LangGraph Adapter

Source references:

• https://github.com/langchain-ai/langgraph
• https://docs.langchain.com/oss/python/langgraph/interrupts

Agent contract mapping

• Graph node -> AEEF agent role execution unit
• Node metadata must include agent_id, allowed_tools, required_gates
• Checkpoint policy must align with escalation and retry limits

Hook mapping

• Pre-node wrapper -> phase=pre hook checks
• Post-node wrapper -> phase=post trace and artifact capture
• Interrupt/HITL edge -> phase=stop escalation checkpoint

Gate enforcement points

• Conditional edges enforce gate decisions (pass, fail, conditional)
• Failed gate loops to remediation node with capped retries
• Human-approval-required gates route to explicit approval node

Handoff mapping

Graph state payload carries:

• task_context
• assumptions
• risks
• skill_usage
• next_decision_required

Validate handoffs with handoff-artifact.schema.json.

Run-ledger mapping

Each node execution emits RunLedgerEntry with:

• tools_used
• skill_ids
• node-level gate_decisions
• artifact_refs

Real example flow

See E2E Feature Flow (LangGraph).