用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/Zafer-Liu/Data-Analysis-Agent --skill dashboard命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | dashboard |
| description | 规划业务数据仪表盘/看板(dashboard) |
| icon | 📊 |
| allowedTools | ["get_schema","query_data","ask_user","propose_dashboard_outline"] |
通过 ask_user 单轮询问收集用户需求,再动态生成 HTML 看板文件。
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.
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.
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.
After generate_dashboard succeeds, the result includes a download-HTML link. Mention it naturally: You can also download the HTML file to view offline.
agent/tools/business/export.py::_tool_propose_dashboard_outlineagent/tools/business/export.py::_tool_generate_dashboard