[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M69BNSNSETNDE2R5AEQAGZVC":3,"skill-same:01M69BNSNSETNDE2R5AEQAGZVC":79},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":26,"publishedAt":44,"linkOnly":45,"markdownContent":46,"markdownBody":47,"frontMatter":48,"refName":50,"commitSha":51,"skillDirectory":52,"forksCount":53,"repositoryHtmlUrl":54,"repositoryHomepageUrl":6,"ownerHtmlUrl":55,"ownerBlogUrl":56,"ownerDescription":57,"updatedAt":58,"install":59,"resources":72},"01M69BNSNSETNDE2R5AEQAGZVC","fabric-metadata-creation",null,"Create reviewable metadata proposals for Microsoft Fabric lakehouse tables and semantic models. Covers schema analysis, concise technical and business descriptions, glossary terms, classifications, Purview-like sensitivity label proposals, PII and sensitive data detection, relationship and lineage hints, data quality rules, CDE candidates, data products, JSON\u002FPDF output, and steward review questions. Use when: metadata creation, data catalog, Purview mapping, glossary inference, data classification, sensitivity labels, stewardship review, governance metadata.","microsoft\u002Fgh-copilot-fabric-agents","microsoft","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F6154722?v=4","01M698VYDXZ6JW5WH3CRV650R8","Microsoft","01M69BNKRQPFJ5Z6ECX2S8E17F",3,"MIT","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fgh-copilot-fabric-agents\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Ffabric-metadata-creation\u002FSKILL.md",[18,23],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000002","文档处理","SCENARIO","使用场景",{"categoryNo":24,"categoryName":25,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发",[27,28,29,30,31,32,33,34,35,36,37,38,9,39,40,41,42,43],"fabric metadata creation","fabric","governance","lakehouse","code","create","data catalog","azure","purview","PDF","microsoft fabric","identity","mcp server","inference","python","review","MCP","2026-11-10 10:12:30",false,"---\nname: fabric-metadata-creation\ndescription: \"Create reviewable metadata proposals for Microsoft Fabric lakehouse tables and semantic models. Covers schema analysis, concise technical and business descriptions, glossary terms, classifications, Purview-like sensitivity label proposals, PII and sensitive data detection, relationship and lineage hints, data quality rules, CDE candidates, data products, JSON\u002FPDF output, and steward review questions. Use when: metadata creation, data catalog, Purview mapping, glossary inference, data classification, sensitivity labels, stewardship review, governance metadata.\"\ncompatibility: \"Requires the Fabric MCP Server VS Code extension for interactive discovery when available. Uses Python helper scripts with azure-identity and requests for Fabric discovery\u002Fexport and markdown plus weasyprint for styled local PDFs. Produces local proposal artifacts only.\"\n---\n\n# Fabric Metadata Creation - Reference\n\nGenerates non-destructive metadata proposals for Microsoft Fabric lakehouse tables, files, views, and semantic models. The result is a steward-review package that can later be mapped into a data catalog such as Microsoft Purview, but it must not apply catalog changes automatically.\n\nThe agent's core principle is traceability: every suggestion must distinguish metadata explicitly found in the source from metadata inferred by the agent, include a confidence level, and carry assumptions or open questions when evidence is incomplete.\n\n## Global Rules\n\n- **Proposal only.** Never modify Fabric assets, Purview assets, glossary terms, labels, classifications, owners, stewards, retention policies, or access controls.\n- **Human review required.** Governance outcomes are suggestions, not decisions.\n- **No invented accountability.** Do not invent real owners, stewards, contacts, responsible teams, legal requirements, or compliance decisions. Use `Unknown`, `TBD`, or an open question when not explicit.\n- **Evidence labeling.** Every metadata object must include `metadata_source` with one of: `explicit`, `inferred`, or `mixed`.\n- **Confidence required.** Every proposal must include `confidence`: `high`, `medium`, or `low`.\n- **Assumptions required.** Every inferred proposal should include `assumptions`; use an empty list only when there are no assumptions.\n- **Open questions required.** Every asset-level and column-level proposal must include review questions when evidence is incomplete.\n- **Conservative language.** Use `candidate`, `suggested`, `likely`, `possible`, or `requires human review` for inferred classifications and sensitivity labels.\n- **Structured output.** Generate only the four final artifacts documented below: proposal JSON\u002FPDF and standalone glossary Markdown\u002FPDF.\n- **Spanish review documents.** Human-readable artifacts must be written in Spanish. Keep original workspace, lakehouse, table, column, measure, semantic model, and glossary candidate names exactly as they appear in the source; do not translate source identifiers.\n\n## Helper Script\n\nUse [.\u002Fscripts\u002Ffabric_metadata.py](.\u002Fscripts\u002Ffabric_metadata.py) for Fabric discovery\u002Fexport, PDF rendering, and output validation. The agent must activate `.venv\u002F` before running it.\n\nInstall dependencies:\n\n```bash\ntest -d .venv || python3 -m venv .venv\nsource .venv\u002Fbin\u002Factivate\npip install azure-identity requests markdown weasyprint\n```\n\nWeasyPrint requires system-level Cairo\u002FPango libraries. They are pre-installed on many Linux environments; on macOS install Pango with `brew install pango`.\n\nCommands:\n\n```bash\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py list-workspaces\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py list-items \u003Cworkspace_id> [\u003Citem_type>]\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py list-lakehouse-tables \u003Cworkspace_id> \u003Clakehouse_id>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py get-lakehouse-table \u003Cworkspace_id> \u003Clakehouse_id> \u003Ctable_name>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py list-semantic-models \u003Cworkspace_id> [\u003Cmodel_name>]\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py export-semantic-model \u003Cworkspace_id> \u003Csemantic_model_id> \u003Coutput_dir>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py render-pdf \u003Cmetadata_proposal.json> \u003Cmetadata_proposal.pdf>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py render-glossary-md \u003Cmetadata_proposal.json> \u003Cglossary_terms.md>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py render-glossary-pdf \u003Cmetadata_proposal.json> \u003Cglossary_terms.pdf>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py render-markdown-pdf \u003Cinput.md> \u003Coutput.pdf>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py validate \u003Coutput_dir>\n```\n\nUse MCP tools first when they are available for interactive discovery. Use the script when MCP does not expose the needed metadata, when exporting semantic model definitions, or when rendering\u002Fvalidating local outputs.\n\n## Output Directory\n\nSave each run to:\n\n```text\n.