Checkpoint Skill
Automatically checkpoint current progress to memory-keeper to prevent catastrophic work loss when context is exhausted.
When to Use
- Every 5-10 tool calls during implementation
- After completing a significant piece of work
- Before starting a large operation
- When switching tasks
- Before ending a session
- When explicitly requested via
/checkpoint
Checkpoint Actions
1. Gather Current State
Collect the following information:
- Current task description from todo list
- List of files modified this session
- Implementation progress (percentage or phase)
- Current blockers or issues
- Next action to take
2. Save to Memory-Keeper
context_save(key: "current-task", value: "<task description>", category: "progress", priority: "high")
context_save(key: "files-modified", value: "<comma-separated file list>", category: "progress")
context_save(key: "implementation-progress", value: "<percentage or phase>", category: "progress")
context_save(key: "next-action", value: "<exact next step>", category: "progress", priority: "high")3. Create Named Checkpoint
context_checkpoint(
name: "checkpoint-<timestamp>",
description: "Task: <task>, Progress: <progress>, Files: <count>, Next: <action>"
)4. Prepare for Compaction (if context is large)
context_prepare_compaction()Checkpoint Frequency Guidelines
| Activity | Checkpoint Frequency | |----------|---------------------| | File creation/modification | After every file | | Running tests | After each test run | | Research/exploration | Every 10 tool calls | | Debugging | After each hypothesis tested | | Multi-step implementation | After each step |
Key Items to Always Save
| Key | Description | Priority | |-----|-------------|----------| | current-task | What you're currently working on | high | | files-modified | All files touched this session | normal | | implementation-progress | How far along (%, phase) | normal | | next-action | Exact next step to take | high | | blockers | Current issues/blockers | high | | todo-state | Serialized todo list | normal |
Checkpoint Output
After checkpointing, confirm with:
Checkpoint saved:
- Task: <current task>
- Progress: <progress>
- Files modified: <count>
- Next action: <next step>Recovery Reference
If context is lost, use /recover to restore state from checkpoints.