Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Power BI custom visual (.pbiviz) development with the pbiviz toolchain and its MCP server. Automatically invoke when the user mentions "custom visual", "pbiviz", "develop a Power BI visual", "powerbi-visuals-tools", "IVisual", "capabilities.json", "visual formatting model", "visual certification", "publish a visual to AppSource", or asks to scaffold, build, debug, package, certify, or publish a .pbiviz developer visual.
.claude/skills/bilal140202-powerbi-custom-visuals/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 122% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 55% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 43% | 0% |
Build distributable Power BI developer visuals: TypeScript/D3 visuals packaged as a .pbiviz file, written against the powerbi-visuals-api and built with the pbiviz CLI (powerbi-visuals-tools). This skill covers the full loop: scaffold, develop, live-preview, package, certify, and publish.
This plugin holds five visual approaches. Choose before building:
yamldeneb-visuals: no-code Vega / Vega-Lite specs inside a report; no packaging, no TypeScript python-visuals: matplotlib / seaborn script visuals (static image, computes at render) r-visuals: ggplot2 script visuals (static image); R's stats ecosystem svg-visuals: SVG drawn by a DAX measure (ImageUrl); inline in tables/cards, no visual sandbox powerbi-custom-visuals (this skill): a real .pbiviz developer visual, reusable across reports and shareable on AppSource; full interactivity, formatting pane, certification
Reach for this skill when the need is a reusable, interactive, distributable visual with its own formatting pane. For a one-off chart in a single report, a sibling skill is usually faster and lighter.
powerbi-visuals-tools engine minimum)pbiviz CLI: run via npx -y powerbi-visuals-tools <command>, or install it globallyturn on developer mode in the Power BI service
pbiviz new <VisualName> creates the project (src/visual.ts,capabilities.json, pbiviz.json, tsconfig.json, package.json, style/, assets/)
src/visual.ts. Implement the IVisual interface;the class name must equal visualClassName in pbiviz.json. Core methods: constructor (set up DOM and services), update (render on data/size/settings change), getFormattingModel (populate the format pane), destroy (cleanup)
capabilities.json: dataRoles (the field wells),dataViewMappings (how roles map to the dataView), and objects (format-pane properties)
powerbi-visuals-utils-formattingmodel(FormattingSettingsService, cards, slices). A card name matches a capabilities.json object; a slice name matches a property
pbiviz start, then add the developer visual to a report and iterateagainst real data
pbiviz package produces the .pbiviz under dist/; import it into a reportFull command flags, the project layout, the IVisual lifecycle, and formatting-model patterns are in references/pbiviz-toolchain.md.
This plugin ships a .mcp.json that starts the pbiviz MCP server (npx -y powerbi-visuals-tools mcp). It grounds development in current, correct API usage instead of guessed code. Prefer its tools while building:
yamlget_available_apis: look up data, formatting, interaction, and utility APIs with examples get_best_practices: API-version management, performance, security, accessibility, testing add_feature: list features available to add (bookmarks, tooltips, drill-down, selection, localization) implement_feature: step-by-step instructions, code templates, and config changes for a feature list_visual_info: extract a project's GUID, version, author, data roles, dependencies check_vulnerabilities: scan dependencies and source for risky patterns (eval, innerHTML) prepare_certification: verify required files, configuration, capabilities, and assets pre-submission
Setup, client config, and per-tool detail are in references/mcp-server.md.
Certification unlocks export to PowerPoint and email subscriptions, and signals a reviewed visual. Key gates: a lowercase certification branch matching the package, eslint-plugin-powerbi-visuals passing, no eval/fetch/XMLHttpRequest, a clean npm audit, and the latest API. Audit with pbiviz package --certification-audit. Publishing goes through Partner Center to AppSource. R-based visuals cannot be certified.
Requirements and the submission flow are in references/certification-publishing.md.
