Power BI superpowers for AI agents

Your AI assistant ships Power BI work that holds up in production. 9 specialist skills, setup for 2 external Microsoft MCP servers, one CLI.

npm install -g bi-superpowers
super install --all --yes

Restart your agent once after install. That's the whole setup.

Works with Codex, Claude Code, GitHub Copilot, Gemini CLI, and Kilo Code — same operating contract across every agent.

Fully open source under MIT. The plugin itself needs no account or license key. Updates, package builds, Microsoft APIs, Power BI Service, and the external MCP servers can require network access or Microsoft credentials.

Status: v1.1.x is the current stable release line. Every public slash command lives under the /bi-* namespace.

Validation baseline: developed and stress-tested first in Codex. Other agents are shipped as compatibility targets unless a release note says otherwise.

Table of Contents

• Fundamentals
• Install From Terminal
• Ask Your Agent To Install It
• Skills Included
• Quick Demos
• Supported Agents
• MCP Servers & External Tools
• CLI Reference
• Plugin File Tree
• Requirements
• Updates
• What Ships To npm
• Development
• Project Validation Harness
• Troubleshooting
• FAQ

Fundamentals

If you're new to AI coding agents, four words you need to know:

• Agent — a conversational AI assistant that reads, writes, and modifies code in your project.
• Plugin — an installable package that gives an agent new capabilities: specialist skills, MCP connections, configuration.
• Skill — a focused expert inside a plugin. Each skill owns one thing (modeling, DAX, performance).
• MCP — Model Context Protocol. An open standard that lets an agent talk to external tools.

BI Agent Superpowers is a plugin: 9 skills, 2 MCPs, one CLI. The agent does the typing; the skills tell it which tools, files, prerequisites, and guardrails apply to Power BI work — so it ships work that holds up in production instead of inventing DAX, PBIR, or setup steps.

Install From Terminal

All Supported Agents

Use this when you want one command path for Codex and the compatibility adapters for Claude Code, GitHub Copilot, Gemini CLI, and Kilo Code.

npm install -g bi-superpowers
super install --all --yes

Restart your agent after installation so it reloads skills and MCP configuration.

Interactive Install

Use this when you want the installer to detect agents and ask where to install.

npm install -g bi-superpowers
super install

Install For One Agent Only

# Claude Code user-level skills + MCP config
super install --agent claude-code --yes

# GitHub Copilot
super install --agent github-copilot --yes

# Codex
super install --agent codex --yes

# Gemini CLI
super install --agent gemini-cli --yes

# Kilo Code
super install --agent kilo --yes

Claude Code Marketplace Install

If you use Claude Code, the marketplace path is the cleanest install experience for that agent. Claude Code is a compatibility target; Codex remains the primary end-to-end validation baseline.

/plugin marketplace add luquimbo/bi-superpowers
/plugin install bi-superpowers

After that, start a new session and run:

/bi-start

Advanced: Optional Local Claude Code Plugin

This is only for Claude Code users who deliberately want repo-local slash-command plugin files. Codex, GitHub Copilot, Gemini CLI, and Kilo Code should use super install above; do not run super kickoff for those agents.

Inside a Power BI project repo:

super kickoff

This generates project-local Claude Code plugin files:

• .claude-plugin/
• .plugin/
• .mcp.json
• commands/
• skills/
• .bi-superpowers.json (ownership and preferences)

Regenerate them later with:

super recharge

Ask Your Agent To Install It

If your agent has terminal access, paste one of these prompts.

Claude Code

Use the marketplace path first:

Install BI Agent Superpowers from the Claude Code marketplace.

Run these Claude Code plugin commands:

/plugin marketplace add luquimbo/bi-superpowers
/plugin install bi-superpowers

After installation, start the plugin with /bi-start and tell me if any restart or reload is needed.

If you also want project-local plugin files in the current repo for Claude Code only:

Set up BI Agent Superpowers for this project as a local Claude Code plugin.

