Generated banner for Powerbi Report CLI
Generated logo for Powerbi Report CLI

Powerbi Report CLI

MicrosoftOfficialPublished by Microsoft

  • Skill
  • Tool integration
  • Data and AI
  • MIT
  • Low risk

From the author

Plan, design, author, preview, publish, and manage Power BI reports across requirements, page design, local PBIR/PBIP edits, validation, screenshots, Fabric upload/download, and rebinding. Use for report lifecycle work; use semantic-model-authoring for model or DAX changes and fabriciq for data questions. Triggers: plan Power BI report, design report page, edit PBIR, preview PBIP, publish report, rebind report
— the description by Microsoft

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

The powerbi-report-cli skill provides a structured, mode-based framework for managing the entire Power BI report lifecycle. It acts as a dispatcher across four distinct phases: planning (requirements and scope), design (visual and layout strategy), authoring (local PBIR/PBIP file manipulation and validation), and management (Fabric item transport and rebinding). The skill enforces strict boundaries between these modes to ensure quality, requiring explicit approval for new reports and utilizing a dedicated CLI for metadata-driven validation and authoring tasks.

What it does

  • Guides users through report requirements, scope, and page planning.
  • Provides design contracts for visual archetypes, layouts, and branding.
  • Enables local editing, validation, and previewing of PBIR and PBIP files.
  • Handles report item transport to and from Microsoft Fabric.
  • Supports re-theming, conditional formatting, and visual configuration.
  • Enforces a mandatory approval gate for greenfield report development.

When to use it

  • When planning a new Power BI report from a semantic model.
  • When you need to define a professional design brief or layout strategy.
  • When editing local PBIR/PBIP files, including visuals, filters, and themes.
  • When validating report structure and formatting before deployment.
  • When publishing or managing report items within a Fabric workspace.

Advantages

  • Ensures consistent report quality through a structured, multi-mode workflow.
  • Prevents premature publishing by enforcing a mandatory approval barrier.
  • Uses a dedicated CLI to eliminate guesswork in PBIR schema and property editing.
  • Maintains clear separation between design intent and technical implementation.
  • Provides robust validation tools to catch errors before Fabric deployment.

Usage examples

“Create a new executive report from this semantic model.”
The agent enters planning mode, inspects the model, generates a report spec, and requests user approval before proceeding.
“Design an operational page and choose the right charts.”
The agent enters design mode, produces a design brief with page archetypes and chart recommendations.
“Add a KPI page to this PBIP, validate it, and take screenshots.”
The agent enters authoring mode, modifies the local PBIP files, runs validation, and captures screenshots for review.
“Publish this local PBIP to my Fabric workspace and rebind it.”
The agent enters management mode, performs the upload to Fabric, and completes the rebind operation.

Install

  1. Download the skill as a ZIP:

    https://mp-staging.autogineer.com/agent-skills/microsoft/powerbi-report-cli/download
  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip powerbi-report-cli.zip -d .agents/skills

Documentation by Microsoft

Telemetry — MANDATORY. Every api.fabric.microsoft.com call must carry x-ms-fabric-skill: powerbi-report-cli (az rest: --headers "x-ms-fabric-skill=powerbi-report-cli"), including every LRO poll, fabric_lro and retry. Snippets omit it — add it anyway.

CRITICAL NOTES

  1. To find workspace details, including its ID, from a workspace name: list all workspaces, then use JMESPath filtering.

  2. To find item details, including its ID, from a workspace ID, item type, and item name: list all items of that type in that workspace, then use JMESPath filtering.

  3. Use powerbi-report-cli for the report artifact: requirements, design, PBIR/PBIP page and visual edits, preview, and publishing. Semantic model, measure, and DAX work belongs to a semantic-model skill. Natural-language questions about report data belong to fabriciq.

Power BI Reports - CLI Skill

This skill owns the Power BI report lifecycle end to end: requirements and scope, visual design, local PBIR/PBIP authoring and preview, and report-item transport to and from Microsoft Fabric.

It is a mode dispatcher and intentionally contains no detailed procedures. Select the mode that matches the request, then read the matching mode reference end to end before acting. The mode file holds the required workflow, commands, payloads, templates, and failure rules.

