CHAPTER 03

Agent Assembly

English summary

Each run compiles an executable agent graph from its resolved identity, authorized tool catalog, prompt, state schema, and an ordered middleware policy stack. Middleware order determines which outcomes remain observable and which policies can short-circuit execution.

01Resolve a per-run assembly identity

02Filter the catalog, then guard execution

03Read middleware as four policy layers

04Explain behavior through ordering

SOURCE TRAIL

Read the implementation

This edition deliberately stops at a concise overview. The links below open the exact DeerFlow source files used by the complete Chinese chapter.

lead_agent/agent.py · assemble_lead_agent lead_agent/agent.py · final lead components middlewares/tool_error_handling_middleware.py · shared base extensions/ordering.py · core_ordering_constraints extensions/stack.py · semantic placement anchors lead_agent/agent.py · lead-specific middleware stack