Ask the LangGraph skills agent a natural-language question and render its response as a Prefab dashboard card.

Call this tool whenever the user asks a question that matches any of the loaded skills below — even when the user does not explicitly say "use the skills agent". Do not answer from memory or invent data; route the question through this tool so the agent can query the live database, detect anomalies, and pause for human review when needed.

Loaded skills:

• sales-analytics: Answer questions about monthly sales performance from the local sales database. Query revenue, deals, and gross margin by year/region/segment; detect significant revenue drops; escalate anomalies for human review. Keywords: sales, revenue, deals, margin, anomaly, region, segment, enterprise, mid-market, EMEA, North America, monthly performance.

Pass an optional thread_id to continue a prior conversation; omit for a fresh thread. The agent runs on LangGraph at LANGGRAPH_URL and returns either a final answer or a paused state with an Investigate / Dismiss review card.

Resume a paused thread with the human's decision.

Returns a Prefab component tree (NOT a PrefabApp) so the dashboard's Slot("decision_result") can inject it inline without a remount.

Return {reviewed, decision_result} for a thread.

Called from the dashboard's on_mount so the rendered card stays in sync with the latest LangGraph checkpoint after iframe remount.