Agent Skills › eosphoros-ai/DB-GPT › dashboard-design

dashboard-design

GitHub

为 dashboard-builder 提供仪表盘设计参考，匹配业务意图选择视觉风格，保持布局层级并适配数据字段，生成查询计划后通过工具构建仪表盘。

skills/dashboard-builder/internal/dashboard-design/SKILL.md eosphoros-ai/DB-GPT

触发场景

用户请求特定仪表盘视觉风格需要为数据选择合适的仪表盘样式

安装

npx skills add eosphoros-ai/DB-GPT --skill dashboard-design -g -y

更多选项

非标准路径

npx skills add https://github.com/eosphoros-ai/DB-GPT/tree/main/skills/dashboard-builder/internal/dashboard-design -g -y

不安装直接使用

npx skills use eosphoros-ai/DB-GPT@dashboard-design

指定 Agent (Claude Code)

npx skills add eosphoros-ai/DB-GPT --skill dashboard-design -a claude-code -g -y

安装 repo 全部 skill

npx skills add eosphoros-ai/DB-GPT --all -g -y

预览 repo 内 skill

npx skills add eosphoros-ai/DB-GPT --list

SKILL.md

Frontmatter

{
    "name": "dashboard-design",
    "author": "DB-GPT Dashboard Lab",
    "version": "1.0.0",
    "description": "Internal visual references for dashboard-builder. Select a dashboard style for the user's data while preserving its layout, colors and readable chart hierarchy."
}

Dashboard Design

This is an internal skill used by dashboard-builder, not a separate user task. Use the selected reference to guide the dashboard's appearance. It never supplies data, SQL, filter values, targets or claims about the selected source.

When the user selects a gallery style, its reference and these instructions are already loaded. Do not reload the skill or scan the other references.
Otherwise, if the user requests a particular visual style, call load_dashboard_style without a selection to list the available references, then load the matching selection. Match business intent and data shape: a command center emphasizes a large-screen overview; logistics emphasizes delivery/cost comparisons; marketing emphasizes campaign/channel outcomes. Do not make users choose a style for a simple request with no visual preference.
Keep the selected reference's slot ids and visual hierarchy. Adapt its suggested titles, metrics and chart kinds to actual fields; explain substitutions in the plan. Never invent appointments, revenue, targets or geography to fit a design.
Use Top 10 for ordinary category rankings and label that scope. Keep complete time ranges for trends; choose a row budget covering all dates and series (default 1000, at most 5000), or use an appropriate coarser time grain. Label bounded detail-table ranges. KPI totals must cover all relevant source rows.
Plan first and wait for confirmation using dashboard-builder. Generate actual queries with the existing Dashboard tools; their data validation still applies. Report the resulting dashboard through those tools so the homepage can open it.
Read only the selected style's extra guidance. Do not load all styles, duplicate the data schema, dump a complete reference into the answer, or produce a second HTML report alongside the dashboard.

The eight active gallery references are listed in the reference index. Each reference separates visual configuration from data queries. To extend a style, edit its reference and guidance here rather than adding a generation flow.