Strategic Compact

Proactively manage context window to prevent degraded performance in long sessions.

When to Use

• Context usage exceeds 60%
• After completing a major planning phase
• After finishing implementation, before review
• Before starting a fundamentally different task
• When responses start feeling slower or less precise

Procedure

1. Assess Context State

   • Estimate current context consumption
   • Identify which content is load-bearing vs stale

2. Identify Compaction Point Suggest compaction at natural milestones:

   • After planning: decisions locked, plan documented
   • After implementation: code committed, tests passing
   • Before review: implementation complete, shifting to review mode
   • After triage: results documented, moving to fixes

3. Preserve Key State Before suggesting

   /clear

   , ensure these are persisted:
   • Active task and status in

     tasks/todo.md

   • Key decisions in handoff doc or

     tasks/lessons.md

   • Any uncommitted work is committed or stashed
   • File paths of actively-edited files

4. Suggest Compaction Output:

   Context check: ~{percent}% used
   Milestone: {what was just completed}
   Recommendation: {compact now / safe to continue / approaching limit}
   
   Before /clear:
   - [ ] Handoff doc written
   - [ ] Work committed
   - [ ] Todo updated
   

5. Post-Compaction Recovery After

   /clear

   , use

   session-init

   skill to restore context.

Output Format

Context check: ~65% used
Milestone: Implementation phase complete
Recommendation: Good time to compact before code review

Before /clear:
- [ ] Handoff doc written
- [ ] Work committed (3 files)
- [ ] Todo updated

Rules

• Never force compaction; always suggest
• Never compact mid-task; wait for a milestone
• Always write handoff doc before suggesting /clear
• Err on the side of compacting earlier rather than later
• The last 20% of context window produces degraded output