In the terminal, run:

npm install -g bi-superpowers
super kickoff

Then verify that .claude-plugin/, .plugin/, .mcp.json, commands/, skills/, and .bi-superpowers.json were created in this repo. Do not modify unrelated files.

GitHub Copilot

Set up BI Agent Superpowers for GitHub Copilot on this machine.

In the terminal, run:

npm install -g bi-superpowers
super install --agent github-copilot --yes

After it finishes, verify that the GitHub Copilot skills directory and MCP config were updated. Then remind me to restart VS Code or reload Copilot so it picks up the new skills and MCP servers.

Codex

Set up BI Agent Superpowers for Codex on this machine.

In the terminal, run:

npm install -g bi-superpowers
super install --agent codex --yes

After it finishes, verify that the shared skills were installed under ~/.agents/skills and that the Codex MCP config was updated. Then summarize the installed skills and MCP servers.

Do not run super kickoff or super mcp-setup for Codex; those commands are only for optional repo-local Claude Code plugin files.

Gemini CLI

Set up BI Agent Superpowers for Gemini CLI on this machine.

In the terminal, run:

npm install -g bi-superpowers
super install --agent gemini-cli --yes

After it finishes, verify that ~/.gemini/skills and the Gemini MCP settings were updated. Then tell me to restart Gemini CLI before using the skills.

Kilo Code

Set up BI Agent Superpowers for Kilo Code on this machine.

In the terminal, run:

npm install -g bi-superpowers
super install --agent kilo --yes

After it finishes, verify that ~/.kilo/skills and ~/.config/kilo/kilo.jsonc were updated. Then tell me to restart Kilo Code before using the skills.

Multi-Agent Setup

Set up BI Agent Superpowers for every supported agent installed on this machine.

In the terminal, run:

npm install -g bi-superpowers
super install --all --yes

After it finishes, summarize which agents were configured, which MCP files were written, and which external skill roots or MCP entries were preserved.

Skills Included

Nine specialist skills. Each owns one thing — and each teaches as it works: it asks in plain business language, explains the why before each step, defines terms on first use, and shows a worked example, so a non-technical user learns while the work gets done.

/bi-start — Session Orchestrator

Opens a clean session, checks for plugin updates, shows the skill menu, and routes the conversation to the right specialist.

Use when starting a new session, when you don't know which skill to invoke, or when you want an update check before work begins.

/bi-kickoff — Project Bootstrap

From zero to a working BI project. Bootstraps a fresh project from the maintained base-template scaffold, guides a short business intake, replaces or adapts the packaged demo into a domain semantic model through MCP, and keeps the project context Codex-first. For an existing PBIP/PBIX project, use /bi-refactor instead.

Use when starting a brand-new Power BI project or bootstrapping a model from base-template. For an existing project, use /bi-refactor.

/bi-refactor — Refactor Specialist

Cleans up an existing project. Diagnoses an existing PBIP/PBIX, proposes template-assisted fixes, and applies semantic-model cleanup through MCP — preserving report topology, never rebuilding visuals, pages, or bookmarks.

Use when modernizing an existing model, migrating it toward base-template patterns, or repairing model issues without touching the report layout.

/bi-modeling — Model Architect

Star schemas, relationships, TMDL. Designs and audits semantic models against a health checklist (date table, fact/dim split, relationship cardinality, hidden technical columns).

Use when designing a new model, refactoring an existing one, or running a health audit before shipping.

/bi-dax — DAX Expert

Measures that compute the right thing. Authors, debugs, and optimizes DAX — measures, calculation groups, SE/FE timing, Sallieri patterns, calculation items.

Use when writing or fixing measures, untangling time-intelligence, or chasing a slow expression with DAX Studio timings.

/bi-performance — Performance Engineer

Diagnose slow models. Fix them. Profiles refreshes and reports with Performance Analyzer, DAX Studio / VertiPaq Analyzer metrics, Tabular Editor 2, and Best Practice Analyzer.

Use when reports are slow, when refresh blows up, or when you need a defensible "this is the bottleneck" answer before optimizing.

