Workspace context
Keep copies of AGENTS.md, SOUL.md, USER.md, IDENTITY.md, TOOLS.md, task notes, templates, and any project-specific memory that would be hard to recreate.
A private assistant is only reliable if you know how to rebuild it after a broken config, lost machine, or failed deployment.
Use this checklist to protect the small set of files and decisions that make an OpenClaw Telegram assistant personal: workspace context, persona files, Telegram access settings, deployment notes, queues, and recovery steps.
Most recovery pain comes from missing context: what the assistant should know, who may talk to it, which channels are connected, and how the project was deployed.
Keep copies of AGENTS.md, SOUL.md, USER.md, IDENTITY.md, TOOLS.md, task notes, templates, and any project-specific memory that would be hard to recreate.
Do not paste secrets into docs. Record where bot tokens, model keys, app passwords, and deployment credentials live, who owns them, and how to rotate them.
Capture the machine, DNS, Vercel/GitHub links, scheduler, queues, and validation commands needed to prove the assistant is working again.
Do a small restore drill before you need it. On a clean machine or separate folder, pull the repo, restore only the documented private files, reconnect credentials from the vault, and verify one end-to-end loop: Telegram message, assistant reply, log evidence, and no accidental public exposure.
The best recovery plan is small enough to use under stress. Back up the files that define the assistant, record where secrets live, and keep a repeatable validation path. If the assistant can be rebuilt without guessing, the setup is much safer to operate.