[FEATURE] create/write agents on the fly.

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

When working on complex, multi-step tasks, Claude Code currently requires agents to be defined manually and statically in advance. This creates a rigid workflow where the orchestrator cannot adapt to emerging subtasks at runtime. In real-world projects — especially large codebases — the nature and scope of required agents often only becomes clear during execution, not before it. This forces developers to either over-provision agents upfront or interrupt the workflow to reconfigure manually.

Proposed Solution

Introduce a dynamic Agent Creator layer that sits between the Orchestrator and the individual Agents:
Orchestrator → Agent Creator → Agents
The Orchestrator analyzes the task and delegates to an Agent Creator, which dynamically spawns, configures, and names purpose-built agents at runtime — each with their own context, tools, and scope. Once created, these agents execute in parallel or sequence and report back to the Orchestrator.
This would enable:

Dynamic decomposition — the Orchestrator doesn't need to know all subtasks upfront
Scoped agents — each agent gets only the tools and context it needs
Reusability — the Agent Creator can reuse existing agents when appropriate
Scalability — complex tasks naturally self-organize into optimal agent topologies

This pattern mirrors how modern agentic frameworks (LangGraph, AutoGen) handle emergent complexity, and would make Claude Code significantly more powerful for large-scale, open-ended engineering tasks.

Alternative Solutions

_No response_

Priority

High - Significant impact on productivity

Feature Category

API and model interactions

Use Case Example

_No response_

Additional Context

_No response_