\u002Fmetadata_proposals\u002F{SOURCE_NAME}\u002F{YYYY-MM-DD_HHmmss}\u002F\n```\n\nRequired files:\n\n| File | Purpose |\n|------|---------|\n| `metadata_proposal.json` | Compact machine-readable metadata proposal |\n| `metadata_proposal.pdf` | Styled A4 proposal PDF rendered from Markdown with WeasyPrint |\n| `glossary_terms.md` | Standalone human-readable glossary proposal |\n| `glossary_terms.pdf` | Styled A4 glossary PDF rendered from Markdown with WeasyPrint |\n\nDo not generate `metadata_proposal.md`, `metadata_proposal.yaml`, `evidence.json`, or other final artifacts. Temporary discovery\u002Fexport files may be used while working, but they must not remain in the final output directory.\n\n## Default Proposal PDF Sections\n\nThe proposal PDF must present these simplified review sections in order. The JSON is the source of truth; the PDF is the human-readable rendering.\n\n1. Resumen ejecutivo\n2. Entrada interpretada\n3. Activos y columnas\n4. Glosario\n5. Sugerencias de gobierno\n6. Relaciones, linaje y calidad de datos\n7. Supuestos y preguntas abiertas\n\nBoth PDFs must include an index\u002Ftable of contents. Sections and subsections must be numbered. Asset sections must be visually isolated with page breaks, and the asset name must appear in the asset table header. Source names that contain Markdown-looking characters, such as `# Case Updates`, must be rendered as literal names rather than Markdown syntax.\n\n## Metadata Categories\n\n### Asset-Level Metadata\n\nUse these compact fields for databases, schemas, tables, views, files, semantic models, measures, dimensions, and relationship assets when applicable:\n\n```text\nasset_name, asset_type, fully_qualified_name, platform, source_system,\ndatabase_name, schema_name, object_type, data_domain, data_product,\nenvironment, lifecycle_status, business_description, technical_description,\ndata_owner, data_steward, glossary_terms, upstream_lineage,\ndownstream_lineage, source_of_truth, security_notes, quality_notes,\nconfidence\n```\n\nAdditional required traceability fields: `metadata_source`, `assumptions`, `open_questions`.\n\n### Column-Level Metadata\n\nUse these compact fields for columns, semantic model columns, calculated columns, measures, and dimensional attributes when applicable:\n\n```text\ncolumn_name, business_name, data_type, nullable, key_role,\nmeasure_or_dimension, business_description, technical_description,\nsemantic_type, classification, sensitivity_label, pii_indicator,\ncritical_data_element_candidate, derivation_logic, allowed_values,\nexample_values, quality_rules, relationship_hints, lineage_hints,\nmasking_recommendation, confidence\n```\n\nAdditional required traceability fields: `metadata_source`, `assumptions`, `open_questions`.\n\n### Business Glossary Term Metadata\n\nUse these compact fields for suggested business glossary terms:\n\n```text\nterm_name, definition, synonyms, related_terms, domain, status,\nowner_or_steward, associated_assets, associated_columns, confidence,\nmetadata_source, open_questions\n```\n\nUse `status: Proposed` unless an explicit existing glossary status is present.\n\nThe glossary must also be published as an independent human-readable output in `glossary_terms.md` and `glossary_terms.pdf`. This glossary is a steward-review artifact, not a Purview import file. It should be readable without opening the full metadata proposal.\n\nRecommended glossary presentation:\n\n1. Title and generation context\n2. Summary by domain or subject area\n3. Alphabetical term list grouped by domain or subject area\n4. For each term: definition, synonyms, related terms, associated assets, associated columns, confidence, metadata source, and open questions\n5. Review guidance explaining that terms are proposed and require steward approval\n\n### Classification Metadata\n\nUse these fields for classification proposals:\n\n```text\nclassification_name, classification_reason, detection_basis, confidence\n```\n\nAllowed classification names include, when evidence supports them: `personal data`, `financial data`, `operational data`, `identifier`, `reference data`, `master data`, `transactional data`, `metric`, `dimension`, `audit field`, `system field`, `derived field`.\n\n### Sensitivity Metadata\n\nUse these fields for Purview-like sensitivity label proposals:\n\n```text\nsuggested_label, label_reason, protection_recommendation, confidence\n```\n\nAllowed suggested labels: `Public`, `Internal`, `Confidential`, `Highly Confidential`, `Restricted`.\n\n### Data Quality Metadata\n\nUse these fields for quality rules and profiling recommendations:\n\n```text\nquality_dimension, quality_rule, severity, confidence\n```\n\nQuality dimensions include: `completeness`, `uniqueness`, `validity`, `accuracy`, `consistency`, `timeliness`, `freshness`, `integrity`, `conformity`, `reasonableness`.\n\n### Lineage and Relationship Metadata\n\nUse these fields for relationship and lineage hints:\n\n```text\nupstream_source, downstream_consumer, relationship_or_join_hint, confidence\n```\n\n### Critical Data Element Metadata\n\nUse these fields for CDE candidates:\n\n```text\ncde_name, business_reason, associated_columns, confidence\n```\n\n### Data Product Metadata\n\nUse these fields for suggested data products or domain groupings:\n\n```text\ndata_product_name, purpose, included_assets, target_consumers,\nowner_or_steward, confidence\n```\n\n## Evidence and Confidence Rules\n\n### Metadata Source\n\n| Value | Meaning |\n|-------|---------|\n| `explicit` | Directly present in Fabric metadata, TMDL, schema, annotations, descriptions, names, or user-provided context |\n| `inferred` | Deduced from names, data types, relationships, sample values, model structure, or common enterprise patterns |\n| `mixed` | Combines explicit evidence with inferred enrichment |\n\n### Confidence Levels\n\n| Confidence | Use When |\n|------------|----------|\n| `high` | Strong explicit evidence exists, or multiple independent signals agree, such as name pattern plus data type plus sample pattern |\n| `medium` | Reasonable inference from names, data types, or model context, but no validating sample or explicit business description |\n| `low` | Weak evidence, ambiguous abbreviations, missing samples, unclear domain, or conflicting signals |\n\nNever raise confidence only because a term sounds plausible. Lower confidence when samples, row counts, relationships, or business context are missing.\n\n## Inference Rules\n\n### Asset Type and Domain\n\nUse table\u002Fmodel names, workspace names, lakehouse names, schemas, and measure names to infer candidate domains. Examples:\n\n| Pattern | Suggested Domain or Context |\n|---------|-----------------------------|\n| `customer`, `client`, `account`, `contact` | Customer, sales, CRM, master data |\n| `order`, `invoice`, `payment`, `booking`, `transaction` | Sales, finance, transactional process |\n| `product`, `item`, `sku`, `catalog` | Product, inventory, reference\u002Fmaster data |\n| `employee`, `staff`, `payroll`, `hr` | Workforce or HR; possible sensitive data |\n| `patient`, `diagnosis`, `doctor`, `visit`, `hospital` | Healthcare; possible regulated data |\n| `flight`, `passenger`, `airport`, `aircraft` | Airline operations |\n| `incident`, `ticket`, `case`, `status`, `priority` | Operations or service management |\n| `calendar`, `date`, `time` | Date\u002Ftime dimension |\n| `audit`, `log`, `event`, `telemetry` | Operational, audit, or observability data |\n\nIf domain is ambiguous, use `Unknown` or a broad proposed domain and add an open question.