Workflow Checkpoint

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

Security Scan

The skill's instructions and requirements are consistent with its stated purpose (saving and resuming workflow progress via local checkpoint files), but it relies on unscoped file I/O and offers no built-in protections for sensitive data, so you should verify how and where checkpoints are stored before installing.

Purpose & Capability

The name/description (workflow checkpointing) matches the SKILL.md: it describes writing and reading local JSON checkpoint files, tracking step status, outputs, artifacts, and resuming workflows. No unrelated binaries, credentials, or installs are requested.

Instruction Scope

Instructions explicitly tell the agent to write and read files at memory/checkpoints/<workflow>.json and to verify artifacts at arbitrary filesystem paths (example paths include /tmp and /tmp/myapp). That behavior is expected for checkpointing, but it's open-ended: the skill does not constrain or sanitize what is saved, does not describe access controls, encryption, or path restrictions, and therefore can cause sensitive data to be persisted unintentionally.

Install Mechanism

Instruction-only skill with no install spec and no code files—lowest install risk (nothing is downloaded or written by an installer).

Credentials

The skill requests no environment variables or credentials (appropriate). However, checkpoint files may contain outputs, artifacts paths, or other state that could include secrets; the SKILL.md warns against saving sensitive data but provides no technical controls to enforce this.

Persistence & Privilege

always is false and the skill does not request long-term platform privileges or modify other skills. Autonomous invocation is allowed by default (normal for skills); nothing in the manifest grants it elevated or persistent privileges beyond typical agent runtime file access.

Assessment

This skill appears to implement local checkpointing and is internally consistent, but before installing: 1) Ask the platform where memory/checkpoints is persisted (ephemeral sandbox vs. durable host disk) and who can read those files. 2) Confirm whether checkpoint files are encrypted at rest and whether ACLs or tenancy isolation prevent other agents or users from reading them. 3) Avoid or filter sensitive inputs when using the skill (the SKILL.md warns about this but does not enforce it). 4) Require retention/cleanup policy (old checkpoints can leak data). 5) Test on a non-sensitive workflow first to verify behavior. If you need stronger guarantees, request that the skill be extended to: whitelist allowed paths, redact or encrypt saved outputs, and document recovery validation steps and limits on artifact verification. If the platform cannot guarantee sandboxed file writes and strict access control, treat checkpoint files as potentially readable by others.

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