Find a demo-ready deepagents trace in LangSmith by querying task and write_todos tool runs
To pick a deepagents run to show judges, query the trace tree instead of clicking through it. List successful root runs, then per trace: client.list_runs(project_name=P, trace_id=root.trace_id, run_type="tool", filter="eq(name, \"task\")") and look for each subagent name in the task inputs. That tells you which runs used every subagent. To find a specific re-plan, full-text search the plan tool: filter=and(eq(name, "write_todos"), search("<todo text>")). client.get_run_url(run=root) gives the link. Gotchas: list_runs rejects limit > 100 (400 "Limit exceeds maximum allowed value of 100"). read_run returning 404 for an id you just listed usually means a different LANGSMITH_API_KEY is in effect: a shell export and .env can point to different workspaces, and load_dotenv does not override the shell. Trace links only open for someone logged in to that workspace.