/bi-connect — MCP Connection Specialist

Wire your agent to Power BI Desktop. Connects through the Microsoft Power BI Modeling MCP so the agent can read tables, columns, measures, and relationships, run DAX, and write user-defined functions using current Microsoft Learn syntax and DAXLIB patterns.

Use when you need live model access, want to execute DAX queries, or are authoring DAX UDFs. Model writes go through MCP; PBIP files stay as saved/exported version-control snapshots.

/bi-powerquery — Power Query Engineer

Create clean Power Query queries for the current model. Builds strict M staging queries for real business data, validates columns and types, and only applies the Data = "Demo" / Data = "Real" append pattern when the current model actually uses it.

Use when wiring real Excel/SQL/SharePoint/API data to a BI model, generating *_Real staging queries, or keeping demo rows while the report switches between demo and real data. The current model is the destination contract; the packaged Andina Nexus/base-template contract is only an example/fallback for that shipped demo.

/bi-scorecard — Scorecards Specialist

Targets, goals, and check-ins in Power BI Service. Builds scorecard blueprints from the base-template metrics catalog, including the Andina Nexus Dirección executive lens and multi-team metric ownership, creates and updates scorecards and goals through REST scripts, and manages check-ins and status rules.

Use when creating Power BI Scorecards/Goals from model metrics, defining owners, cadence, targets, and status rules, or refreshing connected goal values.

The skill follows Microsoft Learn scorecard limits: no BYOK/RLS, max four subgoal levels, no publish-to-web, no SharePoint scorecard visual embed, no cross-tenant B2B sharing, no Multi-Geo, and no embedded analytics.

Quick Demos

Start A Session

/bi-start

Expected result:

• update check;
• menu of 9 skills;
• environment hints;
• next-step recommendation.

Refactor An Existing Power BI Project

/bi-refactor
Diagnose this existing PBIP/PBIX project. Tell me what exists, what is missing, and what to clean up.

Expected result:

• project map;
• risks and missing context;
• recommended cleanup path (model + measures);
• report topology preserved — no destructive rebuilds.

Connect To Power BI Desktop

/bi-connect
Connect to the model currently open in Power BI Desktop and list tables, measures, columns, and relationships.

Expected result:

• MCP connection check;
• semantic model snapshot;
• actionable diagnostics if Desktop or MCP wiring is not ready.

Write A DAX UDF With DAXLIB As Reference

/bi-connect
Write a reusable DAX UDF for gross margin percentage. Use Microsoft Learn for the current UDF syntax and consult DAXLIB patterns before proposing the function.

Expected result:

• current UDF syntax/preview check;
• existing function inspection when a model is connected;
• DAXLIB-aware pattern selection without vendoring external code by default;
• tested function draft and validation steps.

Supported Agents

| Agent | Skill install path | MCP config path | Primary usage | Validation status | |---|---|---|---|---| | Codex | ~/.agents/skills/ | ~/.codex/config.toml | Natural language with skills and MCPs | Primary end-to-end test baseline | | Claude Code | ~/.claude/skills/ or marketplace plugin | ~/.claude.json | Slash commands and natural language | Compatibility target; not the primary end-to-end QA baseline | | GitHub Copilot | ~/.copilot/skills/ | ~/.copilot/mcp-config.json | Natural language with skills and MCPs | Compatibility target; not the primary end-to-end QA baseline | | Gemini CLI | ~/.gemini/skills/ | ~/.gemini/settings.json | Natural language | Compatibility target; not the primary end-to-end QA baseline | | Kilo Code | ~/.kilo/skills/ | ~/.config/kilo/kilo.jsonc | Natural language | Compatibility target; not the primary end-to-end QA baseline |

The installer always writes the universal skill copy to ~/.agents/skills/, then links or copies agent-specific skill directories as needed.

