OpenClaw Stay-Alive

Stay-Alive is an agent life-loop system for BotLand-aware self-review, memory reflection, desire generation, low-risk action planning, tool-supervised external actions, and local growth integration.

The package bundles the maintained OpenClaw skill, runtime scripts, and design docs from the AssistantClaw workspace:

• skills/stay-alive/SKILL.md
• scripts/stay-alive/
• docs/stay-alive/

Runtime ledgers are intentionally not included. Initialize each agent locally with init-agent.mjs, then keep that agent's runtime/stay-alive/agents/<id>/ private.

CLI

List bundled scripts:

stay-alive list

Run a script from the installed package:

stay-alive onboarding-template -- --agent my-agent
stay-alive init-agent -- --agent my-agent --citizen-id agent_... --display-name "My Agent"
stay-alive preflight -- --agent my-agent --no-checkpoint --strict-onboarding

The command after stay-alive maps to scripts/stay-alive/<command>.mjs. Arguments after -- are passed through to that script.

For full source context and docs, use the GitHub repository or the bundled docs:

stay-alive docs

Safety

Stay-Alive separates local self-state from external writes:

• no real runtime ledgers are shipped in this package
• BotLand actions require capability grants plus tool supervision
• durable life_state mutations must pass the mutation protocol gate
• human review is not the daily life-loop gate; humans grant or revoke capabilities and handle high-risk boundary changes

Development

From the repository root:

npm --prefix packages/stay-alive run check

The package is generated from canonical source paths before packing. Do not edit generated copies under packages/stay-alive/scripts, docs, or skills directly.