Build IPPF-branded Power BI artefacts from analytical input. Use when Ane asks to build, scaffold, or update a Power BI dashboard, semantic model, DAX measures, or .pbip project. Applies IPPF Visual Identity 2025 brand, MEL-standard DAX measures, and three standard page recipes (indicator dashboard,

/power bi build Build IPPF branded Power BI artefacts. v1 scope: writes DAX measures into the semantic model via the pbi CLI; writes the IPPF theme JSON into the .pbip folder directly; saves page recipes (indicator dashboard, equity, methodology) as JSON staging artefacts the user applies manually in Power BI Desktop. Full page and visual automation is v2. Tooling reality The pbi CLI (package pbi cli tool , command name pbi ) wraps a Power BI MCP server for semantic model operations only: ✅ pbi measure create — add DAX measures. ✅ pbi table , pbi column , pbi relationship — model structure. ✅ pbi dax — execute and validate DAX. ❌ No pbi theme command. Theme JSON is written directly to the .pbip folder. ❌ No pbi page or pbi visual command. Pages and visuals are created manually in Power BI Desktop using the staging recipe JSONs as a guide. When the upstream tool adds report layer commands, this skill will absorb them. v1 ships with the scope above. Two routes. The v1 flow below ( pbi CLI for measures + theme + staged page recipes) suits an existing model the user keeps building by hand. When the source is an Excel workbook and the user wants a complete, working dashboard handed over (data loaded, measures, theme, and report pages all built), use the full .pbip authoring route in the dedicated section below instead. That route was validated end to end (CERV Portfolio Dashboard, 2026 06 20). Pre flight Run these checks in order. Stop with the indicated message on first failure. 1. pbi installed. Run: pipx list grep pbi cli tool . If empty: stop. Show: "pbi CLI is not installed. Run: pipx install pbi cli tool && pipx ensurepath && pbi skills install . Open a fresh shell so PATH updates pick up." 2. Power BI Desktop running with an active project, connection named. Run pbi connect (bare) once to read the port from the "Auto detected Power BI Desktop on localhost:<port " line. Disconnect: pbi disconnect . Re connect with an explicit name matching the MCP server's internal label: pbi connect d localhost:<port n PBIDesktop <pbip basename <port . The model derived name is required — pbi cli's default save name localhost <port is rejected by the MCP server's measure/table operations with "Connection 'localhost <port ' not found" . If pbi connect reports no active project: stop. Show: "Open Power BI Desktop and load a .pbip project, then re run." If the Analysis Services port shifts mid session (Power BI Desktop sometimes restarts the engine on file reopen, Model view switch, etc.), re run this step with the new port. 3. Active project is .pbip (text format), not legacy .pbix binary. The pbi connect output includes the project path. If it ends in .pbix : stop. Show: "Save your project as .pbip first — File → Save as → set Save as type to Power BI Project. The skill operates on the text based PBIP format only." 4. Brand layer importable. Verify ${WORK FOLDER ROOT}/ane package/reporting/powerbi dashboard/ init .py exists. If missing: stop. Show: "Brand layer not found at ${WORK FOLDER ROOT}/ane package/reporting/powerbi dashboard/ . Verify OneDrive sync." Resolve intent From the prompt and the most recent analytical artefact in this session, resolve: Indicators — keys from MEL DAX LIBRARY . If ambiguous, ask Ane: "Which indicators? Available: {sorted(MEL DAX LIBRARY.keys())}". Audience tier — Tier 1 working brief by default; Tier 2 publication only if the prompt names it. Page set — default (indicator dashboard, equity disaggregation, methodology card) per indicator. Source line — derive from prior artefact; if absent, ask once. Staging directory — default ${PWD}/powerbi build output/ . Holds theme.json, measures.json, and page recipes for the run. Build (Python — produces all artefacts in memory) Apply 1. DAX measures via pbi measure create For each measure, infer the table from the DAX expression's [bracketed] references (heuristic: first Table[Column] reference in the expression, or Measures if the expression only references other measures). The measure name is positional (NOT a name flag). Verified against pbi cli tool 0.5.6 . Surface any error verbatim. Never retry — pbi failures usually mean the table or column referenced does not exist in the model. pbi cli rolls back the transaction on any "table not found"; the measure does NOT land with a red icon (contrary to older Power BI behaviour). Either pre seed a stub table or correct the DAX before retry. 1a. Save discipline — mandatory after EACH pbi model write Tell Ane to switch to Power BI Desktop and press Ctrl+S immediately after each pbi table / pbi measure / pbi column / pbi relationship write. External pbi writes are held in the Analysis Services engine's memory only; the .pbip on disk is updated by Power BI Desktop on File → Save. If the engine restarts before save (port shift, file reopen, sometimes Model view switch), the writes are lost. 2. Theme JSON — direct write into the .pbip folder The .pbip is a text format project. The theme JSON goes into the report folder as a custom theme. Path layout (Power BI Desktop 2.140+): Write the theme JSON to both paths. Then update report.json to register it (Power BI Desktop reads the theme name from report.json metadata; without registration the theme appears in the Themes gallery but does not auto apply). If the user has not enabled the PBIP report format preview in Power BI Desktop (File → Options → Preview features → Power BI Project (.pbip) source control), the theme path may differ. Surface a clear message if <project .Report/ does not exist. Cache bust on retry. Power BI Desktop caches a registered theme by filename across failed