MCP Servers & External Tools

Two official Microsoft MCPs ship pre-wired. One open-source CLI is invoked by /bi-performance.

| Tool | Owner | Transport | Purpose | |---|---|---|---| | powerbi-modeling-mcp | Microsoft (official) | local stdio | Connects to Power BI Desktop through XMLA for model inspection, DAX queries, and model operations. | | microsoft-learn | Microsoft (official) | HTTP | Gives the agent access to Microsoft Learn documentation in context. | | Tabular Editor 2 CLI | Open-source, MIT | local CLI | Best Practice Analyzer runs for diagnostic rule checks; /bi-performance does not perform semantic-model or report writes and searches existing installs before offering to install TE2. |

You can install the MCPs manually, but super install writes the expected MCP config for each selected agent automatically. For Codex installs, it also pins per-tool approval gates for sensitive Power BI Modeling MCP calls: semantic-model writes and trace_operations, because traces can include query text and write JSON files locally. Connection discovery and DAX query/validation remain ungated by default because they do not mutate the model and are needed for smoke tests; stricter teams can still add host-level approvals for those read/query tools. Other agents rely on their own MCP permission prompts or allowlists.

For DAX user-defined functions, /bi-connect also points agents at DAXLIB as an upstream reference:

• https://github.com/daxlib/daxlib
• https://docs.daxlib.org/

BI Agent Superpowers does not vendor DAXLIB source code. Agents consult it as a reference and only import/adapt specific MIT-licensed functions when the user explicitly requests that.

CLI Reference

The package exposes both super and bi-superpowers:

super help
bi-superpowers help

Common commands:

super install                # Install 9 skills + 2 MCPs for selected agents
super uninstall              # Remove the skills + 2 MCPs (mirror of install)
super uninstall --dry-run    # Preview what an uninstall would remove
super kickoff                # Advanced: generate optional local Claude Code plugin files
super recharge               # Regenerate optional local Claude Code plugin artifacts
super create-from-template   # Scaffold a new PBIP project from templates/<name>/ into pbip-files/
super template-path --template base-template  # Print the packaged template path
super powers                 # List available skills
super scan                   # Compare source skills with generated artifacts
super checkup                # Validate skill content
super validate-cases         # List validation playbooks or print one
super smoke-test             # Run the live base-template smoke test directly
super validate-projects      # Validate repo fixtures + local private project refs
super build-desktop          # Build the Claude Desktop .mcpb extension in the current dir
super build-desktop --output dist
super upgrade                # Update the global CLI and print refresh steps
super autoupdate on          # Opt-in: silent self-update on each Claude Code session
super autoupdate off         # Turn the auto-update hook back off
super autoupdate status      # Show whether auto-update is enabled
super help                   # Show full CLI help

Auto-update (opt-in, Claude Code)

Off by default. Run super autoupdate on (or super install --auto-update) to register a Claude Code SessionStart hook. From then on, each new Claude Code session checks at most once per hour and, if a newer version exists, silently upgrades via npm (npm install -g bi-superpowers@latest) and refreshes your agents (super install --all --yes) in a detached background process — so you never install by hand. It is npm-based, so it works even if the GitHub repo is private. Turn it off anytime with super autoupdate off. Other agents (Codex, Copilot, Gemini, Kilo) keep the notify-on-invocation flow plus manual super upgrade.

super create-from-template

super create-from-template --list
super create-from-template --template base-template --name <projectName>
super create-from-template --template base-template --name <projectName> --target /path/to/project
super create-from-template --template base-template --name <projectName> --dry-run
super create-from-template --template base-template --name <projectName> --force
super create-from-template --template base-template --name <projectName> --force --replace-project <oldName>

Used by the bi-kickoff skill PHASE 0 Option A (start from template) so agents do not have to copy files by hand. The command copies templates/<name>/ into <target>/pbip-files/<projectName>.*, renames the PBIP descriptor and the .SemanticModel/ / .Report/ folders, rewrites internal cross-references (.pbip → Report.path, definition.pbir → datasetReference.byPath.path), and regenerates the Fabric logicalId in both .platform files. It refuses to overwrite an existing PBIP unless --force is passed, skips the template's own AGENTS.md, skips local Desktop/editor state such as .pbi, localSettings.json, cache.abf, and DAXQueries, and when forced cleans only the destination project paths before copying again. If there is exactly one PBIP in the destination, --force replaces that project even when the new --name differs; if there are multiple PBIPs, pass --replace-project <oldName> so the command cannot guess the wrong project.

