Skip to content
Get packs

Guide · Agent skill

How to use the Cold Start Brain Bootstrapping agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

Cold Start is an agent skill for GBrain, a personal knowledge store. It walks an AI agent through filling an empty brain with your existing data, one phase at a time, starting with local notes and moving toward contacts, calendar, email, chats and files. This guide explains what the skill does, how to install it, how to trigger it, and what to check before you let it import anything.

Cold Start Brain Bootstrappinggarrytan/gbrain · MIT

Quick start

  1. 1Check GBrain. Confirm GBrain is installed and initialized by running gbrain doctor --json, and clone and sync your brain repository.
  2. 2Install the skill. Download the ZIP, then unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
  3. 3Make the request. Ask the agent something like 'cold start my brain' or 'what should I import first?'.
  4. 4Approve one phase. Begin with existing markdown or Obsidian notes, and give consent before the agent imports anything.
  5. 5Review and stop when ready. Check the pages created, then continue to the next phase or stop. Progress is saved if you need to resume.

What does the cold-start skill make the agent do?

An empty brain is only a database. The skill sequences the data sources most likely to add useful context, so you go from an empty brain to a useful one in a single session. The source ranks sources by information density and ease of import: existing markdown or Obsidian notes come first, then Google Contacts, Google Calendar (90 days), Gmail (recent threads), conversation exports, X/Twitter archives, file archives and meeting transcripts.

Every import phase waits for your consent before it runs. The agent creates and links pages for people, companies, meetings and similar entries as it imports, and it records progress in ~/.gbrain/cold-start-state.json so an interrupted session can resume. You can stop after any phase and still keep what was imported.

  • Asks before each import phase
  • Creates pages under people/, companies/, meetings/, daily/, media/, conversations/ and sources/
  • Tracks progress so a session can resume

How do you install the cold-start skill?

Download the cold-start ZIP from this page. Unzip it into ~/.claude/skills/ for all your projects, or into a project's .claude/skills/ folder to limit it to that project.

If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the ZIP there. The package contains a single file, SKILL.md.

  • Personal install: ~/.claude/skills/
  • Project install: .claude/skills/ inside the project folder
  • Claude apps: Settings → Capabilities → Skills, then upload

What do you need before you run it?

The skill cannot run without the GBrain CLI. Install and initialize GBrain first, then confirm that gbrain doctor --json passes. Clone and sync your brain repository, and make sure the agent has terminal access so it can run gbrain commands.

Two optional inputs are supported: Google Takeout export files, or the gbrain google setup connector. ClawVisor is optional and only works on hosts that have a ClawVisor integration, such as OpenClaw.

  • GBrain installed and initialized
  • Brain repository cloned and synced
  • An agent with terminal access
  • Optional: Google Takeout files or the gbrain google setup connector
  • Optional: ClawVisor, only on supported hosts
gbrain doctor --json

How do you trigger the skill with a natural request?

You don't need a special command. Once the skill is installed, ask the agent in plain language. The skill is designed for the moment just after GBrain setup, when you ask what to do next. These phrases are listed as triggers in the skill file.

Good requests name the goal, for example: 'Cold start my brain', 'Fill my brain with my notes', or 'What should I import first?'. The agent should then propose the first phase and ask for consent before importing.

  • "cold start"
  • "fill my brain"
  • "bootstrap brain"
  • "bootstrap my data"
  • "import my data"
  • "day one" or "get started"
  • "what should I import first"
  • "populate brain" or "now what?"

Which data should you import first?

Start with what is already structured. Existing markdown or Obsidian notes give the highest density of information for the least effort, and the source estimates about 5 minutes for this phase. Next, Google Contacts seeds your people/ directory with names, emails and companies.

For Google data, the source documents the Takeout path first: a Takeout export covers contacts (CSV), calendar (ICS) and Gmail (mbox) as a point-in-time snapshot. The native connector gives live sync instead. Each phase is independent, so you can stop after the first one and keep a useful brain.

