You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Enable agents to compose other agents as sub-tasks, forming a parent-child execution tree with result aggregation and error propagation.
Prior Art
AutoGPT implements nested agent composition as a core feature — agent graphs can contain "Agent Executor" blocks that run other agents as sub-steps. Parent agents receive structured results from child agents and can branch/retry based on outcomes. Supports up to 3 levels of nesting with shared context propagation.
Moltis has a similar concept via its multi-agent orchestration layer.
The current sub-agent system (#66) supports spawning independent sessions, but lacks structured composition — a parent agent can't easily define "run agent A, then if successful run agent B with A's output, otherwise run agent C". Nested composition enables complex workflows like: research agent → analysis agent → report-writing agent, with the parent orchestrating the pipeline.
Summary
Enable agents to compose other agents as sub-tasks, forming a parent-child execution tree with result aggregation and error propagation.
Prior Art
sessions_spawn(feat: Sub-agent / spawn_agent tool for parallel task delegation #66) for basic sub-agent delegation.Motivation
The current sub-agent system (#66) supports spawning independent sessions, but lacks structured composition — a parent agent can't easily define "run agent A, then if successful run agent B with A's output, otherwise run agent C". Nested composition enables complex workflows like: research agent → analysis agent → report-writing agent, with the parent orchestrating the pipeline.
Proposed Design
sessions_spawnwith composition semantics:Acceptance Criteria
Related Issues