From conversation to task and deliverable
A useful Agent platform needs to connect a request with its execution and the result a team can review. We model each execution as an AgentRun within a work scope.
The run is the unit of execution
An AgentRun can start from a conversation, a scheduled trigger, or a background task. It records what is being done and provides a common way to track progress and completion. A longer piece of work can continue as a background task while the original conversation remains available.
Skills define how work is produced
A skill gives the Agent a reusable method for a kind of work. Its contract determines the suitable work scope and the expected outputs. The platform keeps execution and deliverable rendering separate: the run tracks the work, while the resulting artifact determines how people see and use the output.
Results stay reviewable
A completed task can leave files and deliverables in the same scope as the request. People can review the result, continue the conversation, and use it in later work. High-impact conclusions still require human review before use.