  • Phase 1: Existing markdown or Obsidian notes
  • Phase 2: Google Contacts
  • Phase 3: Google Calendar, 90 days
  • Phase 4: Gmail, recent threads
  • Phase 5: ChatGPT or Claude conversation exports
  • Phase 6: X/Twitter archive
  • Phase 7: Dropbox, Drive or local file archives
  • Phase 8: Meeting transcripts

How does the skill keep your credentials safe?

The skill's contract says the agent never holds raw OAuth tokens or API keys. For Google data there are three permitted paths: the native connector (gbrain google setup), where tokens stay in gbrain's credential vault with 0600 file mode; ClawVisor, a hosted gateway that injects credentials at request time and asks for approval on destructive actions; or offline exports such as Google Takeout.

If you decline ClawVisor, the skill does not fall back to direct OAuth. It continues with offline imports or the native connector. The skill declares that it runs scripts and uses the network, so review what the agent runs and keep your credentials out of chat.

  • Runs scripts: yes
  • Uses network: yes
  • Offline exports need no API access
  • ClawVisor needs an agent host with an integration

Tips for a smooth first session

Stop and check the brain after each phase rather than pushing through all eight. Phase 1 alone often gives the agent enough context to be useful. If you use ClawVisor, write broad task purposes; the source says narrow ones such as 'Email triage' get rejected by intent verification, while a full executive-assistant style description works.

The skill writes personal email, contacts and messages into local files. If that is not what you want, skip the email and chat phases.

  • Stop after any phase that gives you enough value
  • Start with offline exports if you want the least exposure
  • Give ClawVisor tasks a broad, specific purpose
  • Review the pages created in each phase before continuing

What people use it for

  • Turning a folder of Obsidian notes into linked people and project pages in your brain
  • Seeding people/ and companies/ pages from a Google Contacts export
  • Adding 90 days of calendar meetings with attendee context from a Takeout ICS file
  • Importing past ChatGPT or Claude conversations to capture your questions and mental models
  • Building a searchable record of your X/Twitter archive and public positions

Frequently asked questions

Is the cold-start skill free to use?

The skill is published under the MIT licence. This page does not list any price for it, and the source does not describe any paid tier. Check the licence file in the repository for the exact terms.

Do I need GBrain installed before I can use this skill?

Yes. The skill requires GBrain installed and initialized, with gbrain doctor --json passing, a cloned and synced brain repository, and an agent that can run gbrain CLI commands. It cannot run without them.

Can the skill import my Gmail without giving an agent my password?

The skill says the agent never holds raw OAuth tokens or API keys. Options are the gbrain google setup connector, which keeps tokens in gbrain's credential vault, ClawVisor on supported hosts, or offline Google Takeout exports. Use offline exports if you want no live account access.

What is the difference between the Takeout route and ClawVisor?

Google Takeout gives you a point-in-time export of contacts, calendar and Gmail that you download yourself. ClawVisor is a hosted credential gateway that gives the agent live, task-scoped access with an audit trail and approval for destructive actions, but it only works on hosts with a ClawVisor integration such as OpenClaw.

What should I do if the import is interrupted?

The skill saves progress in ~/.gbrain/cold-start-state.json so an interrupted session can resume. Ask the agent to continue the cold start, and it can pick up from the saved state.

Can I use this on Codex or Claude Code instead of OpenClaw?

The source says ClawVisor does not work on harnesses such as Codex or Claude Code, because they do not consume its environment variables. On those hosts, use the native gbrain google setup connector or offline exports for Google data, and the local file phases for everything else.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you have just installed GBrain and want a populated brain quickly, especially if you prefer offline exports to live account access. Skip it if you need always-on Gmail or Calendar sync without careful credential setup, if you do not want personal email and messages stored in local files, or if your host cannot run gbrain commands from a terminal.

Download the skill