\n\n### Descriptions\n\nGenerate two descriptions for each asset and column:\n\n- **Technical description**: what the object stores structurally, based on schema and type evidence.\n- **Business description**: what the object likely means to a business user, based on names, model context, relationships, and measures.\n\nDescriptions must be concise, deterministic, and free of unsupported claims. For low-confidence descriptions, explicitly say `Candidate description` or `Likely represents`.\n\n### Keys and Relationships\n\nInfer key metadata with these signals:\n\n| Signal | Suggested Metadata |\n|--------|--------------------|\n| Column named `id`, `{table}_id`, `{entity}_id`, `{entity}_key`, `uuid` | Candidate primary, foreign, natural, or surrogate key depending on context |\n| Single non-null unique ID-like column in a dimension table | Candidate primary key |\n| Fact table column matching a dimension PK name\u002Ftype | Candidate foreign key |\n| Columns named `code`, `number`, `reference`, `dni`, `nif`, `passport`, `iban` | Candidate natural key or business identifier |\n| Integer sequential ID column | Candidate surrogate key |\n| Composite repeated columns such as `order_id` + `line_number` | Candidate composite key |\n\nDefault cardinality to `many-to-one candidate` only when a transactional\u002Ffact-like table points to a reference\u002Fdimension-like table. Use `unknown` when there is insufficient evidence.\n\n### Measure and Dimension Detection\n\n| Pattern | Suggestion |\n|---------|------------|\n| Numeric additive columns: `amount`, `total`, `quantity`, `cost`, `revenue`, `sales`, `importe`, `cantidad` | Metric\u002Fmeasure candidate |\n| Descriptive strings: `name`, `description`, `category`, `status`, `type`, `city`, `country` | Dimension attribute candidate |\n| DAX measure expression exists | Explicit measure |\n| Calculated expression exists | Derived field candidate, include derivation logic |\n\n### Classification Detection\n\n| Pattern | Candidate Classification |\n|---------|--------------------------|\n| `id`, `_id`, `_key`, `uuid`, `guid`, `code`, `number` | Identifier |\n| `email`, `phone`, `mobile`, `telefono`, `address`, `dni`, `nif`, `nie`, `passport`, `name`, `birth_date` | Personal data candidate |\n| `iban`, `bank`, `card`, `salary`, `payroll`, `payment`, `invoice`, `amount`, `revenue`, `margin`, `cost`, `price` | Financial data candidate |\n| `patient`, `diagnosis`, `medical`, `doctor`, `visit`, `treatment` | Regulated or sensitive data candidate |\n| `status`, `category`, `type`, `priority`, `country`, `city`, `currency` | Reference data candidate |\n| `created_at`, `updated_at`, `loaded_at`, `ingestion_time`, `batch_id`, `source_file`, `run_id` | Audit or system field |\n| `score`, `ratio`, `total`, `average`, `margin`, `age`, `duration`, `days_since` | Derived field or metric candidate |\n\nDo not use regex-only matches as definitive evidence. Describe uncertainty in `classification_reason` for sensitive classifications.\n\n### Sensitivity Label Suggestions\n\n| Suggested Label | Use When |\n|-----------------|----------|\n| `Public` | Asset appears intended for open\u002Fpublic use and no sensitive indicators are present. Rarely infer this without explicit context. |\n| `Internal` | Default for ordinary enterprise metadata without sensitive indicators. |\n| `Confidential` | Contains candidate business confidential data, identifiers, contact data, financial values, or internal operational data. |\n| `Highly Confidential` | Contains likely PII, financial identifiers, health-related fields, security-relevant fields, or combinations of sensitive attributes. |\n| `Restricted` | Contains secrets, credentials, tokens, private keys, strong regulated data indicators, or explicitly restricted labels in source metadata. |\n\nTreat every inferred label above `Internal` as requiring human review.\n\n### Privacy and Sensitive Data Signals\n\nPotential PII and sensitive fields include:\n\n```text\nname, first_name, last_name, full_name, email, phone, mobile, address, city,\npostal_code, latitude, longitude, dni, nif, nie, ssn, passport, national_id,\ntax_id, birth_date, gender, patient_id, diagnosis, medical_record, iban,\nbank_account, credit_card, salary, password, secret, token, api_key, private_key,\nip_address, device_id, user_agent\n```\n\nMark these as candidates unless sample values or explicit metadata confirm the pattern.\n\n### Data Quality Rules\n\nSuggest quality rules from type and classification evidence:\n\n| Evidence | Suggested Rule |\n|----------|----------------|\n| Candidate primary key | Completeness and uniqueness checks, expected threshold 100% |\n| Candidate foreign key | Referential integrity check against referenced asset |\n| Required business column | Completeness check with agreed null threshold |\n| Email\u002Fphone\u002Fidentifier pattern | Validity and conformity checks |\n| Date\u002Ftimestamp column | Freshness, timeliness, and valid range checks |\n| Amount\u002Fquantity\u002Fmetric | Reasonableness, non-negative, range, outlier, and aggregation checks |\n| Status\u002Fcategory\u002Fcode | Allowed values and reference set checks |\n| Audit field | Presence, monotonicity, ingestion recency, and run consistency checks |\n\nSeverity guidance:\n\n| Severity | Use When |\n|----------|----------|\n| `high` | Key integrity, sensitive fields, CDE candidates, business-critical metrics |\n| `medium` | Important dimensions, common reporting filters, relationship fields |\n| `low` | Descriptive fields or optional enrichment attributes |\n\n### Critical Data Element Candidates\n\nSuggest a CDE candidate when one or more conditions apply:\n\n- Column is a key identifier used in relationships.\n- Column is a financial, operational, compliance, or customer-impacting metric.\n- Column appears in a semantic model measure or relationship.\n- Column is likely used for regulatory, privacy, security, or executive reporting.\n- Column is needed to join important assets or identify master\u002Freference entities.\n\nAlways describe CDEs as candidates and avoid definitive CDE status.\n\n### Glossary Term Generation\n\nCreate glossary terms by converting physical names to business names and grouping repeated concepts.\n\nRules:\n\n- Convert snake_case and camelCase to title case business terms.\n- Expand common abbreviations only when likely: `id` -> `Identifier`, `qty` -> `Quantity`, `amt` -> `Amount`, `dt` -> `Date`, `cd` -> `Code`, `desc` -> `Description`.\n- Do not invent acronyms unless they are present in the source.\n- Create related terms from relationship context, such as `Customer`, `Customer Identifier`, and `Customer Segment`.