The packaged template is base-template, the canonical dynamic-first reference for new Power BI projects. Dynamic metric format strings stay safe and unscaled for numeric visuals; compact abbreviations live in dedicated text helper measures for KPI labels.

super template-path --template base-template prints the installed template folder path. Skills use it instead of hard-coded developer machine paths.

super mcp-setup is intentionally not part of normal setup. It only refreshes MCP config inside an existing local Claude Code plugin created by super kickoff; it refuses to write repo-local config in arbitrary projects. For Codex, GitHub Copilot, Gemini CLI, Kilo Code, and normal Claude Code user-level installs, use super install --all --yes.

Plugin File Tree

Local Claude Code Plugin

Generated by super kickoff or super recharge:

This tree is optional and Claude Code specific. It is not required for Codex, GitHub Copilot, Gemini CLI, or Kilo Code because those agents use the user-level install created by super install.

project/
├─ .claude-plugin/
│  ├─ plugin.json
│  ├─ marketplace.json
│  └─ skill-manifest.json
├─ .plugin/
│  └─ plugin.json
├─ .mcp.json
├─ commands/
│  ├─ bi-start.md
│  ├─ bi-kickoff.md
│  ├─ bi-connect.md
│  ├─ bi-modeling.md
│  ├─ bi-dax.md
│  ├─ bi-performance.md
│  ├─ bi-powerquery.md
│  └─ bi-scorecard.md
├─ skills/
│  ├─ bi-start/
│  │  ├─ SKILL.md
│  │  └─ scripts/update-check.js
│  ├─ bi-kickoff/
│  │  ├─ SKILL.md
│  │  └─ scripts/update-check.js
│  ├─ bi-connect/
│  │  ├─ SKILL.md
│  │  └─ scripts/update-check.js
│  ├─ bi-powerquery/
│  │  ├─ SKILL.md
│  │  ├─ references/
│  │  └─ scripts/
│  │     ├─ update-check.js
│  │     ├─ new-powerquery-staging.ps1
│  │     └─ test-powerquery-contract.ps1
│  └─ bi-scorecard/
│     ├─ SKILL.md
│     └─ scripts/
│        ├─ update-check.js
│        ├─ powerbi-scorecards-api.ps1
│        ├─ powerbi-goals-api.ps1
│        ├─ powerbi-goal-values-api.ps1
│        ├─ powerbi-goal-status-rules-api.ps1
│        └─ new-scorecard-blueprint-from-base-template.ps1
└─ .bi-superpowers.json

Build Pipeline

flowchart TD
    A["src/content/"] --> B["npm run build:plugin"]
    B --> C["commands/*.md"]
    B --> D["skills/*/SKILL.md"]
    B --> E[".claude-plugin/"]
    B --> F[".plugin/"]
    B --> G[".mcp.json"]
    C --> H["Claude Code slash commands"]
    D --> I["Agent skills"]
    G --> J["Power BI Modeling MCP + Microsoft Learn MCP"]

Requirements

Base Requirements

• Node.js 18 or newer.
• At least one supported AI agent.

For bi-connect

• Windows.
• Power BI Desktop open.
• Power BI Modeling MCP available to the agent.

Updates

Claude Code Marketplace

/plugin update bi-superpowers

npm / CLI Installs

super upgrade
super install --all --yes

If you created a project-local plugin:

cd your-power-bi-project
super recharge

super upgrade intentionally does not run super install or super recharge automatically. User-profile installs and project-local plugins are separate refresh targets.

Each selected agent receives an independent managed copy. Re-running install mirrors only BI Superpowers-owned skill directories and removes stale bundled files without touching external skills.

