AI-powered progress tracking for epic, multi-session tasks in Claude Code. Never lose context. Auto-summarize with Claude Haiku. Continue seamlessly across sessions.
Languages: English | 中文 | 日本語 | 한국어 | Русский | Español
# Add the marketplace
/plugin marketplace add Equality-Machine/claude-marathon
# Install the plugin
/plugin install marathon@marathon
# Done! 🎉Marathon solves one problem: Long tasks lose context when sessions end.
When working on complex tasks that span multiple Claude Code sessions, you lose valuable context:
- What did I do last session?
- What was I about to implement?
- What decisions did I make and why?
Marathon automatically tracks and intelligently summarizes every operation using Claude Haiku:
Before Marathon:
Session 1: Edit 20 files for auth feature ✅
[Context compression happens]
Session 2: "What did I implement yesterday?" ❓
Read through all 20 files manually... 😓
With Marathon:
Session 1: Edit 20 files ✅
→ Auto checkpoint triggers
→ Claude Haiku summarizes: "🎯 Implemented JWT auth system"
[Context compression happens]
Session 2: /marathon-review
→ "Last session: JWT auth 75% complete. Next: Add refresh tokens" ✅
Resume immediately! 🚀
Marathon provides a complete task lifecycle management:
Session Start
↓
📖 /marathon-review ← Restore context from previous sessions
↓
⚙️ /marathon-config ← (Optional) Configure auto-checkpoint threshold
↓
💻 Work on your task...
├─ 🤖 Auto Checkpoint ← Background: AI summarizes every 20 operations
├─ 💾 /marathon-checkpoint ← Manual: Save detailed progress at key moments
└─ 🏁 /marathon-milestone ← Manual: Mark major achievements
↓
✅ /marathon-end ← Complete and archive when task is done
Runs in background automatically - no command needed!
- What: AI batch summarizes recent operations using Claude Haiku
- When: Every N tool calls (default: 20, configurable)
- Cost: < $0.10/month
- Example:
## 📊 Auto Summary (11:03) - [11:02] 📝 Edit: test_manual.txt - [11:03] 🔧 Refactor: auth module (login.js, token.js)
| Command | Purpose | When to Use | Type |
|---|---|---|---|
/marathon-review |
Restore context from previous sessions | Session start - Load past work | Manual |
/marathon-config |
Configure auto-checkpoint threshold | Anytime - Adjust frequency (10-50 calls) | Manual |
| Auto Checkpoint | AI batch summary via Claude Haiku | During work - Every N operations | Automatic |
/marathon-checkpoint |
Save detailed progress report | During work - After key milestones | Manual |
/marathon-milestone |
Mark major achievement | During work - Celebrate important wins | Manual |
/marathon-end |
Complete and archive task | Session end - Final summary & archive | Manual |
Auto vs Manual:
- Auto Checkpoint: Background AI summary, runs automatically every N operations
- Manual Checkpoint (
/marathon-checkpoint): Detailed progress save with your analysis - Milestone (
/marathon-milestone): Celebration marker for major achievements
Day 1: You start task
├─ Marathon creates workspace/session_20260222_*/
├─ Auto-tracks every file edit, git command
└─ After 20 operations → AI summary ✅
Day 3: You continue
├─ /marathon-review → Loads all previous sessions
├─ Shows: What's done? What's next?
└─ You continue with full context ✅
Workspace Structure:
workspace/session_YYYYMMDD_HHMMSS/
├── progress.md # Auto + manual tracking
├── learnings.md # Technical insights
├── session.json # Metadata
└── logs/ # Auto-checkpoint logs
# Working on JWT auth feature...
# (After 20 edits, auto checkpoint triggers)
## 📊 Auto Summary
- [14:10] 🎯 Implement JWT login + token generation
- [14:25] 🧪 Add integration tests
# After major milestone
/marathon-checkpoint
✅ Checkpoint: Login endpoint complete
# When MVP done
/marathon-milestone
🏁 Milestone: JWT Authentication MVP Complete!
# Next session
/marathon-review
🔍 Loaded 3 sessions, 75% complete
🎯 Next: Add token refresh mechanismAdjust auto-checkpoint frequency:
/marathon-configRecommended thresholds:
10-15→ Fast-paced work20-30→ Balanced (default)30-50→ Deep work
✅ Auto-tracking - Set it and forget it ✅ AI-powered - Claude Haiku intelligently summarizes ✅ Cost-effective - <$0.10/month ✅ Zero interruption - Background execution ✅ Multi-project - Directory-aware session filtering ✅ Seamless continuity - Resume exactly where you left off
- Multi-day feature implementations
- Bug investigations spanning sessions
- Research and learning projects
- Team handoffs and context sharing
- Any task longer than one session
- Full Guide: INSTALL.md
- 中文文档: README_ZH.md
- Issues: GitHub Issues
PRs welcome! See CONTRIBUTING.md
MIT License - see LICENSE
Run your marathon with confidence! 🏃♂️💨