\n- Set `owner_or_steward` to `Unknown` or `TBD`, and `status` to `Proposed`, unless explicit.\n\n## Purview-Oriented Mapping\n\nUse this mapping for catalog ingestion planning:\n\n| Proposal Area | Purview-Oriented Target |\n|---------------|-------------------------|\n| Asset metadata | Data map asset attributes, custom attributes, contacts, collections |\n| Business glossary terms | Glossary terms and term relationships |\n| Classification proposals | Custom or built-in classifications for review |\n| Sensitivity proposals | Microsoft Purview Information Protection label proposal, not applied label |\n| Data quality rules | Data quality rule backlog or observability configuration |\n| Relationships and lineage hints | Lineage relationships or process mappings for validation |\n| CDE candidates | Governance critical data element candidate register |\n| Data products | Data product\u002Fdomain grouping proposal |\n\nNever state that a Purview label or classification has been applied.\n\n## Machine-Readable JSON Schema\n\nThe JSON file must follow this compact top-level structure:\n\n```json\n{\n  \"metadata_proposal\": {\n    \"generated_at\": \"ISO-8601 timestamp\",\n    \"generated_by\": \"fabric-metadata-creation agent\",\n    \"proposal_version\": \"1.0\",\n    \"input_interpreted\": {},\n    \"executive_summary\": \"\",\n    \"assets\": [],\n    \"columns\": [],\n    \"glossary_terms\": [],\n    \"classifications\": [],\n    \"sensitivity_labels\": [],\n    \"critical_data_elements\": [],\n    \"data_products\": [],\n    \"relationships_lineage\": [],\n    \"data_quality_rules\": [],\n    \"assumptions\": [],\n    \"open_questions\": []\n  }\n}\n```\n\nEvery asset and column object must include `metadata_source`, `confidence`, `assumptions`, and `open_questions`. Other proposal objects must include `metadata_source` and `confidence`. Use `review_required` only when it materially helps prioritize steward review.\n\n## Output Authoring Rules\n\n- Keep `metadata_proposal.json` compact and machine-readable; do not add legacy YAML, Markdown proposal, or evidence files.\n- Render `metadata_proposal.pdf` from `metadata_proposal.json` using the helper script's Markdown-to-WeasyPrint renderer.\n- Create `glossary_terms.md` as a standalone document that can be reviewed independently from the proposal PDF.\n- Render `glossary_terms.pdf` from structured glossary terms using the helper script's Markdown-to-WeasyPrint renderer; do not manually export the PDF.\n- Use A4 pages with 1.5cm margins, Segoe UI\u002FHelvetica\u002FArial fonts, blue headings, zebra-striped tables, and VS Code-style code blocks.\n- Write review prose, section titles, field labels, assumptions, questions, and generated descriptions in Spanish.\n- Preserve original Fabric identifiers and source values exactly; do not translate table names, column names, measure names, workspace names, lakehouse names, semantic model names, or explicit glossary\u002Fsource terms.\n- Use concise enterprise language. Prefer brief explanations over long generated prose.\n- Include confidence and metadata source in every proposal object.\n\n## Validation Rules\n\nBefore finishing, validate:\n\n1. `metadata_proposal.json` parses as JSON.\n2. `metadata_proposal.pdf` exists and is non-empty.\n3. `glossary_terms.md` exists and includes a standalone glossary title.\n4. `glossary_terms.pdf` exists and is non-empty.\n5. No legacy artifacts exist in the output folder: `metadata_proposal.md`, `metadata_proposal.yaml`, or `evidence.json`.\n6. Every asset and column proposal includes `confidence`, `metadata_source`, `assumptions`, and `open_questions`.\n7. No real owner, steward, contact, legal obligation, retention policy, access policy, or compliance decision was invented.\n\nUse:\n\n```bash\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py validate \u003Coutput_dir>\n```\n\n## Gotchas\n\n- A physical column named `customer_id` is not proof that the field is PII. It is an identifier candidate and may become personal data only when it can identify or link to a person.\n- A field named `name` can describe a person, product, department, status, or location. Use surrounding table context before proposing PII.\n- Financial measures are often confidential business data even when they are not personal data.\n- Health-related names such as `patient`, `diagnosis`, or `treatment` are strong sensitivity signals, but still require human review.\n- Semantic model measures often carry business meaning. Use DAX expressions and format strings as explicit evidence for metric descriptions.\n- Hidden key columns in semantic models are still governance-relevant and should be included in metadata proposals.\n- Semantic model `getDefinition` calls may return `202 Accepted`. The helper must poll the operation and then fetch the operation result payload; the operation status itself does not contain the TMDL parts.\n- Do not assign `Public` unless public availability is explicit or strongly implied by context.\n- Do not assign legal categories such as GDPR, HIPAA, PCI, or SOX as definitive obligations. Use `possible regulatory review area` only when the source context suggests it.\n- If no sample values are available, avoid sample-based claims and reduce confidence.\n","# Fabric Metadata Creation - Reference\n\nGenerates non-destructive metadata proposals for Microsoft Fabric lakehouse tables, files, views, and semantic models. The result is a steward-review package that can later be mapped into a data catalog such as Microsoft Purview, but it must not apply catalog changes automatically.\n\nThe agent's core principle is traceability: every suggestion must distinguish metadata explicitly found in the source from metadata inferred by the agent, include a confidence level, and carry assumptions or open questions when evidence is incomplete.\n\n## Global Rules\n\n- **Proposal only.** Never modify Fabric assets, Purview assets, glossary terms, labels, classifications, owners, stewards, retention policies, or access controls.\n- **Human review required.** Governance outcomes are suggestions, not decisions.\n- **No invented accountability.** Do not invent real owners, stewards, contacts, responsible teams, legal requirements, or compliance decisions. Use `Unknown`, `TBD`, or an open question when not explicit.\n- **Evidence labeling.** Every metadata object must include `metadata_source` with one of: `explicit`, `inferred`, or `mixed`.\n- **Confidence required.** Every proposal must include `confidence`: `high`, `medium`, or `low`.\n- **Assumptions required.** Every inferred proposal should include `assumptions`; use an empty list only when there are no assumptions.\n- **Open questions required.** Every asset-level and column-level proposal must include review questions when evidence is incomplete.\n- **Conservative language.** Use `candidate`, `suggested`, `likely`, `possible`, or `requires human review` for inferred classifications and sensitivity labels.\n- **Structured output.** Generate only the four final artifacts documented below: proposal JSON\u002FPDF and standalone glossary Markdown\u002FPDF.\n- **Spanish review documents.