Anthropic Chat Model: expose "strict": true for client tool definitions

Environment: n8n 2.37.7, @n8n/n8n-nodes-langchain.agent typeVersion 1.9 (AgentV1), lmChatAnthropic 1.6, @langchain/anthropic 1.5.6, model claude-opus-5, adaptive thinking, streaming, prompt caching 5m, four client tools, all subworkflow-backed. Evidence comes from a transparent proxy placed between n8n and api.anthropic.com, capturing 167 exchanges and 267 tool_use blocks over a production-equivalent workload driven through the real agent loop. Problem Anthropic tool use is non-strict by default. In that mode a JSON Schema required list guides generation but does not constrain it, and the model may emit a tool call with no arguments at all. Anthropic documents strict: true as the mechanism that guarantees schema-conforming tool inputs through constrained sampling. n8n cannot send it. A source read of 2.37.7 finds no strict handling anywhere in the langchain nodes: no node parameter, no bindTools(tools, { strict: true }), no withConfig({ strict: true }). The Tools Agent only checks…