stato 0.11.0

Capture, validate, and transfer AI agent expertise.

Stato

The Expertise Layer for AI Agents

Capture, validate, and transfer AI agent expertise.

What Is Stato?

Think: npm for agent expertise, with a built-in compiler.

Stato combines what pip does (install, share, registry) with what a compiler does (validate, reject invalid input) for a new kind of artifact: agent knowledge instead of code.

If you know...	Stato is like...
npm / pip	Install, share, and version expertise packages
TypeScript / GCC	7-pass compiler validates before anything hits disk
Docker	Package expertise so it works on any platform
Git	Snapshot, diff, and merge expertise

What no existing tool does: the agent extracts its own knowledge (crystallization), privacy scanning before export, and a composition algebra for expertise modules.

Compaction is your agent's working memory. Stato is its long-term memory — validated, on disk, portable. Where Mem0/Zep/Letta are services and skills marketplaces distribute single skills, stato distributes whole validated cognitive-state archives (context + plan + memory + skills) that any tool and any teammate can load.

Three ways stato integrates with your agent

A tool needs only one tier to work with stato; each higher tier deepens it.

Tier	What it does	How
Instructions	Static index any tool reads	stato bridge → AGENTS.md, CLAUDE.md, .cursor/rules/*.mdc, copilot, GEMINI.md, SKILL.md
Hooks	Auto-restore state after compaction	stato hooks install → Claude Code, Codex CLI, Gemini CLI
MCP	Live state + validate-gated writes	stato mcp → every MCP client at once

Quickstart per agent

One command wires stato into your agent (it then reads the generated guidance automatically):

Agent	One command
Claude Code	pip install stato && stato bridge --platform claude
Codex CLI	pip install stato && stato bridge --platform agents
Cursor	pip install stato && stato bridge --platform cursor
Gemini CLI	pip install stato && stato bridge --platform gemini
GitHub Copilot	pip install stato && stato bridge --platform copilot
Any MCP client	pip install "stato[mcp]" && stato init --mcp then run stato mcp

Deeper integration (optional): stato skill install --tool <agent> ships the "how to use stato" skill, and stato init --mcp exposes live tools like stato_workspace(task).

Starting from scratch (no .stato/ yet): pip install stato && stato init && stato crystallize, then tell your agent to "read and follow .stato/prompts/crystallize.md", then stato validate .stato/ and stato bridge --platform <agent>.

Install

pip install stato

Install from GitHub (latest development version):

pip install git+https://github.com/genecell/stato.git

Quick Start

Across Sessions (Same Project)

Your agent forgets between sessions. Stato makes its knowledge persist on disk.

# Session 1: agent captures expertise
cd my-project
stato init
stato crystallize
# Saves prompt to .stato/prompts/crystallize.md

# Ask your coding agent to capture its expertise:
#   "Read and follow .stato/prompts/crystallize.md"
# Agent writes .stato/ modules based on what it learned.

# Verify and generate bridge
stato validate .stato/
stato bridge --platform claude

# Session 2 (next day, after /compact, new terminal):
# CLAUDE.md and .stato/ files are still on disk.
# Agent reads CLAUDE.md automatically.
stato resume             # structured recap if needed

No export. No import. Files on disk persist across every session.

Across Projects and People

Transfer expertise to a new project, a teammate, or the community.

# Export
stato snapshot --name scrna-expert --sanitize

# Import into new project
cd ~/new-project && stato init
stato import scrna-expert.stato

# Or install from the community registry
stato registry install genecell/scrna-expert

Composition algebra for working with expertise archives:

Operation	Command	What it does
Snapshot	stato snapshot	Bundle all expertise into a portable archive
Slice	stato slice --module skills/qc	Extract specific skills with dependencies
Graft	stato graft external-skill.py	Add one external skill with validation
Merge	stato merge a.stato b.stato	Combine expertise from multiple sources

