Agent
Domain: Agent Architecture
Motivation
An agent is the unit that actually executes a task: it owns a model, tools, and conversation state, and produces a result. In dsh the dsh-agent package exposes the public Agent handle — create or resume a live agent, send steering input, cancel work, and observe activity — but it does not create model calls itself; you pair it with a driver.
Concrete Example
ctx.agents.create() builds a fresh agent and ctx.agents.resume(id) reopens a persisted one; the agent then runs under a driver such as dsh-agent-loop.
Analogy
A worker on a ticket — given a model, tools, and the history, it does the job and reports back.