agent-harness

A multi-agent orchestrator coordinates planner, worker, and evaluator subagents to automate the plan-execute-evaluate cycle. It manages the task lifecycle across per-task work directories and shares context to drive structured execution.

Yes, you can create domain-specific agents that record lessons and experiences during execution. This evolution history enables self-improvement, allowing the evaluator to drive better future runs across iterations.

During a multi-agent task lifecycle, the orchestrator creates per-task work directories for context sharing. This structured environment allows planner, worker, and evaluator subagents to coordinate end-to-end orchestration.

Scaffolding an agent project uses a structured harness with reusable domain skills and references. This supports the orchestrated plan-execute-evaluate loop, providing end-to-end task lifecycle management and evolution tracking.

Yes, the multi-agent harness supports reusable domain skills and references to scaffold agent projects. These skills integrate with the orchestrator, planner, worker, and evaluator subagents to execute specialized tasks.

An evaluator subagent drives self-improvement by assessing task execution and recording lessons in the evolution history. These experiences are applied across subsequent iterations to optimize future orchestrator runs.