imports — re importing the same filename re shows old validation errors even after the file is fixed. On any retry after a validation failure, write the new theme JSON under a versioned name ( ippf visual identity 2025 v2.json , v3 , etc.) so Power BI parses it as new. 3. Page recipes — staging only in v1 Save each page recipe to the staging directory as JSON. Do NOT attempt to write PBIR page files into the .pbip folder in v1 (PBIR format is still in preview and the schema is unstable). Tell Ane in the chat output exactly which file holds which page recipe and which visuals to create manually: v2 will automate page creation once the upstream pbi cli adds report layer support, or once we wire a direct PBIR writer. Full .pbip authoring from Excel (full handover route) Use when the source is an Excel workbook and the user wants a complete working dashboard, not staged recipes. Author the entire .pbip from a Python generator (one re runnable script), because the pbi CLI cannot import Excel data (DataSourceOperationsTool is skipped in PowerBI compat mode) and has no report page/visual API. Reference build: ${WORK FOLDER ROOT}/scripts/gen cerv dashboard pbip.py . Apply mel wiki/wiki/concepts/edit preservation protocol.md when the target .pbip already exists. Author into a NEW project folder if the user has the target open (avoids the file lock); they close without saving and reopen. Procedure: 1. Mirror the schema. Read an existing empty .pbip 's skeleton and copy its exact versions (compatibilityLevel 1600; report.json 3.3.0; visual 2.9.0; page 2.1.0; pbism 4.2). Require the PBIP + PBIR preview features enabled in Desktop. 2. Inspect the Excel with openpyxl: sheet names, exact header strings (no stripping — trailing spaces break column refs), header row position, and value types per column. Reconcile expected aggregates against any existing in workbook dashboard tabs. 3. Author the SemanticModel TMDL (TAB indented): database.tmdl , model.tmdl (with ref table X lines + relationship blocks), cultures/ , and tables/ .tmdl (columns + measures + an M partition ). The M block is indented deeper than source = . 4. Author the report PBIR : definition.pbir (relative byPath to the SemanticModel — this link survives folder moves), report.json (register the IPPF custom theme + base theme), pages/<id /page.json , and pages/<id /visuals/<id /visual.json . 5. Reconcile measures against source aggregates, then have the user open + Refresh. After load, verify via the CLI: pbi connect , then pbi dax execute "EVALUATE <Table " and read RowCount= (the CLI does NOT print cell values; use a table EVALUATE and read the row count). Power Query M gotchas (each cost a refresh error cycle — apply pre emptively): Excel.Workbook(File.Contents(path), null, true) drops leading fully blank rows , so a fixed Table.Skip(n) misaligns the header. Locate it dynamically: HeaderRow = List.PositionOf(Sheet[Column1], "<key header ") , then Table.Skip(Sheet, HeaderRow) , then Table.PromoteHeaders(.., [PromoteAllScalars=true]) . It is List.PositionOf (list + value), NOT Table.PositionOf (table + record) — the latter throws "cannot convert List to Table". Only hard type columns the dashboard needs (amounts → type number , counts → Int64.Type ). Typing computed / "(auto)" columns throws per cell "N errors" when they return non numeric ("" / "Pending") for not yet filled rows. Leave those untyped. Multi row / merged header sheets: skip PromoteHeaders; use positional Table.Range + Table.SelectColumns({"Column1",...}) + rename. M file paths are literal (no backslash escaping). Put the source path once in a shared expression or inline consistently. PBIR visual JSON (authored blind, all rendered correctly): minimal visual.query.queryState.<role .projections[{field:{Measure Column:{Expression:{SourceRef:{Entity}},Property}}, queryRef:"Entity.Prop", nativeQueryRef:"Prop"}] . Roles: card → Values ; clusteredColumnChart / clusteredBarChart → Category + Y (omit Category for a multi measure chart); tableEx → Values (list). Omit title objects — let the IPPF theme and auto titles style everything. Extras: slicer multi select + Select all = objects.selection[{properties:{singleSelect:{Literal false}, selectAllCheckboxEnabled:{Literal true}}}] ; nav bar = visualType pageNavigator ; logo = visualType image with ResourcePackageItem(RegisteredResources) + copy the PNG into StaticResources/RegisteredResources/ + register in report.json; header band = textbox with visualContainerObjects.background . Layout that reads well (1280×720): compact full width slicer strip (slicers render as dropdowns), KPI cards row, then large charts spanning the full width — avoid a part width slicer row that wastes the band to its right. Set page displayOption to FitToWidth so it fills the monitor and scrolls vertically. Distribution & licensing (advise the user) .pbip is a dev/source control format — it opens empty until refreshed , wrong for novices. For novice viewers, hand out a .pbix (File → Save As → .pbix embeds the data, opens populated, needs no licence). Only Power BI Desktop can write a .pbix ; no CLI/script can, and embedding data needs a live refresh first. Power BI Service publishing lets free tier colleagues view ONLY if the workspace is on Premium (P SKU) or Fabric F64+ capacity ; otherwise every viewer also needs Pro/PPU . Never use "Publish to web" for sensitive (SRHR / finance / sub grantee) data — it is public. The page navigator and buttons navigate on single click in Reading view / published , Ctrl+click in Editing view (inherent Power BI behaviour, not a bug). An Excel source referenced by absolute local/OneDrive path breaks if moved to SharePoint . Repoint via the SharePoint connector for cloud scheduled refresh; refresh re reads the source each time. Output (Tier 1 working brief) Common errors Error Likely cause Fix pbi: command not found after install PATH not updated yet p

power-bi-build