** Human-readable artifacts must be written in Spanish. Keep original workspace, lakehouse, table, column, measure, semantic model, and glossary candidate names exactly as they appear in the source; do not translate source identifiers.\n\n## Helper Script\n\nUse [.\u002Fscripts\u002Ffabric_metadata.py](.\u002Fscripts\u002Ffabric_metadata.py) for Fabric discovery\u002Fexport, PDF rendering, and output validation. The agent must activate `.venv\u002F` before running it.\n\nInstall dependencies:\n\n```bash\ntest -d .venv || python3 -m venv .venv\nsource .venv\u002Fbin\u002Factivate\npip install azure-identity requests markdown weasyprint\n```\n\nWeasyPrint requires system-level Cairo\u002FPango libraries. They are pre-installed on many Linux environments; on macOS install Pango with `brew install pango`.\n\nCommands:\n\n```bash\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py list-workspaces\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py list-items \u003Cworkspace_id> [\u003Citem_type>]\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py list-lakehouse-tables \u003Cworkspace_id> \u003Clakehouse_id>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py get-lakehouse-table \u003Cworkspace_id> \u003Clakehouse_id> \u003Ctable_name>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py list-semantic-models \u003Cworkspace_id> [\u003Cmodel_name>]\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py export-semantic-model \u003Cworkspace_id> \u003Csemantic_model_id> \u003Coutput_dir>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py render-pdf \u003Cmetadata_proposal.json> \u003Cmetadata_proposal.pdf>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py render-glossary-md \u003Cmetadata_proposal.json> \u003Cglossary_terms.md>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py render-glossary-pdf \u003Cmetadata_proposal.json> \u003Cglossary_terms.pdf>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py render-markdown-pdf \u003Cinput.md> \u003Coutput.pdf>\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py validate \u003Coutput_dir>\n```\n\nUse MCP tools first when they are available for interactive discovery. Use the script when MCP does not expose the needed metadata, when exporting semantic model definitions, or when rendering\u002Fvalidating local outputs.\n\n## Output Directory\n\nSave each run to:\n\n```text\n.\u002Fmetadata_proposals\u002F{SOURCE_NAME}\u002F{YYYY-MM-DD_HHmmss}\u002F\n```\n\nRequired files:\n\n| File | Purpose |\n|------|---------|\n| `metadata_proposal.json` | Compact machine-readable metadata proposal |\n| `metadata_proposal.pdf` | Styled A4 proposal PDF rendered from Markdown with WeasyPrint |\n| `glossary_terms.md` | Standalone human-readable glossary proposal |\n| `glossary_terms.pdf` | Styled A4 glossary PDF rendered from Markdown with WeasyPrint |\n\nDo not generate `metadata_proposal.md`, `metadata_proposal.yaml`, `evidence.json`, or other final artifacts. Temporary discovery\u002Fexport files may be used while working, but they must not remain in the final output directory.\n\n## Default Proposal PDF Sections\n\nThe proposal PDF must present these simplified review sections in order. The JSON is the source of truth; the PDF is the human-readable rendering.\n\n1. Resumen ejecutivo\n2. Entrada interpretada\n3. Activos y columnas\n4. Glosario\n5. Sugerencias de gobierno\n6. Relaciones, linaje y calidad de datos\n7. Supuestos y preguntas abiertas\n\nBoth PDFs must include an index\u002Ftable of contents. Sections and subsections must be numbered. Asset sections must be visually isolated with page breaks, and the asset name must appear in the asset table header. Source names that contain Markdown-looking characters, such as `# Case Updates`, must be rendered as literal names rather than Markdown syntax.\n\n## Metadata Categories\n\n### Asset-Level Metadata\n\nUse these compact fields for databases, schemas, tables, views, files, semantic models, measures, dimensions, and relationship assets when applicable:\n\n```text\nasset_name, asset_type, fully_qualified_name, platform, source_system,\ndatabase_name, schema_name, object_type, data_domain, data_product,\nenvironment, lifecycle_status, business_description, technical_description,\ndata_owner, data_steward, glossary_terms, upstream_lineage,\ndownstream_lineage, source_of_truth, security_notes, quality_notes,\nconfidence\n```\n\nAdditional required traceability fields: `metadata_source`, `assumptions`, `open_questions`.\n\n### Column-Level Metadata\n\nUse these compact fields for columns, semantic model columns, calculated columns, measures, and dimensional attributes when applicable:\n\n```text\ncolumn_name, business_name, data_type, nullable, key_role,\nmeasure_or_dimension, business_description, technical_description,\nsemantic_type, classification, sensitivity_label, pii_indicator,\ncritical_data_element_candidate, derivation_logic, allowed_values,\nexample_values, quality_rules, relationship_hints, lineage_hints,\nmasking_recommendation, confidence\n```\n\nAdditional required traceability fields: `metadata_source`, `assumptions`, `open_questions`.\n\n### Business Glossary Term Metadata\n\nUse these compact fields for suggested business glossary terms:\n\n```text\nterm_name, definition, synonyms, related_terms, domain, status,\nowner_or_steward, associated_assets, associated_columns, confidence,\nmetadata_source, open_questions\n```\n\nUse `status: Proposed` unless an explicit existing glossary status is present.\n\nThe glossary must also be published as an independent human-readable output in `glossary_terms.md` and `glossary_terms.pdf`. This glossary is a steward-review artifact, not a Purview import file. It should be readable without opening the full metadata proposal.\n\nRecommended glossary presentation:\n\n1. Title and generation context\n2. Summary by domain or subject area\n3. Alphabetical term list grouped by domain or subject area\n4. For each term: definition, synonyms, related terms, associated assets, associated columns, confidence, metadata source, and open questions\n5. Review guidance explaining that terms are proposed and require steward approval\n\n### Classification Metadata\n\nUse these fields for classification proposals:\n\n```text\nclassification_name, classification_reason, detection_basis, confidence\n```\n\nAllowed classification names include, when evidence supports them: `personal data`, `financial data`, `operational data`, `identifier`, `reference data`, `master data`, `transactional data`, `metric`, `dimension`, `audit field`, `system field`, `derived field`.\n\n### Sensitivity Metadata\n\nUse these fields for Purview-like sensitivity label proposals:\n\n```text\nsuggested_label, label_reason, protection_recommendation, confidence\n```\n\nAllowed suggested labels: `Public`, `Internal`, `Confidential`, `Highly Confidential`, `Restricted`.\n\n### Data Quality Metadata\n\nUse these fields for quality rules and profiling recommendations:\n\n```text\nquality_dimension, quality_rule, severity, confidence\n```\n\nQuality dimensions include: `completeness`, `uniqueness`, `validity`, `accuracy`, `consistency`, `timeliness`, `freshness`, `integrity`, `conformity`, `reasonableness`.