MAESTRO

One ecosystem for reasoning agents. Describe a task, and MAESTRO builds, runs, and evolves the chain — right in the terminal.

Get started The CARL format Agent Skill

Choose your path

MAESTRO → the TUI & CLI The system: drive it from the MAESTRO CARE terminal UI (TUI) or the headless `care` CLI to generate, run, save, and evolve reasoning chains. Start here to work with agents interactively.

CARL → the chain format The format reasoning chains are written in: typed steps, dependencies, and per-step context. Start here to learn how a chain is structured.

Agent Skill → drive MAESTRO from an agent A downloadable agentskills.io skill that lets Claude Code or hermes run the `care` CLI — generate, run, and evolve chains — with portable auto-detection. Start here to install it.

Install the agent skill

Give Claude Code, Codex, hermes, or OpenClaw the Maestro skill in one command:

uvx maestro-install skill        # asks which agent; or --agent claude | codex | hermes | openclaw | all

It downloads and unpacks the skill into the agent’s skills directory:

Claude Code → ~/.claude/skills/maestro
Codex → ~/.codex/skills/maestro
hermes → ~/.hermes/skills/maestro
OpenClaw → ~/.openclaw/skills/maestro

More: Agent Skill · Install & Use.

How it fits together

MAESTRO is the product on top of a four-part stack. Each part owns one stage of a chain’s life.

MAGE — generation

Turns a natural-language task into a CARL chain.

CARL — the chain format

The format every chain is written in: steps, dependencies, and context queries.

GigaEvo Memory — persistence

Stores chains, agents, skills, and run history.

GigaEvo Platform — evolution

Runs a genetic search over chains and promotes the winners.

In short: CARL is the format, MAESTRO is the system that runs it. MAGE writes a chain in the CARL format; MAESTRO runs and evolves it — from the MAESTRO CARE TUI or the care CLI.