Mode Selection

ModeUse when the request...Existing route examplesRead first
planningcreates a new report/dashboard and needs requirements, model/dependency inspection, scope, a page plan, and approval before implementationcreate a report, build a dashboard, create from a semantic model, plan then implement, walk me through creating a reportreferences/planning.md, then references/planning-part-02.md
designdecides what a report should look like: tone, signature, page archetype, chart type, layout, color, typography, theme direction, accessibility, brand application, redesign, or critiquedesign a Power BI report, make the dashboard professional, choose a chart, apply a brand, redesign the report, create a design briefreferences/design.md
authoringreads or edits local PBIR/PBIP files, validates them, previews the report, or captures screenshotsedit PBIR, create/add a report page or visual, format a visual, add filters/slicers/bookmarks/themes, validate PBIR, preview/reload/screenshot in Desktop or servicereferences/authoring.md
managementmoves a report definition to or from Fabric or manages the workspace report itempublish/upload/download PBIR or PBIP, list reports, get/update/delete a report, rebind a reportreferences/management.md

Mode Boundary Rules

  • Classify by intent, not by which file or tool is already open.

  • design decides what the report should look like; authoring writes the PBIR that realizes it. Choosing a chart type is design; encoding it into visual.json is authoring.

  • Restyling or reformatting an existing local report is authoring, even when the request uses design language. Use design when the user wants advice or a design contract rather than a file change.

  • authoring changes local files and performs preview/validation. It does not publish to Fabric.

  • management transports definitions and manages Fabric report items. It does not invent or directly author PBIR content.

  • planning owns the guided requirements-to-approval flow for a new report. A focused edit to an existing report goes directly to authoring.

A greenfield request can span modes in this order:

planning -> design -> authoring -> management

Handle one mode at a time. Announce each switch and read the new mode reference before starting that part. The planning approval gate is a turn boundary: never build or publish until the user explicitly approves the locked spec in a later reply.

GREENFIELD APPROVAL BARRIER — MANDATORY. When a new-report request asks to plan, build, and/or publish but the user has not approved a locked spec in a prior reply, set the current phase to approval_pending and keep planning as the only active mode. In this phase, do not load authoring or management, edit/scaffold PBIR or semantic-model files, generate synthetic data, call Fabric write APIs, or publish. Persist _brief/report-spec.md, ask for explicit approval, and end the turn. The build/publish wording in the original request is desired future scope, not approval to cross this barrier.

Required Deliverable by Mode

ModeCompletion requirement
planningPersist _brief/report-spec.md or the user-named equivalent, ask for explicit approval, and stop before implementation.
designProduce the complete Design Brief: contract, including design_identity and a page archetype/layout contract for every page. Do not edit PBIR or call Fabric APIs.
authoringPersist the requested PBIR/PBIP changes, validate each logical batch, then complete the selected preview and screenshot-review workflow required by the authoring reference. Do not publish.
managementComplete the requested Fabric report operation, including LRO polling and documented readback verification.

Rules

MUST
  • Select the narrowest mode that fully covers the current request.

  • Read references/<mode>.md end to end before the first command or file edit in that mode. When an already-loaded mode explicitly routes one narrow cross-mode operation to a named topic reference, treat that topic reference as the complete handoff; do not load the target mode's umbrella file unless the work expands beyond that operation.

  • In planning mode, also read references/planning-part-02.md before producing _brief/report-spec.md; it contains the required canonical design contract and approval checks.

  • Across a multi-turn task, retain a compact working state containing resolved workspace/item/model IDs, local and downloaded paths, and references already read. Reuse it in later turns unless the user explicitly requests a fresh query or the preceding operation invalidated a value.

  • Preserve every argument and safety boundary documented by the selected mode.

  • Announce and perform an explicit mode switch when the request crosses a boundary.

  • Use the powerbi-report-author CLI metadata and validation surfaces instead of guessing PBIR schemas, roles, formatting properties, selectors, or enum values.

  • Keep local edits local unless the user explicitly requests publishing.

