workflow-checkpoint（工作流程检查点）

Workflow Checkpoint System 💾

Save and recover from any point in multi-step AI workflows. Never lose progress mid-task.

Why This Matters

AI Agents executing multi-step workflows often fail mid-way:

• Task 3 of 5 fails → all progress lost
• Session restarts → must start from scratch
• Token overflow → workflow interrupted
• Tool errors → uncertain where we left off

This skill eliminates that problem with automatic checkpointing.

How It Works

Core Protocol

For every multi-step workflow:

1. PLAN → Write steps to checkpoint file
2. EXECUTE → After each step, save:
   - Which step completed
   - What output was produced
   - What artifacts were created
   - Current state/data
3. VERIFY → Check step result
4. CHECKPOINT → Update progress file
5. RECOVER → On failure, resume from last checkpoint

Checkpoint File Format

Save to memory/checkpoints/<workflow-name>.json:

{
  "workflow": "deploy-website",
  "startedAt": "2026-04-21T07:30:00Z",
  "totalSteps": 5,
  "completedSteps": [1, 2, 3],
  "currentStep": 4,
  "status": "in_progress",
  "steps": {
    "1": {
      "name": "Clone repository",
      "status": "done",
      "output": "/tmp/myapp cloned successfully",
      "timestamp": "2026-04-21T07:31:00Z"
    },
    "2": {
      "name": "Install dependencies",
      "status": "done",
      "output": "npm install completed",
      "timestamp": "2026-04-21T07:33:00Z"
    },
    "3": {
      "name": "Build project",
      "status": "done",
      "output": "build/ directory created",
      "timestamp": "2026-04-21T07:35:00Z"
    },
    "4": {
      "name": "Deploy to server",
      "status": "failed",
      "error": "Connection timeout",
      "timestamp": "2026-04-21T07:38:00Z"
    },
    "5": {
      "name": "Verify deployment",
      "status": "pending"
    }
  },
  "artifacts": ["/tmp/myapp/build/", "/tmp/myapp/config/"],
  "lastCheckpoint": "2026-04-21T07:38:00Z"
}

Recovery Protocol

When resuming a failed workflow:

1. Read checkpoint file
2. Identify last completed step
3. Skip completed steps (verify artifacts still exist)
4. Resume from failed/pending step
5. Update checkpoint

Usage Examples

Before a complex task:

I'm about to execute a 5-step workflow: deploy-website.
Steps: 1) Clone repo 2) Install deps 3) Build 4) Deploy 5) Verify
Saving checkpoint to memory/checkpoints/deploy-website.json

After each step:

Step 2/5 complete: Install dependencies
Output: npm install completed, 142 packages
Checkpoint updated: completedSteps [1,2]

On failure:

Step 4/5 FAILED: Deploy to server
Error: Connection timeout to 192.168.1.100:22
Checkpoint saved. Can resume from step 4.
Retrying... (attempt 1/3)

On recovery:

Resuming workflow: deploy-website
Last checkpoint: Step 3 completed at 07:35
Skipping steps 1-3 (verified artifacts exist)
Resuming from step 4: Deploy to server

Integration with Other Skills

Works great with:

• EVR - Verify each step before checkpointing
• Error Recovery - Auto-retry failed steps from checkpoint
• Memory Guard - Checkpoints persist across sessions

Anti-Patterns

❌ Don't save checkpoints for single-step tasks ❌ Don't save sensitive data in checkpoint files ❌ Don't skip verification when resuming ❌ Don't forget to clean up old checkpoints

License

MIT