powerbi-review-semantic-model
Skill3 starsChanged 7 months agoMITView on GitHub
Review, audit, and validate Power BI semantic models against quality, performance, and best practice standards. Automatically invoke when the user asks to "review a semantic model", "audit a semantic model", "check model quality", "optimize my model", "validate model design", "check AI readiness", "prepare model for Copilot", or mentions model validation or quality assessment.
In this repository58 other files it gives its agents.
- powerbi-audit-tenant-settingsSkill
- powerbi-bpa-rulesSkill
- powerbi-connect-pbidSkill
- powerbi-connectSkill
- powerbi-create-pbi-reportSkill
- powerbi-c-sharp-scriptingSkill
- powerbi-custom-visualsSkill
- powerbi-cyber-securitySkill
- powerbi-data-catalog-lineageSkill
- powerbi-data-governance-traceabilitySkill
- powerbi-data-transformationSkill
- powerbi-dax-masterySkill
- powerbi-dax-performanceSkill
- powerbi-daxSkill
- powerbi-deneb-visualsSkill
- powerbi-doctorSkill
- powerbi-duckdb-lakehouseSkill
- powerbi-fabric-appsSkill
- powerbi-fabric-app-templatesSkill
- powerbi-fabric-capacitySkill
- powerbi-fabric-cliSkill
- powerbi-fabric-pipelinesSkill
- powerbi-fabricSkill
- powerbi-lineage-analysisSkill
- powerbi-measure-glossarySkill
- powerbi-medallion-architectureSkill
- powerbi-modelSkill
- powerbi-modifying-theme-jsonSkill
- powerbi-paginated-reportsSkill
- powerbi-pbi-cliSkill
- powerbi-pbip-formatSkill
- powerbi-pbir-cliSkill
- powerbi-pbir-format-enhancedSkill
- powerbi-performance-scaleSkill
- powerbi-power-querySkill
- powerbi-project-managementSkill
- powerbi-python-visualsSkill
- powerbi-refresh-semantic-modelSkill
- powerbi-report-authoringSkill
- powerbi-report-conversionSkill
- powerbi-report-designSkill
- powerbi-reportSkill
- powerbi-report-structureSkill
- powerbi-report-themingSkill
- powerbi-review-reportSkill
- powerbi-r-visualsSkill
- powerbi-security-rlsSkill
- powerbi-source-integrationSkill
- powerbi-spark-livySkill
- powerbi-standardize-naming-conventionsSkill
- powerbi-star-schema-modelingSkill
- powerbi-svg-visualsSkill
- powerbi-te2-cliSkill
- powerbi-te-cliSkill
- powerbi-te-docsSkill
- powerbi-testing-validationSkill
- powerbi-time-series-dataSkill
- powerbi-tmdlSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
Before analyzing TMDL, collect metadata and understand the business context.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Warning: This skill is incomplete and still in progress, but may provide value already as-is -- Kurt
Reviewing Semantic Models
Structured evaluation of Power BI semantic models against quality, performance, and best practice standards. Produces actionable findings with prioritized recommendations.
Review Workflow
Step 0: Gather Context
Before analyzing TMDL, collect metadata and understand the business context.
Run the model info script:
python3 scripts/get_model_info.py -w <workspace-id> -m <model-id>This returns: storage mode, model size, connected reports, deployment pipeline, endorsement status, sensitivity label, data sources, refresh schedule, last refresh, and capacity SKU.
Ask the user:
- What business process does this model represent?
- Who are the primary consumers? (report developers, analysts, executives, AI/Copilot users?)
- Are they the developer of both the model and its reports, or only one?
- Is the model in development, testing, or production?
- Where should findings be documented? (scratchpad, agent-docs, wiki, etc.)
Understanding the business context is critical. A model for 3 analysts has different requirements than one consumed by Copilot across the organization. The audit categories and their severity shift based on this context.
Step 1: Analyze Model Structure
Inspect the model definition to evaluate its structure. The approach depends on available tooling -- use whatever is available to read the model's tables, columns, measures, relationships, and expressions. Do not prescribe a specific tool; common options include Tabular Editor, the te-cli, fab export to TMDL, or programmatic access via APIs.