What Ships To npm

The npm package includes:

bin/
.claude-plugin/
.plugin/
.mcp.json
commands/
skills/
src/content/
templates/
desktop-extension/
README.md
LICENSE
package.json

The npm package does not include:

validation/
validation.local.example.json
validation.local.json
node_modules/
*.test.js
**/.pbi/localSettings.json

templates/ ships the canonical templates/base-template/ PBIP package with AGENTS.md, semantic model, and report definition. It is the "template-as-reference" used by template-aware specialist skills and the source for the bi-kickoff "create from template" branch invoked via super create-from-template, which copies templates/base-template/ into <user-repo>/pbip-files/<projectName>.* with renaming, cross-reference rewriting, and regenerated logicalIds. The base template includes Power Query pre-converted currency base columns, packaged sample queries grouped under Sample Dataset\Dimensiones and Sample Dataset\Hechos, _GC Métrica / _GC Tipo cálculo calculation groups, guarded metric dispatch, metric-aware dynamic dimensions through _PC Dimensión, a Equipos métricas[Equipo] slicer for filtering the metric selector by team, multi-team metric memberships through _Aux Métrica-Equipo, nine Equipos (eight operating teams plus Dirección) for the Andina Nexus product-and-services demo company, team-specific model-view diagrams, safe unscaled numeric model formats, compact KPI label helpers, cleaned key metadata, and a Data = "Demo" contract on appendable business tables. Demo facts now share full-calendar weekly drivers from the FechaInicio parameter through the current model date; sales, project execution, finance, marketing, operations, payroll, returns, tasks, and customer acquisition are derived from shared business drivers instead of independent random grids. Demo IDs are entity-coded text values such as ClienteA25D and shared helper values such as EQUA00A, so relationships stay on the original ID columns while real rows can append with Data = "Real" and their own source IDs; visible customer and country labels are capped at 13 values each so dynamic-dimension slicers stay readable without shrinking the customer cohort. /bi-powerquery is current-model first: it creates user-side staging queries for the active model and uses the packaged Demo/Real contract only when the active model is that demo or a generated project with the same pattern. The template itself does not ship unloaded *_Real queries. /bi-scorecard uses base-template as a metrics-catalog reference for Service scorecards; it does not add a universal targets model to the template. validation/ is repo-only. It stores sanitized validation descriptors and playbooks for project QA; private project paths live in validation.local.json, which must never be committed.

Development

Install dependencies:

npm install

Regenerate plugin artifacts:

npm run build:plugin

Validate locally:

npm test
npm run lint
npm run format:check
npm pack --dry-run --ignore-scripts

Source of truth:

src/content/

Generated artifacts:

commands/
skills/
.claude-plugin/
.plugin/
.mcp.json

Edit source files first, then regenerate generated outputs.

Project Validation Harness

In a source checkout, the repo can validate BI Agent Superpowers against real Power BI work without committing customer repositories.

Public fixtures live in:

templates/base-template/

Sanitized validation metadata lives in:

validation/projects/
validation/cases/

Private local project paths live only in:

validation.local.json

That file is gitignored. Copy validation.local.example.json when you want to map client projects on your own machine. Descriptors ending in .example.json are templates for humans and are ignored by validation runs, including --strict.

Run the structural validation:

super validate-projects --project base-template

Run every descriptor, treating skipped private projects as failures:

super validate-projects --strict

For a live Power BI Desktop smoke test of the canonical template, use validation/cases/base-template-smoke-test.md after the structural validation passes. super smoke-test will probe the open Desktop sessions automatically and keep the one that matches the expected evidence, including seven metrics per team, full FechaInicio → current-day calendar coverage, zero business invariant failures, and complete metric plus compatible metric-dimension coverage.

To discover the repo playbooks or print one in full:

super validate-cases
super validate-cases --case base-template-smoke-test

This command is non-destructive. It checks project paths and required files; it does not open Power BI Desktop, edit PBIP files, or copy private repos.

Publishing

Publishing is handled by GitHub Actions. Do not run npm publish locally.

