Browse / Second brain / brainprotocol

brainprotocol

giobi

Brain Protocol — open specification for AI agent workspaces

Open giobi/brainprotocol → Does not match the four rules.

aiai-assistantclaudedigital-gardenknowledge-managementnote-takingpkmproductivitysecond-brainzettelkasten

Built for Claude Code

ships .claude and nothing for other agents.

In its own words

README.mdgiobi/brainprotocol

Open specification for structured AI agent workspaces.

Markdown-first, git-versioned, LLM-agnostic, human-readable.

The AI model is replaceable. The brain is not.

What is it

Brain Protocol is a convention — not a framework, not software. It defines how to organize markdown files so any AI agent can read them, write to them, and accumulate knowledge over time.

Your brain is yours: portable, readable, versioned. Works with Claude, Gemini, GPT-4o, Cursor, Aider, or anything that reads files.

Quick Start

Open with your AI agent. It will read boot/soul.md and start the interactive onboarding.

Structure

Every .md file has YAML frontmatter:

Skills

Skills are installable modules: commands, agents, automations, integrations. Install with /brain install .

Skills are protocol-level — they work on any brain, with any LLM. Anyone can publish a skill registry (a GitHub repo, local path, or API endpoint with a manifest.yaml).

This repo includes four core skills:

/brain — skill package manager
/bye — session closing (log, commit, push)
/save — mid-session checkpoint
/devil — devil's advocate mode

Compatibility

Works with any AI agent that reads markdown:

Claude Code (Anthropic)
Gemini CLI (Google)
GPT-4o / ChatGPT
Cursor, Windsurf, Aider
Codex CLI, Mistral, Llama