\n\n### Lineage and Relationship Metadata\n\nUse these fields for relationship and lineage hints:\n\n```text\nupstream_source, downstream_consumer, relationship_or_join_hint, confidence\n```\n\n### Critical Data Element Metadata\n\nUse these fields for CDE candidates:\n\n```text\ncde_name, business_reason, associated_columns, confidence\n```\n\n### Data Product Metadata\n\nUse these fields for suggested data products or domain groupings:\n\n```text\ndata_product_name, purpose, included_assets, target_consumers,\nowner_or_steward, confidence\n```\n\n## Evidence and Confidence Rules\n\n### Metadata Source\n\n| Value | Meaning |\n|-------|---------|\n| `explicit` | Directly present in Fabric metadata, TMDL, schema, annotations, descriptions, names, or user-provided context |\n| `inferred` | Deduced from names, data types, relationships, sample values, model structure, or common enterprise patterns |\n| `mixed` | Combines explicit evidence with inferred enrichment |\n\n### Confidence Levels\n\n| Confidence | Use When |\n|------------|----------|\n| `high` | Strong explicit evidence exists, or multiple independent signals agree, such as name pattern plus data type plus sample pattern |\n| `medium` | Reasonable inference from names, data types, or model context, but no validating sample or explicit business description |\n| `low` | Weak evidence, ambiguous abbreviations, missing samples, unclear domain, or conflicting signals |\n\nNever raise confidence only because a term sounds plausible. Lower confidence when samples, row counts, relationships, or business context are missing.\n\n## Inference Rules\n\n### Asset Type and Domain\n\nUse table\u002Fmodel names, workspace names, lakehouse names, schemas, and measure names to infer candidate domains. Examples:\n\n| Pattern | Suggested Domain or Context |\n|---------|-----------------------------|\n| `customer`, `client`, `account`, `contact` | Customer, sales, CRM, master data |\n| `order`, `invoice`, `payment`, `booking`, `transaction` | Sales, finance, transactional process |\n| `product`, `item`, `sku`, `catalog` | Product, inventory, reference\u002Fmaster data |\n| `employee`, `staff`, `payroll`, `hr` | Workforce or HR; possible sensitive data |\n| `patient`, `diagnosis`, `doctor`, `visit`, `hospital` | Healthcare; possible regulated data |\n| `flight`, `passenger`, `airport`, `aircraft` | Airline operations |\n| `incident`, `ticket`, `case`, `status`, `priority` | Operations or service management |\n| `calendar`, `date`, `time` | Date\u002Ftime dimension |\n| `audit`, `log`, `event`, `telemetry` | Operational, audit, or observability data |\n\nIf domain is ambiguous, use `Unknown` or a broad proposed domain and add an open question.\n\n### Descriptions\n\nGenerate two descriptions for each asset and column:\n\n- **Technical description**: what the object stores structurally, based on schema and type evidence.\n- **Business description**: what the object likely means to a business user, based on names, model context, relationships, and measures.\n\nDescriptions must be concise, deterministic, and free of unsupported claims. For low-confidence descriptions, explicitly say `Candidate description` or `Likely represents`.\n\n### Keys and Relationships\n\nInfer key metadata with these signals:\n\n| Signal | Suggested Metadata |\n|--------|--------------------|\n| Column named `id`, `{table}_id`, `{entity}_id`, `{entity}_key`, `uuid` | Candidate primary, foreign, natural, or surrogate key depending on context |\n| Single non-null unique ID-like column in a dimension table | Candidate primary key |\n| Fact table column matching a dimension PK name\u002Ftype | Candidate foreign key |\n| Columns named `code`, `number`, `reference`, `dni`, `nif`, `passport`, `iban` | Candidate natural key or business identifier |\n| Integer sequential ID column | Candidate surrogate key |\n| Composite repeated columns such as `order_id` + `line_number` | Candidate composite key |\n\nDefault cardinality to `many-to-one candidate` only when a transactional\u002Ffact-like table points to a reference\u002Fdimension-like table. Use `unknown` when there is insufficient evidence.\n\n### Measure and Dimension Detection\n\n| Pattern | Suggestion |\n|---------|------------|\n| Numeric additive columns: `amount`, `total`, `quantity`, `cost`, `revenue`, `sales`, `importe`, `cantidad` | Metric\u002Fmeasure candidate |\n| Descriptive strings: `name`, `description`, `category`, `status`, `type`, `city`, `country` | Dimension attribute candidate |\n| DAX measure expression exists | Explicit measure |\n| Calculated expression exists | Derived field candidate, include derivation logic |\n\n### Classification Detection\n\n| Pattern | Candidate Classification |\n|---------|--------------------------|\n| `id`, `_id`, `_key`, `uuid`, `guid`, `code`, `number` | Identifier |\n| `email`, `phone`, `mobile`, `telefono`, `address`, `dni`, `nif`, `nie`, `passport`, `name`, `birth_date` | Personal data candidate |\n| `iban`, `bank`, `card`, `salary`, `payroll`, `payment`, `invoice`, `amount`, `revenue`, `margin`, `cost`, `price` | Financial data candidate |\n| `patient`, `diagnosis`, `medical`, `doctor`, `visit`, `treatment` | Regulated or sensitive data candidate |\n| `status`, `category`, `type`, `priority`, `country`, `city`, `currency` | Reference data candidate |\n| `created_at`, `updated_at`, `loaded_at`, `ingestion_time`, `batch_id`, `source_file`, `run_id` | Audit or system field |\n| `score`, `ratio`, `total`, `average`, `margin`, `age`, `duration`, `days_since` | Derived field or metric candidate |\n\nDo not use regex-only matches as definitive evidence. Describe uncertainty in `classification_reason` for sensitive classifications.\n\n### Sensitivity Label Suggestions\n\n| Suggested Label | Use When |\n|-----------------|----------|\n| `Public` | Asset appears intended for open\u002Fpublic use and no sensitive indicators are present. Rarely infer this without explicit context. |\n| `Internal` | Default for ordinary enterprise metadata without sensitive indicators. |\n| `Confidential` | Contains candidate business confidential data, identifiers, contact data, financial values, or internal operational data. |\n| `Highly Confidential` | Contains likely PII, financial identifiers, health-related fields, security-relevant fields, or combinations of sensitive attributes. |\n| `Restricted` | Contains secrets, credentials, tokens, private keys, strong regulated data indicators, or explicitly restricted labels in source metadata. |\n\nTreat every inferred label above `Internal` as requiring human review.\n\n### Privacy and Sensitive Data Signals\n\nPotential PII and sensitive fields include:\n\n```text\nname, first_name, last_name, full_name, email, phone, mobile, address, city,\npostal_code, latitude, longitude, dni, nif, nie, ssn, passport, national_id,\ntax_id, birth_date, gender, patient_id, diagnosis, medical_record, iban,\nbank_account, credit_card, salary, password, secret, token, api_key, private_key,\nip_address, device_id, user_agent\n```\n\nMark these as candidates unless sample values or explicit metadata confirm the pattern.