把自然语言描述转为结构化 JSON,并由 mcp-diagram-generator MCP 服务生成 Draw.io、Mermaid 或 Excalidraw 图表文件。
编程
grafana-foundation-sdk
试用Build Grafana dashboards as code with the grafana-foundation-sdk typed builders (TypeScript or Go). Use when creating, modifying, or generating Grafana dashboard JSON programmatically, converting hand-written dashboard JSON to typed code, building monitoring dashboards, or working with Prometheus/Loki queries in dashboards.
它能做什么
The grafana-foundation-sdk provides strongly typed builder libraries for defining Grafana dashboards as code. Instead of writing raw JSON (which is error-prone and hard to review in diffs), you compose dashboards using chained builder calls that produce valid Grafana JSON.
技能文档
Grafana Foundation SDK
The grafana-foundation-sdk provides strongly typed builder libraries for defining Grafana dashboards as code. Instead of writing raw JSON (which is error-prone and hard to review in diffs), you compose dashboards using chained builder calls that produce valid Grafana JSON.
The SDK is auto-generated from Grafana's internal CUE schemas via the cog tool. It supports Go, TypeScript, Python, PHP, and Java. This skill focuses on TypeScript (primary) and Go (secondary) since those are the most common choices for infrastructure teams.
When to use this skill
- Creating new Grafana dashboards from scratch
- Converting existing hand-written dashboard JSON to typed code
- Adding panels, variables, or queries to dashboards
- Building reusable dashboard components (helper functions for common panel patterns)
- Generating dashboards dynamically based on service lists or configs
Installation
The SDK is published as concrete v0.0.x tags (latest: v0.0.16). Pin explicitly - it is pre-1.0 and the API churns between releases (see Known Gotchas).
TypeScript:
npm install '@grafana/grafana-foundation-sdk@~0.0.16'
# or
pnpm add '@grafana/grafana-foundation-sdk@~0.0.16'
Go:
go get github.com/grafana/grafana-foundation-sdk/go@v0.0.16
Core Architecture
Everything follows the builder pattern: create a builder, chain configuration methods, call .build() (TS) or .Build() (Go) to produce the final object. The output is standard Grafana dashboard JSON - compatible with Grafana's API, file-based provisioning, and Kubernetes ConfigMaps.
Each panel type, query type, and variable type lives in its own package. You import only what you need:
// Each concern has its own import
import { DashboardBuilder, RowBuilder } from '@grafana/grafana-foundation-sdk/dashboard';
import { PanelBuilder as TimeseriesBuilder } from '@grafana/grafana-foundation-sdk/timeseries';
import { PanelBuilder as StatBuilder } from '@grafana/grafana-foundation-sdk/stat';
import { DataqueryBuilder as PromQueryBuilder } from '@grafana/grafana-foundation-sdk/prometheus';
import { DataqueryBuilder as LokiQueryBuilder } from '@grafana/grafana-foundation-sdk/loki';
Quick Start - TypeScript
import { DashboardBuilder, RowBuilder, QueryVariableBuilder } from '@grafana/grafana-foundation-sdk/dashboard';
import { PanelBuilder as StatBuilder } from '@grafana/grafana-foundation-sdk/stat';
import { PanelBuilder as TimeseriesBuilder } from '@grafana/grafana-foundation-sdk/timeseries';
import { DataqueryBuilder } from '@grafana/grafana-foundation-sdk/prometheus';
import * as common from '@grafana/grafana-foundation-sdk/common';
const dashboard = new DashboardBuilder('My Service Overview')
.uid('my-service-overview')
.tags(['my-service'])
.editable()
.refresh('30s')
.time({ from: 'now-24h', to: 'now' })
.timezone('browser')
.withVariable(
new QueryVariableBuilder('service')
.label('Service')
.query('label_values(up{namespace="default"}, job)')
.datasource({ type: 'prometheus', uid: 'prometheus' })
.refresh(1)
.includeAll(true)
.allValue('.*')
.sort(1)
)
.withRow(new RowBuilder('Overview'))
.withPanel(
new StatBuilder()
.title('Request Rate')
.datasource({ type: 'prometheus', uid: 'prometheus' })
.withTarget(
new DataqueryBuilder()
.expr('sum(rate(http_requests_total{job=~"$service"}[5m]))')
.legendFormat('req/s')
)
.unit('reqps')
.decimals(1)
.height(4)
.span(6)
.colorMode(common.BigValueColorMode.Background)
.graphMode(common.BigValueGraphMode.Area)
.reduceOptions(
new common.ReduceDataOptionsBuilder().calcs(['lastNotNull'])
)
)
.withPanel(
new TimeseriesBuilder()
.title('Request Rate Over Time')
.datasource({ type: 'prometheus', uid: 'prometheus' })
.withTarget(
new DataqueryBuilder()
.expr('sum by (job)(rate(http_requests_total{job=~"$service"}[5m]))')
.legendFormat('{{job}}')
)
.unit('reqps')
.fillOpacity(15)
.height(8)
.span(12)
);
// Output the dashboard JSON
console.log(JSON.stringify(dashboard.build(), null, 2));
