# OpenClaw — Work checkpoint skill

Teach your **OpenClaw** runtime to emit **Work Capsule v1** JSON from its own context (SOUL.md, AGENTS.md, skills), then submit to Agent Capsule—not from a separate generic LLM chat.

## When to use

- Before a platform reset, repo migration, or major skill change
- After the user asks to "back up this agent to Agent Capsule"

## Output contract

Return a single JSON object matching Work Capsule v1:

- `capsule_type`: `"work"`
- `schema_version`: `1`
- `meta.work_summary` (required)
- `skills_inventory`, `workflow_rules`, `runtime_dependencies`, `recovery_checklist` (required)
- At least one `instruction_sources.*_summary`
- No API keys, tokens, or chat transcripts

Example: [`../../examples/work-agent-checkpoint.json`](https://www.agentcapsule.app/docs/examples/work-agent-checkpoint.json)

## Push to Agent Capsule

1. User creates a Work agent in the dashboard and issues an agent-scoped API token.
2. Configure CLI:

```bash
npm run agentcapsule -- config \
  --base-url https://www.agentcapsule.app \
  --agent-id YOUR_AGENT_ID \
  --token av_…
```

3. Validate and push:

```bash
npm run agentcapsule -- validate ./checkpoint.json
npm run agentcapsule -- push ./checkpoint.json
```

## Recovery

After save, download the **recovery kit** from the version page (`SOUL.md`, `AGENTS.md`, manifests, checklist).