Quick Start

Once installed, all agents and rules load automatically in every session.

Choose the workflow before choosing maestria

maestria adds coordination and review sessions. For small changes or expensive, slow models, direct execution or a single skill load is often better. Read When to use maestria first.

Start a Session

Launch OpenCode V2 as usual. The plugin loads with it.

Investigate with the main loop

Trace unfamiliar code with the recon skill in the main session:

With the recon skill, explore the auth module and trace the login flow.

For unfamiliar, multi-file, or high-risk work, hand it to the maestria agent:

@maestria I need to add rate limiting to the API. Plan and implement it.

The maestria agent works as one main loop: it explores, plans, and implements the change itself, loading the recon, architecture-plan, and diagnose skills when the task needs them, and sends meaningful changes to @reviewer for an independent check before delivery. Modes adjust that loop: the full-method mode loads recon and architecture-plan before implementing with required review, the research-only mode investigates without editing or dispatching, and the fast mode works directly on familiar code. Pick a mode with its keyword in the prompt or with /fein, /sonar, or /blitz. See the Agent Reference.

For longer autonomous runs, the independent OpenCode Goal Plugin adds Codex-style goal mode; maestria neither requires nor activates it.

Next Steps

• Agent Reference - Full docs for the main loop and reviewer
• Workflow Patterns - Main-loop and review patterns