Quick Start

Start a new Kimi Code session after installing, so the orchestrator skill and system prompt load.

Choose the route that fits the task

Delegation adds model work and waiting time. Use When to Use maestria before sending a small task or an expensive model through the full pipeline.

Delegate One Focused Task

Give the orchestrator a concrete task:

Use builder to add a JSDoc comment to the add function in src/math.ts.

The orchestrator loads the builder skill and routes the task to Kimi Code’s coder subagent.

Run a Staged Fix

For a task with an unknown cause and a clear need for review:

The login flow throws on an empty email. Trace the bug, fix it, and review the fix.

The usual sequence is:

1. diagnose finds and explains the cause.
2. builder applies the confirmed fix.
3. reviewer checks the result.

Review Several Similar Items

For three or more independent, uniform items, ask for a parallel review:

Review these five files for security issues: src/auth.ts, src/api.ts, src/db.ts, src/routes.ts, and src/middleware.ts.

Kimi Code can use AgentSwarm to fan out the same reviewer task. Make the swarm call the only tool call in that turn.

Plan a Larger Change

Ask planner for ordered milestones before implementation:

Use planner to draft a migration plan from REST to GraphQL.

The plan should contain phases, success criteria, and rollback points. Give it to the orchestrator for execution, or to builder one phase at a time.

Native Workflow Commands

Use a namespaced command when you want an explicit route:

/maestria:fein
/maestria:sonar
/maestria:blitz

• fein runs the full pipeline.
• sonar stops after research and design.
• blitz skips optional reconnaissance and design when the host permits direct execution.

Next Steps

• Installation
• When to Use maestria
• Workflow Patterns
• Skill Reference