  • Resolve Fabric workspace and item IDs by listing and filtering rather than guessing identifiers.

  • Produce the artifact the user requested; reading or summarizing a reference is not task completion.

PREFER
  • Infer facts already present in the prompt, semantic model, existing PBIP, or prior approved spec instead of asking again.

  • Load only the selected mode and its directly relevant topic references.

  • Reuse reference guidance already loaded in the current task. Open a new reference only when the requested operation or an observed mismatch requires it.

  • Preserve the user's existing project structure, schemas, host, operation, binding, and delivery target unless the selected mode explicitly requires a change.

AVOID
  • Acting from this dispatcher without reading the selected mode reference.

  • Reloading this dispatcher, a mode file, or a topic reference already read in the current task solely because the conversation advanced to a later turn.

  • Loading a retired sibling skill name; planning, design, authoring, and management are modes of this skill.

  • Editing PBIR or calling Fabric APIs in planning or design.

  • Publishing from authoring, or authoring PBIR content from management.

  • Bypassing the planning approval gate because the original request also asked to build or publish.

<!-- BEGIN GENERATED REFERENCE INDEX -->

Complete Reference Index

Open every relevant reference directly from this index. Files named part-XX continue the named topic.

Authoring
Design
Mode Guides

<!-- END GENERATED REFERENCE INDEX -->

Examples

User requestMode
"Create a new executive report from this semantic model."planning
"Design an operational page and choose the right charts."design
"Add a KPI page to this PBIP, validate it, and take screenshots."authoring
"Publish this local PBIP to my Fabric workspace and rebind it."management

Route Compatibility

The merged skill preserves the four previous route surfaces. Treat the following phrases as explicit aliases for the corresponding mode:

ModePreserved route phrases
planningcreate a report; build a report; create a dashboard; create a new report from scratch; create a report from a semantic model or dataset link; build a report for this model; make me a dashboard; plan then implement; walk me through creating a report
designdesign Power BI report; make dashboard look professional; choose chart type; apply brand to report; redesign report; create design brief; Power BI report design archetype
authoringedit PBIR; create Power BI report page; add visual to PBIP; format report visual; validate Power BI report; preview Power BI report; Power BI Desktop preview status; preview report in service; preview report in browser; screenshot report page; screenshot all report pages; list preview hosts; implement dashboard in PBIP; add dashboard visuals to report; implement an approved PBIP report spec; edit PBIR pages/visuals; edit a Power BI report
managementupload PBIR report definition; upload Power BI report; download PBIR definition; publish PBIR definition; publish Power BI report to Fabric; manage Power BI reports; list workspace reports

Files

references/authoring.md

Contents

  • Critical notes

  • Must/Prefer/Avoid

    • MUST

    • PREFER

    • AVOID

  • Quick Start Workflow

<!-- Mode reference for the `powerbi-report-cli` skill. Loaded on demand from `skills/powerbi-report-cli/SKILL.md` when the request matches the `authoring` mode. -->

AUTHORING COMPLETION GATE — MANDATORY FOR REPORT EDITS

Schema validation and screenshot validation are paired. Whenever this skill runs powerbi-report-author validate on a report, a successful schema result MUST immediately continue to loading the validated PBIR, capturing every affected page, and reviewing the screenshots. Schema success alone is never task completion, even when the user did not explicitly request a screenshot.

Any task that creates or modifies rendered report content is incomplete until:

  1. powerbi-report-author validate "<folder>" succeeds.

  2. The latest edited PBIR is loaded into the selected preview host.

  3. Every affected page is captured to a screenshot.

  4. The screenshot files are reviewed using references/authoring/screenshot-review.md.

  5. Visible issues are fixed and the validate → load → screenshot → review loop is repeated.

Capture each page changed by page, visual, filter, slicer, bookmark, or layout edits. Use --all-pages for report-wide changes such as themes or global formatting. Do not report the authoring task complete without successful, fresh screenshot review. If the latest PBIR cannot be loaded, the affected pages cannot be captured, or the screenshots cannot be reviewed, report screenshot validation as blocked rather than claiming completion.

