Horizon Track

Track long-running objectives that span multiple sessions, days, or weeks.

When to use

When an objective is too large for a single session — multi-week features, research programs, migration projects, or any work that requires persistent progress tracking across conversations.

Steps

1. Initialize horizon — define the objective, target date, and 3-7 milestones
2. Store horizon — call mcp__claude-flow__memory_store with namespace horizons and key horizon-[name]:

   {
     "objective": "...",
     "created": "2026-04-28",
     "targetDate": "2026-05-15",
     "milestones": [
       {"id": "m1", "name": "...", "criteria": "...", "status": "pending"},
       {"id": "m2", "name": "...", "criteria": "...", "status": "pending"}
     ],
     "currentMilestone": "m1",
     "sessions": []
   }

3. Session check-in — at the start of each session:
   • Recall horizon: mcp__claude-flow__memory_retrieve key horizon-[name] namespace horizons
   • Review milestone status
   • Assess drift (are we still on track?)
   • Plan this session's contribution
4. Work and record — as work progresses:
   • Update milestone status
   • Record session summary
   • Store intermediate findings
5. Session check-out — at the end of each session:
   • Update horizon state in memory
   • Record what was accomplished
   • Note blockers or scope changes
   • Estimate remaining effort
6. Milestone completion — when a milestone is done:
   • Verify completion criteria met
   • Store learned patterns via mcp__claude-flow__hooks_intelligence_pattern-store
   • Advance to next milestone
7. Drift detection — flag when:
   • Progress rate suggests target date will be missed
   • Scope has grown beyond original definition
   • Dependencies have changed
   • Approach needs fundamental rethinking

Memory namespaces

• horizons — active horizon definitions and state
• horizon-sessions — per-session summaries keyed by [horizon]-[date]
• horizon-learnings — patterns and insights discovered during the horizon