\n\n### Data Quality Rules\n\nSuggest quality rules from type and classification evidence:\n\n| Evidence | Suggested Rule |\n|----------|----------------|\n| Candidate primary key | Completeness and uniqueness checks, expected threshold 100% |\n| Candidate foreign key | Referential integrity check against referenced asset |\n| Required business column | Completeness check with agreed null threshold |\n| Email\u002Fphone\u002Fidentifier pattern | Validity and conformity checks |\n| Date\u002Ftimestamp column | Freshness, timeliness, and valid range checks |\n| Amount\u002Fquantity\u002Fmetric | Reasonableness, non-negative, range, outlier, and aggregation checks |\n| Status\u002Fcategory\u002Fcode | Allowed values and reference set checks |\n| Audit field | Presence, monotonicity, ingestion recency, and run consistency checks |\n\nSeverity guidance:\n\n| Severity | Use When |\n|----------|----------|\n| `high` | Key integrity, sensitive fields, CDE candidates, business-critical metrics |\n| `medium` | Important dimensions, common reporting filters, relationship fields |\n| `low` | Descriptive fields or optional enrichment attributes |\n\n### Critical Data Element Candidates\n\nSuggest a CDE candidate when one or more conditions apply:\n\n- Column is a key identifier used in relationships.\n- Column is a financial, operational, compliance, or customer-impacting metric.\n- Column appears in a semantic model measure or relationship.\n- Column is likely used for regulatory, privacy, security, or executive reporting.\n- Column is needed to join important assets or identify master\u002Freference entities.\n\nAlways describe CDEs as candidates and avoid definitive CDE status.\n\n### Glossary Term Generation\n\nCreate glossary terms by converting physical names to business names and grouping repeated concepts.\n\nRules:\n\n- Convert snake_case and camelCase to title case business terms.\n- Expand common abbreviations only when likely: `id` -> `Identifier`, `qty` -> `Quantity`, `amt` -> `Amount`, `dt` -> `Date`, `cd` -> `Code`, `desc` -> `Description`.\n- Do not invent acronyms unless they are present in the source.\n- Create related terms from relationship context, such as `Customer`, `Customer Identifier`, and `Customer Segment`.\n- Set `owner_or_steward` to `Unknown` or `TBD`, and `status` to `Proposed`, unless explicit.\n\n## Purview-Oriented Mapping\n\nUse this mapping for catalog ingestion planning:\n\n| Proposal Area | Purview-Oriented Target |\n|---------------|-------------------------|\n| Asset metadata | Data map asset attributes, custom attributes, contacts, collections |\n| Business glossary terms | Glossary terms and term relationships |\n| Classification proposals | Custom or built-in classifications for review |\n| Sensitivity proposals | Microsoft Purview Information Protection label proposal, not applied label |\n| Data quality rules | Data quality rule backlog or observability configuration |\n| Relationships and lineage hints | Lineage relationships or process mappings for validation |\n| CDE candidates | Governance critical data element candidate register |\n| Data products | Data product\u002Fdomain grouping proposal |\n\nNever state that a Purview label or classification has been applied.\n\n## Machine-Readable JSON Schema\n\nThe JSON file must follow this compact top-level structure:\n\n```json\n{\n  \"metadata_proposal\": {\n    \"generated_at\": \"ISO-8601 timestamp\",\n    \"generated_by\": \"fabric-metadata-creation agent\",\n    \"proposal_version\": \"1.0\",\n    \"input_interpreted\": {},\n    \"executive_summary\": \"\",\n    \"assets\": [],\n    \"columns\": [],\n    \"glossary_terms\": [],\n    \"classifications\": [],\n    \"sensitivity_labels\": [],\n    \"critical_data_elements\": [],\n    \"data_products\": [],\n    \"relationships_lineage\": [],\n    \"data_quality_rules\": [],\n    \"assumptions\": [],\n    \"open_questions\": []\n  }\n}\n```\n\nEvery asset and column object must include `metadata_source`, `confidence`, `assumptions`, and `open_questions`. Other proposal objects must include `metadata_source` and `confidence`. Use `review_required` only when it materially helps prioritize steward review.\n\n## Output Authoring Rules\n\n- Keep `metadata_proposal.json` compact and machine-readable; do not add legacy YAML, Markdown proposal, or evidence files.\n- Render `metadata_proposal.pdf` from `metadata_proposal.json` using the helper script's Markdown-to-WeasyPrint renderer.\n- Create `glossary_terms.md` as a standalone document that can be reviewed independently from the proposal PDF.\n- Render `glossary_terms.pdf` from structured glossary terms using the helper script's Markdown-to-WeasyPrint renderer; do not manually export the PDF.\n- Use A4 pages with 1.5cm margins, Segoe UI\u002FHelvetica\u002FArial fonts, blue headings, zebra-striped tables, and VS Code-style code blocks.\n- Write review prose, section titles, field labels, assumptions, questions, and generated descriptions in Spanish.\n- Preserve original Fabric identifiers and source values exactly; do not translate table names, column names, measure names, workspace names, lakehouse names, semantic model names, or explicit glossary\u002Fsource terms.\n- Use concise enterprise language. Prefer brief explanations over long generated prose.\n- Include confidence and metadata source in every proposal object.\n\n## Validation Rules\n\nBefore finishing, validate:\n\n1. `metadata_proposal.json` parses as JSON.\n2. `metadata_proposal.pdf` exists and is non-empty.\n3. `glossary_terms.md` exists and includes a standalone glossary title.\n4. `glossary_terms.pdf` exists and is non-empty.\n5. No legacy artifacts exist in the output folder: `metadata_proposal.md`, `metadata_proposal.yaml`, or `evidence.json`.\n6. Every asset and column proposal includes `confidence`, `metadata_source`, `assumptions`, and `open_questions`.\n7. No real owner, steward, contact, legal obligation, retention policy, access policy, or compliance decision was invented.\n\nUse:\n\n```bash\npython .github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py validate \u003Coutput_dir>\n```\n\n## Gotchas\n\n- A physical column named `customer_id` is not proof that the field is PII. It is an identifier candidate and may become personal data only when it can identify or link to a person.\n- A field named `name` can describe a person, product, department, status, or location. Use surrounding table context before proposing PII.\n- Financial measures are often confidential business data even when they are not personal data.\n- Health-related names such as `patient`, `diagnosis`, or `treatment` are strong sensitivity signals, but still require human review.\n- Semantic model measures often carry business meaning. Use DAX expressions and format strings as explicit evidence for metric descriptions.\n- Hidden key columns in semantic models are still governance-relevant and should be included in metadata proposals.