Microsoft publishes the full source of official and certified visuals. Start from the tutorials (circlecard, sampleBarChart) and read a production visual close to the target (gantt, sankey, sunburst, timeline, ChicletSlicer) for real patterns. A curated list with what each one teaches is in references/community-examples.md.
references/pbiviz-toolchain.md ... CLI commands, project structure, IVisual lifecycle, capabilities.json, formatting model, key npm packagesreferences/mcp-server.md ... the bundled pbiviz MCP: setup and its toolsreferences/certification-publishing.md ... certification requirements and AppSource publishingreferences/community-examples.md ... official and certified example visuals worth readingexamples/vscode-mcp.json ... standalone .vscode/mcp.json for a visual project outside this pluginexamples/capabilities-objects.json ... a capabilities.json objects block wired to a formatting card| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | pass→pass | 9,879 | 8,990 | -9% | 1 | 1 | 0% | 2,084 | 2,894 | +39% | 0 | 0 | — |
case-02 | pass→pass | 10,311 | 7,262 | -30% | 1 | 1 | 0% | 1,998 | 2,803 | +40% | 0 | 0 | — |
case-01 | fail→pass | 16,289 | 10,014 | -39% | 1 | 1 | 0% | 2,849 | 3,074 | +8% | 0 | 0 | — |
case-04 | pass→pass | 12,800 | 8,951 | -30% | 1 | 1 | 0% | 2,481 | 3,176 | +28% | 0 | 0 | — |
case-05 | pass→pass | 14,058 | 7,961 | -43% | 1 | 1 | 0% | 2,353 | 2,868 | +22% | 0 | 0 | — |
case-06 | fail→pass | 4,508 | 3,261 | -28% | 1 | 1 | 0% | 923 | 2,049 | +122% | 0 | 0 | — |
case-07 | pass→pass | 6,746 | 2,723 | -60% | 1 | 1 | 0% | 1,327 | 1,870 | +41% | 0 | 0 | — |
case-08 | pass→pass | 8,205 | 5,932 | -28% | 1 | 1 | 0% | 1,627 | 2,517 | +55% | 0 | 0 | — |
case-09 | pass→pass | 2,957 | 3,633 | +23% | 1 | 1 | 0% | 542 | 2,042 | +277% | 0 | 0 | — |
case-10 | pass→pass | 12,987 | 8,973 | -31% | 1 | 1 | 0% | 2,683 | 3,239 | +21% | 0 | 0 | — |
case-11 | pass→pass | 2,899 | 3,106 | +7% | 1 | 1 | 0% | 504 | 1,890 | +275% | 0 | 0 | — |
case-12 | pass→pass | 3,825 | 2,809 | -27% | 1 | 1 | 0% | 732 | 1,851 | +153% | 0 | 0 | — |
case-13 | pass→pass | 4,776 | 2,150 | -55% | 1 | 1 | 0% | 999 | 1,768 | +77% | 0 | 0 | — |
case-14 | fail→pass | 6,361 | 2,538 | -60% | 1 | 1 | 0% | 1,187 | 1,845 | +55% | 0 | 0 | — |
case-15 | fail→pass | 14,699 | 4,958 | -66% | 1 | 1 | 0% | 3,099 | 2,316 | -25% | 0 | 0 | — |
case-16 | pass→pass | 12,093 | 9,150 | -24% | 1 | 1 | 0% | 2,138 | 3,031 | +42% | 0 | 0 | — |
case-17 | pass→pass | 6,563 | 2,249 | -66% | 1 | 1 | 0% | 1,074 | 1,681 | +57% | 0 | 0 | — |
case-18 | pass→pass | 3,537 | 3,037 | -14% | 1 | 1 | 0% | 486 | 1,652 | +240% | 0 | 0 | — |
case-19 | pass→pass | 8,518 | 6,916 | -19% | 1 | 1 | 0% | 1,514 | 2,706 | +79% | 0 | 0 | — |
case-20 | fail→pass | 6,174 | 1,619 | -74% | 1 | 1 | 0% | 1,112 | 1,593 | +43% | 0 | 0 | — |
case-21 | fail→pass | 6,954 | 2,317 | -67% | 1 | 1 | 0% | 1,338 | 1,793 | +34% | 0 | 0 | — |
case-22 | fail→pass | 7,914 | 3,128 | -60% | 1 | 1 | 0% | 1,441 | 1,922 | +33% | 0 | 0 | — |
case-23 | fail→pass | 7,017 | 1,721 | -75% | 1 | 1 | 0% | 1,382 | 1,704 | +23% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +35 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.