Workers and skills
Route specialized work to workers with identity, context, skills, and permissions so HQ behaves like a team instead of one generic assistant.

Specialists
Workers carry domain instructions and skills carry repeatable workflows.

Choose a worker
A worker has identity, context, skills, and permissions. That is what makes it different from a one-off agent prompt.
Use /run when a task clearly belongs to a domain worker, and check whether the worker's context and permissions match the job.
Use the skill
Skills encode procedures such as planning, QA, handoff, deployment, learning, and content production.
When a task matches a named skill, let the worker follow that skill's workflow instead of reinventing the process in the prompt.


Train one worker deeply
The first worker should come from a recurring task that involves judgment you have already accumulated. Depth beats a roster of shallow workers.
Pick one repeated task, explain what good looks like, let the worker attempt it, then capture precise corrections with /learn.
Keep the chapter executable.
Routes specialized work through a worker that carries domain instructions and memory.
Turns reusable corrections or insights into future HQ behavior.
What to keep in mind.
Worker-first habit
Use a specialist when the task has a clear domain such as design, QA, content, security, finance, data, or deployment.
First worker question
What recurring task do you do that involves judgment you have already accumulated?
