The Expertise Layer for AI Agents

The Expertise Layer for
AI Agents

Capture, validate, and transfer what your AI agents learn. Across sessions, platforms, and teams.

View on GitHub
stato demo

$ stato init

Initialized .stato/ in ~/my-project


$ stato crystallize

# Agent writes expertise to .stato/ modules...


$ stato status

+-----------------------+--------+---------+--------+

| Module | Type | Version | Status |

+-----------------------+--------+---------+--------+

| skills/qc_filtering | skill | 1.2.0 | valid |

| skills/normalization | skill | 1.1.0 | valid |

| plan.py | plan | 1.0.0 | 3/5 |

+-----------------------+--------+---------+--------+


$ stato bridge --platform claude

Generated CLAUDE.md (2 skills, 5 working rules)

🧠

Never Lose Knowledge

Expertise persists across sessions on disk as validated Python files. They survive /compact, session restarts, and context window limits. Your agent starts every session with full knowledge.

📦

Share and Compose

Snapshot, import, and merge expertise across projects and teams. Template mode preserves learned parameters while resetting progress. Privacy scanning before every export.

🔀

Any Platform

Bridge generators create CLAUDE.md, .cursorrules, AGENTS.md, or generic README files. Claude Code, Cursor, Codex from one source. Even transfer from web AI conversations.

If You Already Know These Tools

Stato combines familiar concepts from the tools you use every day.

📦

Like npm/pip

Install, share, and version expertise packages from a registry. stato registry install scrna-expert

🎯

Like a Compiler

7-pass validation pipeline rejects invalid modules. Error codes, auto-corrections, type checking.

🗃

Like Docker

Package expertise so it runs on any platform. Claude Code, Cursor, Codex, or any agent that reads files.

🔀

Like Git

Snapshot, diff, and merge expertise. Track how agent knowledge evolves across sessions and projects.

What no existing tool does: the agent extracts its own knowledge. Plan in any web AI, build in any coding agent. Privacy scanning before every export.

How It Works

Four steps from zero to portable expertise.

1

Crystallize

Run stato crystallize and paste the prompt into your agent. It captures its own expertise into validated modules.

2

Persist

.stato/ files live on disk. They survive session restarts, /compact, and context window limits. Your agent starts every session with full expertise.

3

Compose

Snapshot, slice, graft, and merge expertise across projects. Privacy scanner checks for secrets before you share.

4

Bridge

Generate platform-specific files for any coding agent. Claude Code, Cursor, Codex, or generic. Same expertise, every platform.

Session 1 vs Session 2: Same Agent, Different Knowledge

Without stato, every session starts from scratch. With stato, expertise persists.

😕 Fresh Session (no stato)

"The compiler likely uses a multi-pass

validation approach:


1. Parse the Python source

2. Check the structure

3. Validate types


I'd recommend using AST parsing for

static analysis..."


Generic advice. No specifics. Guessing.

🤓 Next Session (with stato)

"The compiler runs a 7-pass pipeline:


1. Syntax (ast.parse, E001)

2. Structure (find class, E002)

3. Type Inference (name + fields)

4. Schema Check (E003, E004)

5. Type Check (W001-W003 auto-fix)

6. Execute (sandbox, E005-E006)

7. Semantic (DAG check, E008-E010)


Auto-corrections apply in reverse-line-

order to avoid offset drift."


Specific. Accurate. Implementation-level detail.

Same agent, same question. Stato makes session 2 start where session 1 left off.

How Stato Compares

Stato is the only tool that validates, composes, and transfers agent expertise across platforms.

Feature Stato CLAUDE.md SkillKit MemGPT CrewAI
Typed modules (skill, plan, memory, context) ~ ~
Compiler validation (7-pass pipeline)
Composition (snapshot, slice, graft)
Privacy scanner (19 patterns, review gate)
Cross-platform bridges (4 platforms) ~
Web AI import (bundle files)
Context restoration (stato resume) ~

Get Started in 60 Seconds

Choose your starting point.

across sessions

# Session 1: capture expertise

$ pip install stato

$ stato init

$ stato crystallize

# Agent writes .stato/ modules from its own knowledge

$ stato bridge --platform claude

Generated CLAUDE.md


# Session 2 (next day, after /compact):

# Agent reads CLAUDE.md automatically

$ stato resume

# Full context restored

Bridge Web AI to Coding Agents

Transfer expertise from Claude.ai, Gemini, or ChatGPT into Claude Code, Cursor, or Codex. One file, one command.

💬

Web AI

Claude.ai, Gemini, ChatGPT

📄

stato_bundle.py

Single Python file

📦

stato import-bundle

Validate + write

🤖

Any Platform

Claude Code, Cursor, Codex

web AI transfer

$ stato crystallize --web --raw | pbcopy

# Prompt copied. Paste into Claude.ai / Gemini / ChatGPT

# Save the AI's output as stato_bundle.py


$ stato import-bundle stato_bundle.py


Bundle contents:

Skills: 2 (data_loading, preprocessing)

Plan: yes

Memory: yes

Context: yes


skills/data_loading.py

skills/preprocessing.py

plan.py

memory.py

context.py


Result: 5 imported, 0 failed

Generated CLAUDE.md