菜单
Validate a Power BI semantic model with the real Best Practice Analyzer (BPA) via the free Tabular Editor 2 CLI, using a shareable BPA rules file.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | pbi-quality-rules |
| description | Validate a Power BI semantic model with the real Best Practice Analyzer (BPA) via the free Tabular Editor 2 CLI, using a shareable BPA rules file. |
Run the Best Practice Analyzer (BPA) — the industry-standard model linter — against a PBIP semantic model. This skill uses the real BPA engine from Tabular Editor 2 (free, open source) instead of a homegrown checker, so results match what Power BI developers expect and the rules are portable.
bpa-rules.json (next to this file) holds the team's BPA rules in Tabular Editor's native
format (ID, Name, Category, Severity, Scope, Expression in Dynamic LINQ over the
TOM). Severity: 3 = error, 2 = warning, 1 = info. Edit this file to add or tune rules;
it loads directly into Tabular Editor's BPA UI too.
Requires Tabular Editor 2 (free). Desktop should be closed so disk reflects the latest state.
# Windows (TabularEditor.exe on PATH). Point at the .SemanticModel folder (TMDL) or model.bim.
TabularEditor.exe "Sales.SemanticModel" -A ".agent/skills/pbi-quality-rules/bpa-rules.json" -V
-A <rulesfile> runs the BPA with the given rules; -V writes violations to the console
(and sets a non-zero exit code on error-severity hits — useful in CI).te CLI (preview) accepts the same rules file if installed instead.Parse the console output, then present errors and warnings only (omit info unless asked).
The kit stays usable without the tool. Parse the TMDL files directly and evaluate the same
rules by hand: for each rule, read its Scope and Expression from bpa-rules.json and apply
the equivalent check to the parsed objects (measures, tables, columns, relationships). This is
less robust than the real BPA — say so, and recommend installing Tabular Editor 2 for accuracy.
error, offer an immediate fix. For warning, suggest but
don't force. Group by category so the report is scannable.Use when the user wants to create, add, extend, or modify an agent, skill, or workflow in this kit (e.g. "create a skill for our dbt naming standard", "add an agent for X", "new workflow"). Decides whether to build an agent, a skill, or nothing, enforces lean quality standards, and keeps every installed AI-tool folder (.agent, .claude, .github, .opencode, AGENTS.md) identical in each tool's native format.
Use when the user wants to build a presentation, deck, slides, pitch, readout, or interactive report from data work — exploratory analysis, dashboards, business plans, proposals. Covers as-code HTML formats (reveal.js deck, flexible scrollable deck, interactive site, consulting-style PDF, Marp) plus a chart cookbook, how to choose between them, and ready-to-edit starters. Default engine is reveal.js. For native editable PowerPoint use the generating-pptx skill.
Use when the user needs an EDITABLE PowerPoint (.pptx) deliverable — "make a PowerPoint", "export to PPTX", "editable slides", "send me a .pptx". Generates native, fully editable PowerPoint (real text boxes, shapes, native charts with embedded data, tables) via python-pptx — not images. Brand colors map from DESIGN.md. For HTML decks/sites/PDF use building-html-presentations instead.
Use when the user wants output to follow a company's brand / visual identity, or asks to set one up — "use our brand", "make it on-brand", "here's our identity", "create a DESIGN.md". Reads and applies a DESIGN.md (Google Labs open spec): brand tokens for colors, typography, spacing, radius, shadows, and components, so decks, sites, and docs come out consistently on-brand. Scaffolds a DESIGN.md when none exists.
Use when the user asks to document, write docs for, or generate a runbook, data dictionary, metric definition, ADR, or Power BI dashboard catalog for a data project. Provides the team's canonical documentation templates as separate files.
Edit the Power BI report layer as code — pages, visuals, bookmarks, and themes — by writing PBIR JSON files in a PBIP project (offline, Desktop closed).