SCREENSHOT ARTIFACT LIFECYCLE — BEFORE THE FIRST MANDATORY CAPTURE

Before constructing or running the first screenshot command for mandatory authoring validation, read and follow Screenshot artifact lifecycle (see authoring/preview-part-03.md, section screenshot-artifact-lifecycle). Classify a supplied location by intent: save/export/keep/retain paths are retained output and are never the temporary validation parent or cleanup target; only a location explicitly supplied for temporary validation replaces the default parent. Otherwise resolve Windows Local Application Data and automatically use its Power BI Report Authoring\Screenshots directory. Do not ask the user for a location or permission to save mandatory validation screenshots. Apply the lifecycle's exposure warning and explicit-confirmation requirements to any risky user-supplied location. Review mandatory-validation screenshots from the workflow-owned child and delete only that child before every handled exit.

After screenshot review passes on the service host, automatically display the validated report in a visible service preview, even when the user did not separately request it. Reuse the same folder, workspace GUID, and semantic- model GUID. Run visible-mode service status first, then open/attach under the normal safety contract. A headless validation instance and a visible instance are separate; do not change or repeat the completed screenshot operation. Leave the visible preview open for the user. If display fails, preserve the successful validation result but report visible service display as blocked. This automatic display step is service-only. It adds no Desktop command and does not change Desktop status, open/attach, reload, screenshot, failure, or lifecycle behavior.

Loading and screenshot capture are separate CLI invocations. Never combine --reload and --screenshot in one command; rendering and capture together can exceed the 30-second command timeout. For authoring verification, follow the selected host's status → open/attach or reload → screenshot flow. If the selected host reports unsaved changes and blocks loading the latest PBIR, do not use a stale screenshot to validate the edits.

When the selected host's status reports explicit boolean details.hasUnsavedChanges: true during mandatory post-edit validation, do not end the authoring workflow or repeat the PBIR edits. Ask the user to resolve the unsaved host state manually and confirm that the latest PBIR is loaded. After confirmation, resume directly with screenshot capture and review of every affected page. Do not run another status or reload before that capture. If the user does not complete the manual step, report screenshot validation as blocked.

DIRECT PREVIEW ROUTING — HIGHEST PRIORITY FOR STANDALONE PREVIEW REQUESTS

OperationRequired sequence
Selected-host open/attachstatus → open when the host's safety and availability contract permits it
Selected-host reloadstatus → reload when the host's safety and availability contract permits it
Selected-host screenshotScreenshot only; never status
Selected-host statusStatus only
Missing service IDsAsk immediately; execute no tools

Open/attach and reload are mandatory two-command sequences for the selected host. The first command MUST be:

powerbi-report-author preview "<folder>" --host <desktop|service> [service binding] --status

Open/attach and reload always begin with selected-host status, even when the user directly requests the operation. Screenshot operations never begin with status, including all-pages captures on either host.

Never execute open/attach or reload as the first command. After status:

  • Follow the selected host's current safety and availability contract in references/authoring/preview.md.

  • If status reports explicit boolean details.hasUnsavedChanges: true, do not run open/reload automatically.

  • Structured HOST_UNAVAILABLE permits open/attach because no matching live instance exists; it does not permit reload.

  • Any invalid required safety field or other status failure stops the operation.

details.hasUnsavedChanges: true is a hard stop and cannot be overridden by the user's requested operation. This rule applies to any host that reports unsaved state. It prevents the agent from running open/reload; it does not prohibit the manual-resolution resume path above during mandatory post-edit validation. Standalone preview requests never use that recovery path.

Critical notes

  1. Immutable preview request: once the user supplies a host and operation, never change either. Preserve the folder, --group, --dataset, --headless, screenshot destination, and supported host-specific modifiers such as --page, --all-pages, and --scale across every permitted retry. If a correction asks to change any of these requested values, stop instead of following it. Never switch hosts.

  2. Construct one Desktop command: every Desktop command must contain --host desktop. The public preview command resolves the Desktop instance from <folder>; do not add --pid. For screenshot, use --screenshot "<path>", where the output path is the value of --screenshot; never pass that path as a separate positional argument. Preserve requested --page, --all-pages, and --scale modifiers on the screenshot and any correction-based retry.