Expected release flow:

npm version patch --no-git-tag-version
npm run build:plugin
npm test
npm run lint
npm run format:check
npm audit --omit=dev --audit-level=moderate
node bin/cli.js validate-projects --project base-template
node bin/cli.js smoke-test --json
node bin/cli.js scan --verbose
npm pack --dry-run --ignore-scripts
git add package.json package-lock.json commands/ skills/ .claude-plugin/ .plugin/ .mcp.json
git commit -m "chore: release vX.Y.Z"
git tag -a vX.Y.Z -m "vX.Y.Z"
git push origin main --follow-tags

The publish workflow runs on v* tag pushes. Manual dispatch is constrained to an already-tagged release commit on main. CI refuses mismatched tag/package versions and publishes from GitHub Actions without npm provenance because this repository is private. Do not create retrospective tags for already-published versions.

Troubleshooting

super is not found after npm install

Check the global package:

npm list -g bi-superpowers

Try the long binary name:

bi-superpowers help

If npm's global bin directory is not in PATH, fix PATH or use your package manager's global binary setup instructions.

The Agent Does Not See The Skills

1. Restart the agent.
2. Run super install --all --yes again.
3. Check the relevant skill directory.
4. If using a local Claude Code plugin, run super recharge inside that repo.

Power BI Desktop Does Not Connect

1. Open Power BI Desktop.
2. Open the target .pbix or PBIP project.
3. Confirm the Power BI Modeling MCP is installed.
4. Restart the agent after MCP installation.

FAQ

Does This Replace Power BI Desktop?

No. Power BI Desktop remains the visual and report authoring tool. The plugin helps agents inspect project files and use MCPs for semantic-model work, but PBIP files are treated as version-control snapshots. Agents must not patch .Report/** PBIR JSON, visual.json, bookmarks, slicer/card settings, themes, or report resources directly.

Existing-visual dimension/measure binding changes are the only report-side exception: they may happen in Power BI Desktop manually, or through a future dedicated rebind command that proves a binding-only diff with dry-run, backup, validation, and an explicit source-to-target mapping. No safe report rebind command currently ships with this plugin, so agents must not patch PBIR JSON by hand, even for a "simple measure replacement."

Microsoft Learn's PBIP project guidance also marks report.json, mobileState.json, semanticModelDiagramLayout.json, and semantic-model diagramLayout.json as unsupported external-edit surfaces during preview. The plugin treats those files as diagnostic-only.

Report topology is locked. Agents must never delete, rename, move, or recreate report pages, visuals, mobile layouts, or bookmarks to make a broken report look "fixed." They must not rebuild a "minimal functional report." Any added, deleted, or renamed path under .Report/** is a hard stop for semantic-model skills; the correct action is to preserve the report shell and hand the visual repair back to Power BI Desktop.

Does It Work Outside Windows?

The CLI and several skills can run outside Windows. Workflows that depend on Power BI Desktop, especially bi-connect, are Windows-first.

Do I Need Claude Code?

No. Codex is the primary development and end-to-end validation baseline for this plugin. Claude Code has a useful compatibility path through marketplace/local plugin artifacts, but it is not the canonical test base unless a release note explicitly says that release was tested there.

Can I Use Only The MCP Servers?

Yes. You can install Microsoft Learn MCP and Power BI Modeling MCP manually. This repo packages them with skills and operational workflows so agents know when and how to use them.

Does This Send Data To A SaaS Backend?

No. The package installs local files, writes local agent config, and uses MCP servers configured on your machine. There is no BI Agent Superpowers SaaS backend.

Contributing

Issues and pull requests are welcome:

• GitHub: https://github.com/leverio/acadevor/tree/main/plugins/bi-superpowers
• Issues: https://github.com/leverio/acadevor/issues
• npm: https://www.npmjs.com/package/bi-superpowers

Before large changes, read:

• AGENTS.md
• ARCHITECTURE.md
• src/content/
• bin/

License

MIT © Lucas Sanchez (@luquimbo)

Open-source Power BI Desktop toolkit for AI coding agents. Built by Lucas Sanchez.

bi-superpowers.dev · npm · GitHub