add for Loop Mode
   • Value add for Task Mode
   • Complexity cost analysis

4. Recommendation

   • Integrate / Don't integrate / Partial
   • Rationale with evidence
   • Implementation path (if proceeding)

Decision Criteria

Notes

• TASK-34 exists as backlog item - this POC provides evaluation
• Native tasks confirmed in Claude Code v2.1.16+
• Key distinction: ephemeral (native) vs persistent (Navigator)
• TodoWrite already provides in-conversation tracking

Last Updated: 2025-01-23