Key Patterns
1. Helper functions for repeated panel configurations
The biggest win from using the SDK is creating reusable helpers that encode your team's conventions:
function promDs() {
return { type: 'prometheus', uid: 'prometheus' } as const;
}
function lokiDs() {
return { type: 'loki', uid: 'loki' } as const;
}
function promQuery(expr: string, legend?: string) {
const q = new DataqueryBuilder().expr(expr);
if (legend) q.legendFormat(legend);
return q;
}
function statPanel(title: string, expr: string, opts?: { unit?: string; decimals?: number; color?: string }) {
const panel = new StatBuilder()
.title(title)
.datasource(promDs())
.withTarget(promQuery(expr))
.height(4)
.span(4)
.colorMode(common.BigValueColorMode.Background)
.graphMode(common.BigValueGraphMode.Area)
.reduceOptions(new common.ReduceDataOptionsBuilder().calcs(['lastNotNull']));
if (opts?.unit) panel.unit(opts.unit);
if (opts?.decimals !== undefined) panel.decimals(opts.decimals);
// Thresholds can be set via .thresholds() if needed
return panel;
}
2. Template variables
// Query variable - populated from Prometheus labels
new QueryVariableBuilder('service')
.label('Service')
.query('label_values(http_server_duration_count{namespace="myapp"}, job)')
.datasource({ type: 'prometheus', uid: 'prometheus' })
.refresh(2) // 1=on dashboard load, 2=on time range change
.includeAll(true)
.allValue('.*')
.sort(1) // 1=alphabetical asc
// Custom variable - static key:value pairs
new CustomVariableBuilder('level')
.label('Log Level')
.query('All : .+, Error : error|fatal, Warning : warn, Info : info, Debug : debug')
.current({ text: 'All', value: '.+' })
Reference variables in queries with standard Grafana syntax: $service, $__range, $__rate_interval, $__auto.
3. Panel sizing
Panels use height(h) (grid rows) and span(w) (out of 24 columns):
- Full width:
.span(24) - Half width:
.span(12) - Third width:
.span(8) - Quarter width:
.span(6) - Typical stat panel:
.height(4).span(4) - Typical timeseries:
.height(8).span(12)
4. Thresholds
import { ThresholdsConfigBuilder } from '@grafana/grafana-foundation-sdk/dashboard';
// First step must have no value (it's the base)
new StatBuilder()
.thresholds(
new ThresholdsConfigBuilder()
.mode(common.ThresholdsMode.Absolute)
.steps([
{ value: null as any, color: 'green' },
{ value: 80, color: 'yellow' },
{ value: 95, color: 'red' },
])
)
5. Field overrides
new TimeseriesBuilder()
.overrideByName('Revenue', [
{ id: 'color', value: { fixedColor: 'green', mode: 'fixed' } },
])
.overrideByRegexp('.*5..', [
{ id: 'color', value: { fixedColor: 'red', mode: 'fixed' } },
])
6. Rows (including collapsed)
// Regular row
.withRow(new RowBuilder('Traffic'))
// Collapsed row with nested panels
.withRow(
new RowBuilder('Business Details')
.collapsed()
.withPanel(/* ... */)
.withPanel(/* ... */)
)
7. Loki log and metric queries
import { DataqueryBuilder as LokiQueryBuilder } from '@grafana/grafana-foundation-sdk/loki';
// Log query
new LokiQueryBuilder()
.expr('{namespace="myapp", app=~"$service", level=~"$level"}')
.refId('A')
// Metric query from logs
new LokiQueryBuilder()
.expr('sum by (buyer_wallet)(count_over_time({namespace="myapp"} | event="request" [$__range]))')
.legendFormat('{{buyer_wallet}}')
.refId('A')
8. Transformations
Transformations are applied as raw objects since the SDK doesn't have typed builders for all transformation types:
new TableBuilder()
.withTransformation({
id: 'reduce',
options: {
reducers: ['lastNotNull'],
mode: 'seriesToRows',
includeTimeField: false,
labelsToFields: true,
},
})
.withTransformation({
id: 'organize',
options: {
excludeByName: { Field: true },
renameByName: { buyer_wallet: 'Buyer Wallet', 'Last not null': 'Requests' },
},
})
.withTransformation({
id: 'sortBy',
options: { sort: [{ field: 'Requests', desc: true }] },
})
Generating Output
The .build() call returns a plain object matching Grafana's dashboard JSON schema. Serialize it however you need:
// Standard JSON file (for provisioning or ConfigMaps)
const fs = require('fs');
const dashboard = builder.build();
fs.writeFileSync('dashboard.json', JSON.stringify(dashboard, null, 2));
// Kubernetes resource manifest (for Grafana's k8s API)
const manifest = {
apiVersion: 'dashboard.grafana.app/v1beta1',
kind: 'Dashboard',
metadata: { name: dashboard.uid },
spec: dashboard,
};
console.log(JSON.stringify(manifest, null, 2));
Known Gotchas
These are sharp edges discovered from real usage and open issues on the SDK repo:
-
SDK is v0.0.x (public preview) - Used by Grafana Labs in production but the API can change between releases. Pin your version explicitly. Best suited for Grafana >= 12, works with >= 10.
