Checkpoint

State Persistence LangGraph-inspired

Enables durable execution through state checkpointing, allowing workflows to resume from any point after interruption, failure, or planned pause.

Overview

The Checkpoint Agent implements a checkpointing system inspired by LangGraph's durable execution model. It enables long-running workflows to persist state at defined points, allowing for:

Fault Recovery: Resume from last checkpoint after crashes or failures
Human-in-the-Loop: Pause for approval, then continue seamlessly
Time-Travel Debugging: Restore any previous state for investigation
Session Continuity: Pick up work across conversation boundaries

Key Concepts

Checkpoint Types

Type	When Created	Use Case
Phase Checkpoint	At workflow phase transitions	Major milestone recovery
Step Checkpoint	Before/after significant operations	Fine-grained recovery
User Checkpoint	Manually triggered by user	Intentional save points
Auto Checkpoint	Periodic (configurable interval)	Background protection

Checkpoint Data

Each checkpoint captures:

Workflow State: Current phase, step, and decision history
Agent Context: Active agent, pending tasks, memory references
File State: Hash of modified files for integrity verification
Metadata: Timestamps, checkpoint ID, parent checkpoint reference

Commands

/checkpoint save

Manually create a checkpoint:

/checkpoint save "Before database migration"

Checkpoint Created: chk_abc123
- Phase: Implementation
- Step: 12 of 25
- Files tracked: 8
- Resume command: /checkpoint restore chk_abc123

/checkpoint list

View available checkpoints:

/checkpoint list

Checkpoints for workflow: wf_xyz789
ID          | Type   | Phase          | Created       | Size
------------|--------|----------------|---------------|-------
chk_abc123  | user   | Implementation | 10 min ago    | 2.4KB
chk_auto_45 | auto   | Implementation | 25 min ago    | 2.1KB
chk_phase_2 | phase  | Architecture   | 1 hour ago    | 1.8KB

/checkpoint restore

Resume from a checkpoint:

/checkpoint restore chk_abc123

Restoring checkpoint chk_abc123...
- Phase: Implementation
- Step: 12 of 25
- Verifying file integrity... OK
- Loading agent context... OK

Ready to continue from: "Implement user authentication"

Integration Points

System	Integration
Conductor	Automatic phase checkpoints at workflow transitions
Time-Travel	Uses checkpoints as replay starting points
Handoff	Checkpoint before agent handoffs for rollback
Memory	Stores checkpoint metadata in memory system

Use Cases

Long Implementation Sessions: Save progress during multi-hour coding sessions
Risky Operations: Checkpoint before migrations, refactors, or security changes
Approval Gates: Pause workflow pending human review, resume after approval
Context Overflow: When context fills, checkpoint and start fresh session
A/B Testing: Checkpoint, try approach A, restore, try approach B

← All Agents Planner Agent →