dashboard

# 仪表盘规划 通过 ask_user 单轮询问收集用户需求，再动态生成 HTML 看板文件。 ## 三阶段工作流 ### Phase 1 — 理解数据 Call get_schema ONCE. Run 1-3 lightweight queries to understand tables and columns. After data exploration, IMMEDIATELY proceed to Phase 2 in the SAME turn. Do NOT stop or wait. ### Phase 2 — Collect requirements via ask_user (SINGLE question) Call ask_user ONCE with a single multi-select question asking what the user wants on the dashboard. Derive 4-6 options from actual data columns. Use real column names. Ask in user language. After ask_user returns, IMMEDIATELY proceed to Phase 3 in the SAME turn. Do NOT ask again. ### Phase 3 — Design and propose (SAME turn as Phase 2 answer) Build 2-6 widgets using ONLY metrics the user selected. Each widget needs valid SQL, ONLY real table/column names. NEVER fabricate names. KPI_Card: SQL returns 1 row; col1=value, col2=subtitle, col3=trend_pct. Grid: w=3,h=2. Bar_Chart / Line_Chart: field_mapping: x, y Grouped_Bar_Chart: field_mapping: x, value_cols=[col1,col2,...] Stacked_Bar_Chart: field_mapping: x, y=[col1,col2,...] Pie_Chart: field_mapping: label, value Scatter_Plot: field_mapping: x, y, color(opt) Area_Chart: field_mapping: x, y Heatmap: field_mapping: x, y, value Layout: KPI cards at y=0 (w=3,h=2 each), charts below at y=2+. Total width=12 units. Call propose_dashboard_outline(name=..., widgets=[...]). Output NOTHING after the tool call. ## CRITICAL RULES - Call ask_user EXACTLY ONCE. Never call it twice. - After the user answers ask_user, you MUST call propose_dashboard_outline in the same turn. - Do NOT output text after ask_user or propose_dashboard_outline. - Do NOT manually call generate_chart — the dashboard system handles chart creation. ## HTML export note After generate_dashboard succeeds, the result includes a download-HTML link. Mention it naturally: You can also download the HTML file to view offline. ## Implementation reference - Proposal tool entry: `agent/tools/business/export.py::_tool_propose_dashboard_outline` - Generation tool entry: `agent/tools/business/export.py::_tool_generate_dashboard`