L4 Functional Sub-Agent Architecture

L4 functional agents are stateless, ephemeral specialist agents spawned for a specific task, completing it, and returning the result to the manager agent that called them. No persistent identity, no memory, no Telegram presence.

The problem they solve: L2 manager agents occasionally need deep specialist knowledge such as SEO audits, PDF generation, or competitive research, that does not belong embedded in a general-purpose manager SOUL.md. Cramming specialist knowledge into manager configs causes context bloat and staleness. Every agent invocation carries the knowledge whether it is needed or not.

L4 agents are pure functions. They receive a task brief, do the work, and return a structured result. Their SOUL.md contains only role definition, task input format, and output format.

What is included: L4 SOUL.md templates for 5 specialist types (SEO auditor, PDF builder, content writer, competitive analyst, QA reviewer), the task brief format, the result return format, the spawning protocol for L2 managers, and the architectural decision guide.

Best for OpenClaw operators running an L1/L2 hierarchy who need specialist capability on demand without adding permanent agents.