Quick Start

Once the plugin is loaded (see Installation & Setup), its components are available in your session as namespaced identifiers - no extra setup.

Choose the workflow before choosing the orchestrator

Maestria adds coordination and review sessions. For small changes or expensive, slow models, direct execution or one specialist is often the better path. Read When to use Maestria first.

Start a Session

Launch Claude Code with the plugin loaded:

claude --plugin-dir ./packages/claude-code

Delegate to a Specialist

Call any of the 7 specialist agents directly by their namespaced name:

@maestria:adventurer Explore the auth module and trace the login flow.

The specialists are adventurer, architect, builder, diagnose, planner, reviewer, and writer. See the Specialist Reference for role details.

Use the Orchestrator

For unfamiliar, multi-file, or high-risk work, run the orchestrator skill to route the work through the right specialists:

/maestria:orchestrator I need to add rate limiting to the API. Plan and implement it.

The orchestrator is the routing layer - it decomposes the task, delegates to specialists, and integrates the results. It does not implement work it has delegated.

Select a Pipeline Mode

Three workflow commands control the pipeline depth. They are namespaced and invoked like any Claude Code command:

Command	Pipeline	Use case
/maestria:fein	Full pipeline: recon → design → implement → review	Production-grade work, mandatory review
/maestria:sonar	Research only: owning specialist → optional distinct specialist → STOP	Understanding before committing
/maestria:blitz	Fast path: direct or maestria:builder (skip optional ceremony; required review remains)	Quick fixes, known territory

/maestria:fein add rate limiting to the API.

What to Expect

• Global rules - the maestria:global-rules skill is configured to preload into every specialist agent at startup, staging the universal rules contract into context.
• Read-only roles - @maestria:adventurer, @maestria:planner, and @maestria:reviewer have the Write and Edit tools denied; they explore, plan, and review but never implement.
• Advisory, not security - persona constraints in skills and prompts are guidance. disallowedTools is the only runtime enforcement; do not treat prompt rules as a security boundary.

Native candidate - verify behavior

The plugin validates cleanly with claude plugin validate --strict, but runtime behavior (skill preload resolution, command pipelines, namespaced invocation) is not yet tested end to end. If a component does not behave as documented, report it - upstream Claude Code docs are moving and unpinned.

Next Steps

• Specialist Reference - Full docs for all 7 specialists plus the orchestrator
• Workflow Patterns - Orchestration and pipeline patterns