\n- Semantic model `getDefinition` calls may return `202 Accepted`. The helper must poll the operation and then fetch the operation result payload; the operation status itself does not contain the TMDL parts.\n- Do not assign `Public` unless public availability is explicit or strongly implied by context.\n- Do not assign legal categories such as GDPR, HIPAA, PCI, or SOX as definitive obligations. Use `possible regulatory review area` only when the source context suggests it.\n- If no sample values are available, avoid sample-based claims and reduce confidence.\n",{"name":5,"description":7,"compatibility":49},"Requires the Fabric MCP Server VS Code extension for interactive discovery when available. Uses Python helper scripts with azure-identity and requests for Fabric discovery\u002Fexport and markdown plus weasyprint for styled local PDFs. Produces local proposal artifacts only.","main","a54dec65de5a1d2a62bd5cbd9e5003adca5f5a84",".github\u002Fskills\u002Ffabric-metadata-creation",1,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fgh-copilot-fabric-agents","https:\u002F\u002Fgithub.com\u002Fmicrosoft","https:\u002F\u002Fopensource.microsoft.com","Open source projects and samples from Microsoft","2026-06-04 21:31:34",{"source":8,"skillName":5,"commands":60,"prompt":70,"zipAvailable":71},[61,64,67],{"tool":62,"command":63},"npx","npx skills add microsoft\u002Fgh-copilot-fabric-agents --skill fabric-metadata-creation",{"tool":65,"command":66},"bunx","bunx skills add microsoft\u002Fgh-copilot-fabric-agents --skill fabric-metadata-creation",{"tool":68,"command":69},"pnpm","pnpm dlx skills add microsoft\u002Fgh-copilot-fabric-agents --skill fabric-metadata-creation","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fgh-copilot-fabric-agents\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Ffabric-metadata-creation\u002FSKILL.md",true,[73],{"path":74,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":77,"sourceHtmlUrl":78},"scripts\u002Ffabric_metadata.py","SCRIPT","脚本",46398,"https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fgh-copilot-fabric-agents\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Ffabric-metadata-creation\u002Fscripts\u002Ffabric_metadata.py",[80,103,121,137],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":98,"publishedAt":44},"01M69BP4V2GV4WZVHWX5A65QEW","fabric-synthetic-data","Generate and upload synthetic data to Microsoft Fabric lakehouse tables. Covers star-schema generation, realistic fake data (names, dates, IDs, transactions), configurable row counts, referential integrity between tables, Parquet export, and OneLake upload. Use when: synthetic data, test data, fake data, sample data, generate tables, populate lakehouse, seed data, mock data.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fgh-copilot-fabric-agents\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Ffabric-synthetic-data\u002FSKILL.md",[86,89,90,95],{"categoryNo":87,"categoryName":88,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000010","后端与接口",{"categoryNo":24,"categoryName":25,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":91,"categoryName":92,"taxonomyType":93,"taxonomyTypeDesc":94},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",{"categoryNo":96,"categoryName":97,"taxonomyType":93,"taxonomyTypeDesc":94},"01JBSEEDCAT000000000000033","数据分析师",[99,28,100,43,41,39,101,9,37,102,31,30],"synthetic data","fabric synthetic data","test","generate",{"skillNo":104,"name":105,"nameZh":6,"description":106,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":107,"categories":108,"keywords":116,"publishedAt":44},"01M69BP1F17XK0TFKMDS23QNEV","fabric-semantic-model","Infer and generate Power BI semantic models (TMDL format) from Microsoft Fabric lakehouse tables. Covers star-schema detection, table classification (fact\u002Fdimension), relationship inference, DAX measure generation, data type mapping, and deployment via REST API. Use when: create semantic model, build data model, star schema, Power BI model, TMDL, fact table, dimension table.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fgh-copilot-fabric-agents\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Ffabric-semantic-model\u002FSKILL.md",[109,110,113,114,115],{"categoryNo":87,"categoryName":88,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":111,"categoryName":112,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000013","分析与可视化",{"categoryNo":24,"categoryName":25,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":91,"categoryName":92,"taxonomyType":93,"taxonomyTypeDesc":94},{"categoryNo":96,"categoryName":97,"taxonomyType":93,"taxonomyTypeDesc":94},[28,117,118,9,39,37,119,120,32,102,31,30,40,43],"fabric semantic model","dax","deployment","power bi",{"skillNo":122,"name":123,"nameZh":6,"description":124,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":125,"categories":126,"keywords":128,"publishedAt":44},"01M69BNXPJX1D29F4SAD54K1HZ","fabric-optimization-review","Review and optimize data storage, data models, and semantic models in Microsoft Fabric lakehouses. Covers Delta table optimization (V-Order, Z-Order, compaction, partitioning), data model anti-pattern detection, semantic model DAX review, relationship optimization, AI\u002Fagentic readiness (descriptions, synonyms, linguistic schema, display folders), workspace audit (Fabric best practices, dimensional modeling, Direct Lake), and actionable recommendation reports. Use when: optimization, performance, review, tuning, best practices, V-Order, Z-Order, partitioning, compaction, DAX optimization, data model review, AI readiness, agentic, NLQ, Copilot, natural language, workspace audit, Direct Lake, dimensional modeling, audit.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fgh-copilot-fabric-agents\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Ffabric-optimization-review\u002FSKILL.md",[127],{"categoryNo":24,"categoryName":25,"taxonomyType":21,"taxonomyTypeDesc":22},[42,28,129,130,131,43,41,132,39,133,9,118,134,135,37,136,31],"fabric optimization review","best practices","performance","agentic","storage","dax optimization","copilot","pattern detection",{"skillNo":138,"name":139,"nameZh":6,"description":140,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":141,"categories":142,"keywords":145,"publishedAt":44},"01M69BNNR7HC8MMMB1W3DS371C","fabric-data-cleaner","Algorithm reference and self-contained PySpark notebook templates for cleaning Microsoft Fabric lakehouse tables. Covers profiling, duplicate detection, null analysis, type validation, statistics, IQR outlier detection, date format validation, Spanish DNI\u002FNIE checksum validation, and email\u002Fphone format checks. Each notebook can be uploaded and run independently in Fabric.","https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fgh-copilot-fabric-agents\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Ffabric-data-cleaner\u002FSKILL.md",[143,144],{"categoryNo":24,"categoryName":25,"taxonomyType":21,"taxonomyTypeDesc":22},{"categoryNo":96,"categoryName":97,"taxonomyType":93,"taxonomyTypeDesc":94},[146,147,28,39,9,148,37,149,150,31,30,151,43],"fabric data","fabric data cleaner","vscode","pyspark","run","validation"]