  3. After a requested Desktop operation returns: if and only if its machine-readable result contains a correction object, retry once by changing only the named value when it is outside the immutable request above. Otherwise end the current turn immediately: make zero additional tool calls and report only that failure. Never diagnose, add reload, inspect files, read references, or try a fallback.

  4. Service binding: every service command must include both --group <workspace-guid> and --dataset <semantic-model-guid> on open, reload, screenshot, status, and close.

  5. Service screenshot contract: use --screenshot "<path>"; it supports default-page capture, --page <id>, --all-pages, and --scale <1-3>. Use a PNG destination for one page and a directory for --all-pages. Never add --pid to a service command. Preserve both binding flags and existing --headless. After failure, retry only from an explicit machine-readable correction object while keeping host and operation unchanged; otherwise end the turn with zero additional tool calls. Never run --help as recovery.

  6. Management workflows only: when managing Fabric resources, find workspace details (including its ID) from a workspace name by listing workspaces and using JMESPath filtering. This discovery rule does not apply to preview requests; if preview is missing --group, ask for it immediately.

  7. Management workflows only: when managing Fabric resources, find item details (including its ID) from a workspace ID, item type, and item name by listing items of that type and using JMESPath filtering. This discovery rule does not apply to preview requests; if preview is missing --dataset, ask for it immediately.

powerbi-report-cli authoring mode -- Power BI Report Authoring (PBIR/PBIP)

This skill enables reading, editing, and creation of Power BI report definition files in the PBIR (Power BI Report) format used by PBIP (Power BI Project) files.

Must/Prefer/Avoid

MUST
  • Desktop terminal rule (highest priority): after executing a selected Desktop attach/open, reload, status, or screenshot command, a failure without an explicit machine-readable correction object ends the current turn. Emit zero further tool calls. Parser errors, usage text, hints, suggested help, and unknown-option messages are not structured corrections. This rule overrides all host discovery, status, troubleshooting, retry, validation, file inspection, reference-reading, and general error guidance. The sole exception is structured HOST_UNAVAILABLE from the mandatory status preflight for open/attach, which proceeds to the plain open/attach command as specified in the top-level routing table; the same result remains terminal for reload.

  • The Desktop terminal rule covers every screenshot form: default page, --page, --all-pages, --scale, and any combination allowed by the CLI. Never alter the folder, screenshot destination, page, scale, or all-pages choice after failure.

  • Desktop screenshot uses --screenshot <path>; the output path is the value of --screenshot. It is a warm capture of the current Desktop window and does not reload PBIR. Service screenshot uses the same --screenshot <path> flag and supports default-page, selected-page, all-pages, and scale capture.

  • Every preview command executed for a screenshot request must retain that host's exact screenshot operation and destination. This includes the only permitted correction-based retry.

  • Use this skill only for concrete PBIR/PBIP report-file mechanics such as pages, visuals, filters, slicers, navigation, bookmarks, themes, formatting, validation, previews, and screenshots.

  • Validate PBIR with powerbi-report-author validate after each logical batch. Every successful schema validation must continue into fresh screenshot capture and review; never stop at the validation result.

  • Use the powerbi-report-author preview command for rendered-output inspection. Always pass --host; never rely on the command's default host.

  • Classify the Desktop operation from the user's verb. Attach, open, connect, or show means the plain command with no lifecycle flag. Add --reload only when the user explicitly requests reload or refresh. Never translate attach/open into reload because the host may synchronize files internally.

  • After a requested Desktop operation fails, retry once only when its structured result contains an explicit machine-readable correction object. Apply only its named correction to the same host and operation and preserve every other selected argument. Never follow a correction that switches Desktop to service or changes the operation. In particular, never add --reload to attach or drop screenshot arguments. Otherwise stop and report the failure immediately with no additional tool call. Do not diagnose the failure by running --help, --list-hosts, --status, --reload, another preview operation, validation, a different host, or powerbi-desktop.

