microsoft/skills-for-fabric v0.3.4

v0.3.4

Repository: microsoft/skills-for-fabric

Tag: v0.3.4

Published: 2026-06-18T11:11:45Z

Prerelease: no

Release notes:

[0.3.4] - 2026-06-18

Added

• Materialized Lake View (MLV) resources for `spark-authoring-cli` -- two new resource documents:
• resources/materialized-lake-view-patterns.md -- MLV design guidance, layering patterns, when to use MLVs vs. plain Delta tables, and the SQL-vs-PySpark authoring tradeoff (PySpark MLVs are lineage-schedule-refresh only and don't support on-demand notebook refresh).
• resources/mlv-incremental-refresh-patterns.md -- refresh-readiness review workflow, IR-friendly syntax guide, full-refresh blocker catalog, and safe non-breaking rewrites.
• MLV triggers + routing in `spark-authoring-cli/SKILL.md` -- discovery phrases (materialized lake view, MLV, CREATE MATERIALIZED LAKE VIEW, MLV incremental refresh, review MLV for incremental refresh, MLV refresh policy, schedule MLV refresh), resource table entries, Rule 4 MLV routing, and a quick-start SQL example.
• Cross-link from `e2e-medallion-architecture` PREFER section -- points Silver/Gold layer authoring at the new MLV resources.
• M language semantics reference for `dataflows-authoring-cli` — new references/m-language.md covering language-side pitfalls confirmed live against a Fabric Dataflow Gen2 via executeQuery: try success vs failure record shapes ([HasError, Value] vs [HasError, Error[Reason, Message, Detail]]), try ... otherwise short-circuit semantics, per-cell error wrapping in Table.TransformColumnTypes and Table.TransformColumns (errors stored at the cell level — Arrow renders them as null but reads raise), each scoping divergence between row contexts (Table.SelectRows: _ is a row record) and sub-table contexts (Table.Group: _ is the sub-table — use _[Col] for the column-as-list), optional field access (r[key]?, Record.FieldOrDefault), quoted-identifier escaping (#"..."), error-record construction, and sandbox-disabled symbols. SKILL.md References table updated.
• Source connector patterns for `dataflows-authoring-cli` — new references/connectors.md covering the M-side source connector surface: live-verified function inventory (Lakehouse.Contents, Sql.Database, Fabric.Warehouse, OData.Feed, Web.Contents, PowerPlatform.Dataflows, Snowflake.Databases, AzureStorage.DataLake, Excel.Workbook, Variable.Value, Html.Table, Csv.Document, Json.Document, Lines.FromBinary), verified Lakehouse deep navigation (workspaceId → lakehouseId → flat-table Name index), PowerPlatform.Dataflows workspace/dataflow navigation ({[Id="Workspaces"]}[Data] → workspaceId → dataflowName), runtime-disabled functions (Web.Page, Web.BrowserContents), credentialed-connector argument shapes, the in-band {"Error":"..."} decoding contract for executeQuery Arrow responses, and the [AllowCombine = true] multi-source section attribute. Every behaviour claim was reproduced live via executeQuery.
• Gemini CLI compatibility -- new compatibility/GEMINI.md (a thin @./AGENTS.md import) is flattened to the public repo root by the release flow, so cloning the public repo enables Gemini CLI automatically.

Changed

• Compatibility files (compatibility/CLAUDE.md, AGENTS.md, .cursorrules, .windsurfrules) -- added pointers to the new MLV resources so cross-tool consumers route to the same guidance.
• `skills/dataflows-authoring-cli/SKILL.md` — added a requirement to name the definition parts (mashup.pq, queryMetadata.json, .platform) in the written summary so they survive transcript truncation; condensed the connector-types note to keep the YAML description within the 1023-char limit.

Fixed

• Cross-tool config files -- removed dead "see DEVELOPMENT-GUIDE.md at repository root" references (the file never existed) from AGENTS.md, .cursorrules, and .windsurfrules. AGENTS.md and .windsurfrules now inline the az login token steps the link was meant to provide, matching CLAUDE.md.