-
instant()andrange()are mutually exclusive in Prometheus - Calling.instant()setsinstant=trueANDrange=false. Calling.range()does the opposite. Use.rangeAndInstant()if you need both. -
Loki
range()/instant()are deprecated - Use.queryType('range')or.queryType('instant')instead. Similarly,.resolution()is deprecated in favor of.step(). -
First threshold step must have
value: null- This is the base/default color. Omitting it produces invalid JSON. -
Panel IDs are auto-assigned - You don't set
idon panels. Grafana assigns them at import time. Similarly,gridPos.x/yare computed fromheight()andspan(). -
Transformations are plain objects - The SDK has no typed builders for transformations. Pass them as raw
{ id, options }objects via.withTransformation(). -
CustomVariable quirk - When provisioning via Grafana's API,
CustomVariableBuilderrequires the.query()field with comma-separated key:value pairs (e.g.,'All : .+, Error : error') for options to persist, even when.values()is also used. -
Go:
cog.ToPtr()is essential - Many struct fields are pointer types. Usecog.ToPtr[T](value)for nullable fields (thresholds, datasource refs). TypeScript doesn't have this issue. -
Go:
Build()returns error - Always check it. TypeScript's.build()returns the object directly with compile-time type safety instead. -
No typed query builders for plugin datasources - Only core datasources (Prometheus, Loki, Tempo, Elasticsearch, CloudWatch, etc.) have builders. For third-party plugins, define custom query types by implementing the
Builderinterface. -
Dashboard schema v1 vs v2 - This skill targets the v1 dashboard (
@grafana/grafana-foundation-sdk/dashboard, k8s apiVersiondashboard.grafana.app/v1beta1). A newer schema v2 ships asdashboardv2beta1(k8s apiVersiondashboard.grafana.app/v2beta1) with its own builders. v2beta1 is still stabilizing and has known sharp edges (e.g. transforms, annotation positioning, SQL expressions in Go) - prefer v1 unless you specifically need v2 layouts. Most query/panel builders are shared; some expose aQueryV2Builder/VisualizationV2Buildervariant for v2. -
Builders are only type-checked if wired into a tsconfig - The SDK gives compile-time safety only when the generator file is actually type-checked. A generator sitting under a non-package directory (e.g. a Helm chart dir) that no
tsconfigincludes is silently unchecked, so type errors surface only at.build()runtime. Also: the SDK's output targets ES2024/bundlermodule resolution, which an older globaltscchokes on - run the project-local compiler (npx tsc), not a stale global one. -
Regenerate JSON after every generator edit - The deployed dashboard is the generated JSON, not the
.ts/.gosource. Edit the generator, re-run it, and commit the regenerated JSON together; never hand-edit the generated JSON (the next regen silently overwrites it). A repo rule ("never edit the dashboard JSON directly") is worth adding.
Reference Files
For detailed API reference and complete examples, see:
references/typescript-api.md- Full TypeScript API with all panel types, query builders, and configuration optionsreferences/patterns.md- Common dashboard patterns, recipes, and a complete example converting this project's dashboard to SDK code
相关技能
通过托管 OAuth 访问 Microsoft Graph Excel 接口,读写 OneDrive 中的工作簿、工作表、区域、表格与图表。
编写、调试与调优 Playwright 测试,涵盖定位器策略、追踪诊断与 CI 友好的超时配置。
以 AI 机器人身份加入视频会议,提供语音、虚拟形象与屏幕共享四种模式。
执行 Git 操作(提交、分支、合并、变基、冲突解决与恢复)时强制套用安全规则。
从 AdMapix API 拉取广告创意、应用、榜单和收入预估等数据,原样返回结构化 JSON。
tenequm 的更多技能
浏览全部技能TanStack Query、Router、Start 在 React 全栈项目中的类型安全参考模式。
用 MPP 协议在 HTTP 402 上做机器对机器支付,覆盖 TypeScript、Python、Rust 三套 SDK。
构建高质量 Agent Skills 的实操指南,覆盖 SKILL.md 结构、frontmatter、描述写法与单文件 / references/ 取舍。
Lance v11.0.0-beta.6 与 Rust/Python 引擎的固定版本参考资料,同时覆盖 v10.0.0 稳定线。
用 Wrangler CLI 在 Cloudflare 全球边缘网络上开发并部署 JavaScript、TypeScript、Python 或 Rust 代码。
用 Swift 6.3 构建原生 macOS 应用,覆盖 SwiftUI、SwiftData、并发与端侧 AI。