  • Follow the top-level Direct preview routing table exactly. Detailed command construction and error handling for standalone preview requests are in preview.md (see authoring/preview.md).

  • Derive preview arguments from the request, conversation, and report context. Ask only when required information is missing or ambiguous; never infer tenant-scoped workspace or semantic-model IDs.

  • Treat preview as inspection: service preview does not publish local work in progress. Never treat approval of edits, validation, screenshots, or preview as permission to publish. Never publish or overwrite the remote report with local changes until the user explicitly gives permission to do it. Publish only when the current request explicitly asks to publish, upload, push, or deploy the local changes. If that intent is absent, ambiguous, or you are unsure, keep the changes local and do not publish — do not ask to publish and do not make any remote write. Modifying a report with the service host means local edits plus validate plus preview only; never publish as part of that flow. Discard changes only when explicitly requested.

  • Treat host availability and capabilities as runtime facts. Respect HOST_UNAVAILABLE and CAPABILITY_UNSUPPORTED; do not bypass the preview command by calling an underlying host CLI directly.

  • Use CLI capability lookup before writing visual roles, formatting objects, enum values, selectors, or expression encodings.

PREFER
  • Start from an approved Design Brief: or _brief/report-spec.md for greenfield report builds.

  • Route visual-design uncertainty to the design mode before writing files.

  • For semantic model metadata or model-side changes, use a semantic-model authoring skill, Power BI Modeling MCP, or local TMDL files when available.

AVOID
  • Do not guess PBIR JSON from memory when CLI metadata or reference files are available.

  • Do not use only this skill for open-ended design, report planning, or Fabric report item CRUD; pair it with the design mode, the planning mode, or the management mode.

Quick Start Workflow

  1. Plan/design routing → for greenfield builds, read the planning mode first; for theming, visual style, layout, redesigns, or critiques, read the design mode. Return here for PBIR mechanics. Before authoring, use the Design Brief: yaml block from _brief/report-spec.md (or an approved inline Design Brief: block in the conversation) as implementation context.

  2. Set up/update CLI → before first use, confirm powerbi-report-author is available; see CLI Setup in authoring-part-02.md.

  3. Understand the model → use the Semantic Model MCP Server/skill if available, or read TMDL files directly for table/column/measure names

  4. Preview context → if Desktop and an operation are already selected, execute the Critical Notes protocol immediately. Otherwise derive the host and required arguments from available context; see Edit → Validate → Preview → Review in authoring-part-03.md.

  5. Route by intent → use Topic Files and Examples in authoring-part-02.md to pick the relevant guide.

  6. Use CLI metadata → use Authoring Metadata & Validation CLI in authoring-part-03.md for exact visual roles, formatting objects, property names, enum values, and selector requirements; do not infer these from memory.

  7. Check common pitfalls → read Anti-Patterns and Pitfalls in authoring-part-04.md before editing or validating when the change touches visuals, bindings, filters, formatting, layout, or rendered output.

  8. Validate → run powerbi-report-author validate <path-to-.Report-dir> after every logical batch of PBIR changes, then always continue to step 8; see Validation in authoring-part-03.md.

  9. Verify rendering → for every rendered-output change, load the latest PBIR, capture every affected page, and review the screenshot files using the powerbi-report-author preview surface with an explicit --host; see Edit → Validate → Preview → Review in authoring-part-03.md and Screenshot Review (see authoring/screenshot-review.md). Do not proceed until both validation and visual review pass. For dashboard/report builds, page scaffolding is not completion — each requested page needs data-bound visuals.

  10. Report back → give the user a concise summary of what was done and any issues encountered (major and minor).

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Security

What our scanner noticed· Low risk

  • Phrasing that asks for actions without confirmation
  • Instructions that use the network

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Published by
Microsoft
Publisher
Microsoft (Official)
Source
microsoft/skills-for-fabric · skills/powerbi-report-cli · commit f8fac11
Licence
MIT License (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE)
  • ATTRIBUTION.md (written by AutoGineer)
Synced
Also distributed in
microsoft/skills-for-fabric

AutoGineer is not affiliated with or endorsed by Microsoft.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download powerbi-report-cli.zip