Agent Skills › lightdash/lightdash › lightdash-analytics

lightdash-analytics

GitHub

基于 Lightdash MCP 服务器进行数据探索、指标查询及可视化分析。支持通过语义层回答业务问题,生成图表仪表板,并管理 dbt 内容变更流程,确保数据查询的准确性与规范性。

plugins/lightdash/skills/lightdash-analytics/SKILL.md lightdash/lightdash

触发场景

回答业务数据分析问题 创建或修改 Lightdash 图表和仪表板 查询受治理的指标数据

安装

npx skills add lightdash/lightdash --skill lightdash-analytics -g -y
更多选项

非标准路径

npx skills add https://github.com/lightdash/lightdash/tree/main/plugins/lightdash/skills/lightdash-analytics -g -y

不安装直接使用

npx skills use lightdash/lightdash@lightdash-analytics

指定 Agent (Claude Code)

npx skills add lightdash/lightdash --skill lightdash-analytics -a claude-code -g -y

安装 repo 全部 skill

npx skills add lightdash/lightdash --all -g -y

预览 repo 内 skill

npx skills add lightdash/lightdash --list

SKILL.md

Frontmatter
{
    "name": "lightdash-analytics",
    "description": "Use when answering business questions, exploring governed metrics, querying Lightdash data, or creating Lightdash charts and dashboards through the Lightdash MCP server."
}

Lightdash Analytics

Use the Lightdash MCP server as the source of truth for the user's governed semantic layer.

Discover before querying

  1. Set the active project if the server requires one.
  2. Use route_agent when available; otherwise list the available explores.
  3. Inspect the selected explore and fields before composing a metric query.
  4. Use verified content as a reference when it matches the request.

Never invent explore names, field IDs, metric definitions, filter values, or query UUIDs. Search field values when a string filter must match an existing value.

Answering questions

Prefer run_metric_query for questions that fit the semantic layer. Use run_sql, when it is available, only when the requested analysis cannot be represented through a governed explore.

If a metric query is still running, poll get_query_result with the returned query UUID. When a visual would clarify the result, render a completed metric query with render_chart.

State the metric, time period, filters, and any important caveats in the answer. Treat an empty result as a valid result rather than a failed query.

Creating content and changing analytics

Before creating a chart or dashboard, inspect the relevant schema and existing content. Validate the query first, then use the server's creation workflow.

For dbt or content-as-code changes, work in a branch and use the Lightdash CLI workflow: preview the change, validate it, review it, then merge. Do not deploy or start AI writeback unless the user explicitly asks for that external change.

Built-in Lightdash skills

The MCP server can expose additional Lightdash skills and references. Use list_skills when the client does not surface MCP resources directly, then read only the skill or reference relevant to the task.

版本历史

  • 71d06ec 当前 2026-08-20 16:07

同 Skill 集合

.claude/skills/add-onboarding-tour/SKILL.md
.claude/skills/add-scope-walkthrough/SKILL.md
.claude/skills/agent-harness/SKILL.md
.claude/skills/breakup-pr/SKILL.md
.claude/skills/creating-pull-requests/SKILL.md
.claude/skills/debug-local/SKILL.md
.claude/skills/deprecate-endpoint/SKILL.md
.claude/skills/fix-vulnerability/SKILL.md
.claude/skills/frontend-style-guide/SKILL.md
.claude/skills/graphite/SKILL.md
.claude/skills/har-replay/SKILL.md
.claude/skills/ld-permissions/SKILL.md
.claude/skills/renovate-pr/SKILL.md
skills/lightdash-release-review/SKILL.md
skills/upgrade-preflight/SKILL.md
skills/developing-in-lightdash/SKILL.md

元信息

文件数
0
版本
bd3048b
Hash
d8701071
收录时间
2026-08-20 16:07

首页 - Wiki
Copyright © 2011-2026 iteam. Current version is 2.156.0. UTC+08:00, 2026-11-21 16:35
浙ICP备14020137号-1