Across Platforms

Same expertise, different coding agents. One command generates every bridge:

stato bridge --platform all

# Creates the file each tool reads automatically:
#   AGENTS.md                           -> Codex, and most agents (cross-tool standard)
#   CLAUDE.md                           -> Claude Code
#   .cursor/rules/stato.mdc             -> Cursor
#   .github/copilot-instructions.md     -> GitHub Copilot
#   GEMINI.md                           -> Gemini CLI
#   README.stato.md                     -> anything else / humans

From Web AI to Coding Agent

Plan architecture in Claude.ai or ChatGPT. Build in any coding agent.

stato crystallize --web
# Paste prompt into web AI -> get bundle -> save as stato_bundle.py
stato import-bundle stato_bundle.py
stato bridge --platform cursor

Why Stato Exists

AI coding agents are powerful but stateless. Every session starts from zero. Expertise earned in one session, one project, or one platform stays trapped there.

Stato treats agent expertise like code: captured in structured modules, validated by a 7-pass compiler, composed with algebraic operations, and portable across any platform. Your agent's knowledge becomes a permanent, shareable, validated artifact.

Read the full story ->

Features

Feature	Description
Crystallize	Agent captures its own knowledge into structured modules
7-Pass Compiler	Validates syntax, structure, types, schema, semantics before writing
Composition Algebra	Snapshot, slice, graft, merge expertise archives
Cross-Platform Bridges	AGENTS.md, CLAUDE.md, .cursor/rules/*.mdc, Copilot, GEMINI.md, SKILL.md from one source
Web AI Bridge	Import expertise from Claude.ai, ChatGPT, Gemini conversations
Privacy Scanner	19 patterns detect secrets, emails, paths before export
Resume	Restore full context after /compact or session restart
Convert	Migrate existing CLAUDE.md, .cursorrules, SKILL.md into stato
Registry	Search and install community expertise packages
Diff	Field-level comparison between module versions

CLI Reference

Command	Description
stato init	Initialize a stato project
stato crystallize	Save prompt for agent to capture expertise
stato crystallize --print	Print full crystallize prompt to terminal
stato crystallize --web	Generate prompt optimized for web AI
stato validate	Run 7-pass compiler on modules
stato status	Show all modules, plan progress, warnings
stato bridge	Generate platform bridge files
stato resume	Generate context recap for session restoration
stato diff	Compare module versions
stato snapshot	Export expertise as portable archive
stato import	Import modules from .stato archive
stato import-bundle	Import from web AI bundle file
stato inspect	Preview archive contents
stato slice	Extract specific modules with dependencies
stato graft	Add external module with validation
stato merge	Combine two archives with conflict resolution
stato convert	Migrate from CLAUDE.md, .cursorrules, SKILL.md
stato registry list	List available packages
stato registry search	Search packages by keyword
stato registry install	Install a community package

Full documentation: stato.hiniki.com | USAGE.md

Comparison

Capability	Stato	Plain CLAUDE.md	SkillKit	MemGPT	CrewAI
Validated modules	7-pass compiler	No validation	No validation	No	No
Cross-platform	4 platforms	Claude only	Claude only	OpenAI only	Framework-locked
Composition algebra	snapshot, slice, graft, merge	Manual copy	No	No	No
Privacy scanning	19 patterns	None	None	None	None
Web AI bridge	Import from any chat	No	No	No	No
Agent self-capture	Crystallize prompt	Human-authored	Human-authored	Auto (opaque)	Config files
Package registry	GitHub-based	No	No	No	No
Session resume	Structured recap	Re-read file	No	Built-in	No

Registry

Browse and install shared expertise packages:

stato registry list
stato registry search "bioinformatics"
stato registry install scrna-expert

Contributing

https://github.com/genecell/stato

Issues and PRs welcome.

Contact

Min Dai - [e-mail]

Fishell Laboratory, Harvard Medical School and the Broad Institute of MIT and Harvard.

License

MIT