Step 2: Audit Categories
Evaluate findings across categories, ordered by severity:
Critical
- Bidirectional relationships (ambiguity risk)
- Circular dependencies between tables
- Missing data types on columns
- Tables without relationships (orphaned)
Memory and Size
- High-cardinality columns with large dictionaries (GUIDs, transaction IDs, composite keys)
- IsAvailableInMdx enabled on hidden or high-cardinality columns (wastes memory on attribute hierarchies unused by DAX; disable for columns not consumed via Analyze in Excel / MDX)
- Unsplit DateTime columns (near-unique precision creating massive dictionaries)
- Auto Date/Time tables (hidden LocalDateTable_* bloating memory)
- Inappropriate data types (Double for currency, String for numeric)
- Calculated columns that could be measures
- Unused columns or tables (no references in measures or visuals or other downstream items)
Data Reduction
- Unfiltered history in fact tables (no date-range filter or incremental refresh)
- Columns that aren't necessary for reporting or calculations or consumption
- Pre-summarization opportunities (detail grain not needed for reporting)
- Columns better handled upstream (i.e. calculations not done in calc columns or PQ)
DAX Anti-Patterns (for systematic DAX query optimization, use the dax skill)
- Filtering tables instead of columns in CALCULATE (causes both correctness and performance issues)
- Unhandled division by zero (use DIVIDE() or explicit zero-check; note: plain
/is fine when the denominator is guaranteed non-zero and can be faster) - Iterators with callbacks or nested iterators over large tables (use aggregators like SUM/AVERAGE when possible; iterators over large tables are fine if the expression is Storage Engine-pushable)
- Missing KEEPFILTERS around non-equality filter predicates in CALCULATE
Measure Hygiene
- Implicit measures used where explicit measures should exist
- Report-scoped extension measures that should be model-level
- Duplicate or overlapping measures with ambiguous names
Documentation
- Tables or columns missing descriptions
- Missing display folders for measures
- Inconsistent naming conventions (use the
standardize-naming-conventionsskill)
Design
- Star schema violations (direct fact-to-fact relationships, snowflake patterns)
- Missing or misconfigured date table: must be marked (
dataCategory: Timein TMDL, with a key Date column), have continuous daily dates (no gaps), span the full range of fact data, and relate to fact tables via a single-column relationship. Missing any of these causes time intelligence functions (DATEADD, SAMEPERIODLASTYEAR, TOTALYTD) to return BLANK - Excessive columns per table (>30 suggests denormalization issues)
- Many-to-many relationships without bridging tables
- Multiple fact tables relating to the same dimension via different keys without a shared conformed dimension (causes slicers on one fact to not filter the other)
- Inactive relationships without corresponding USERELATIONSHIP in measures (orphaned relationships that suggest incomplete modeling)
Direct Lake (if applicable)
- Delta table health (parquet file count, V-Order, row group sizes)
- DirectQuery fallback risk (RLS definitions, SQL endpoint views)
AI and Copilot Readiness (see references/ai-readiness.md)
- Duplicate field names across tables (confuses Copilot/data agents)
- Missing AI instructions
- Missing or inadequate descriptions for AI consumption
- Complex patterns (disconnected tables, many-to-many, inactive relationships) are valid model design but AI may struggle with them
Step 3: Performance Analysis
For performance-specific analysis, see references/performance.md.
Step 4: Report Findings
Produce a structured markdown report with:
- Summary table of finding counts by severity
- Detailed findings with file locations and line numbers where possible
- Specific remediation recommendations for each finding
- Prioritized action list (critical first)
Using the Semantic Model Reviewer Agent
Dispatch the semantic-model-auditor agent to perform the structural audit. The agent handles export, analysis, and reporting autonomously.
Notes
- The structural audit analyzes model metadata -- it does not execute DAX queries or check data quality
- For DAX query performance testing, see
references/performance.md - For DAX optimization, use the
daxskill - For companion report review, use the
review-reportskill in the reports plugin
References
references/ai-readiness.md-- Copilot/Data Agent preparation: AI instructions, descriptions, schema, verified answersreferences/performance.md-- Performance testing methodology, unused column detection, memory analysisscripts/get_model_info.py-- Quick model metadata overview (storage mode, size, reports, pipeline, endorsement, data sources)
Related Skills
dax-- DAX performance optimizationreview-report(reports plugin) -- Companion skill for report-level reviewstandardize-naming-conventions-- Naming audit and remediationlineage-analysis-- Downstream report discoveryrefresh-semantic-model-- Refresh monitoring and troubleshooting
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

