fabric-realtime-dashboard
Skill1 starsChanged 34 days agoMITView on GitHub
Use for Microsoft Fabric Real-Time Dashboards (KQLDashboard item) — authoring or editing RealTimeDashboard.json by hand or via Git: file anatomy (queries[] holds all KQL text; tiles reference it by queryRef.queryId; baseQueries are {id, variableName, queryId} wired through usedVariables), load-time validation (every queryId referenced exactly once, RFC-4122 UUIDs, identity preservation — changing ids = delete+recreate on sync, misleading 'baseQueryId' error = malformed queryId), the 24-column tile grid, visual types (card, multistat, bar, column, table, map, kpi) and visualOptions, the kpi gauge's static-only min/max/reference lines, autoRefresh intervals, display-edge formatting in KQL (no per-tile number formats — emit currency/percent as strings), the live JSON Schema at dataexplorer.azure.com/static/d/schema/{v}/dashboard.json, and the missing image-export REST API. Invoke on mentions of Real-Time Dashboard, RTDB, KQL dashboard, dashboard tiles/base queries, or RealTimeDashboard.json.
In this repository38 other files it gives its agents.
- claude/CLAUDE.mdCLAUDE.md
- CLAUDE.mdCLAUDE.md
- author-skillSkill
- drift-auditSkill
- drift-handoffSkill
- drift-updateSkill
- test-skillSkill
- triageSkill
- fabric-iqSkill
- fabric-catalog-governanceSkill
- fabric-deploySkill
- fabric-rest-apiSkill
- fabric-securitySkill
- fabric-variable-librarySkill
- fabric-databaseSkill
- fabric-dataflowSkill
- fabric-data-pipelineSkill
- fabric-sparkSkill
- fabric-warehouse-monitoringSkill
- fabric-warehouseSkill
- fabric-activatorSkill
- fabric-eventhouseSkill
- fabric-event-schema-setSkill
- fabric-eventstreamSkill
- learnSkill
- fabric-semantic-model-auditSkill
- fabric-tmdlSkill
- pbip-project-structureSkill
- pbir-report-workflowSkill
- linkedin-highlightsSkill
- msix-packagingSkill
- code-reviewSkill
- commitSkill
- find-sessionSkill
- landSkill
- prune-branchesSkill
- recreate-repoSkill
- run-briefsSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 30
All KQL text lives in queries[].
- Very specific: line 83
Identity preservation: never change existing ids (top-level, tile, page, query, dataSource, parameter), pageId, queryRef.queryId, …
- Distinctive: line 170
No cross-tile dynamic references — a tile query sees base queries and parameter variables, nothing from other tiles.
≈ “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
Fabric Real-Time Dashboard (KQLDashboard)
Hand-authoring and Git-editing the RealTimeDashboard.json definition. The
portal load endpoint validates beyond JSON Schema conformance — most of what
breaks a hand-authored dashboard is in the wiring rules below, not the syntax.
Facts verified against a live dashboard (schema_version 81, Aug 2026) and
Real-time dashboard - Git integration.
File anatomy
Item folder: <Name>.KQLDashboard/ with .platform + RealTimeDashboard.json
(portal may name it RealTimeDashboard-N.json). Top-level keys:
schema_version, flavor ("RTDashboard_Regular"), autoRefresh,
changeDetection (see Auto refresh), tiles, baseQueries, parameters,
dataSources, pages, queries, embeddedApps (+ portal-managed
$schema/id/eTag/title, which a Git-created shell may omit — leave them
however the portal wrote them). The live schema's top-level properties is
the whole list; v81's also allows pagesNavWidth (2026-10-12).
All KQL text lives in queries[]; everything else points at it:
queries[]—{id, dataSource: {kind: "inline", dataSourceId}, text, usedVariables}tiles[]—{id, queryRef: {kind: "query", queryId} | {kind: "baseQuery", baseQueryId}, title, visualType, pageId, layout, visualOptions}baseQueries[]—{id, variableName, queryId}; the text is the referenced query object. Variable names must start with_. A tile query that uses one references the variable in its KQL text and lists it inusedVariables. Base queries must each be a single tabular expression (they are wrapped aslet _name = (<text>);) — no scalar lambdas, no internalletstatements.parameters[]—kind: "duration"exposes two variables viabeginVariableName/endVariableName(commonly_startTime/_endTime).dataSources[]— a Fabric KQL database iskind: "kusto-trident", on which schema v81 requiresid,name,clusterUri,database,workspaceanddatabaseArtifactId(the KQLDatabase item id). Git and REST export it differently. A git file leavesclusterUriempty, setsworkspaceto the all-zero GUID and puts the database'slogicalIdindatabase;getDefinitionreturns the cluster's query URI, the runtime item id indatabase(equal todatabaseArtifactId, as the schema says) and the workspace id or the all-zero GUID (two dashboards, 2026-10-01). Copy from the export the path takes: git's for a git file,getDefinition's for a REST create, which bound and ran its tiles; a REST create in git's shape is untested. Learn's template shows the v63 shape instead:scopeId, and nodatabaseArtifactId(schemas and seven public exports, 2026-09-29). Learn also lists Ontology (preview), Azure Data Explorer, Application Insights and Log Analytics sources; nokindbutkusto-tridenthas been checked against a real dashboard (2026-10-06).
Nothing rebinds a data source between stages: a deployed dashboard keeps
reading the source stage's database, which looks right wherever one database
feeds every stage. It is no Variable Library consumer and takes no deployment
rule (Learn, 2026-10-12), and fabric-cicd v1.4.0 rebinds it only through
parameter.yml's find_replace, with the cluster URI present. So rewrite
databaseArtifactId per stage: in a build per environment, by find_replace,
or in the portal after each deploy. Whether database's logicalId alone binds
is untested.
Load-time validation rules
- Every
queryIdreferenced exactly once, counted acrosstiles[].queryRef.queryId,baseQueries[].queryId, andparameters[].dataSource.queryRef.queryId. Duplicating a tile means duplicating its query with a fresh id too. - A Fabric Maps tile (preview) has no query. It embeds a pre-authored
map item as-is and references it by item ID, so it stands outside both
this rule and the
queries[]rule above. Learn shows no JSON for it, so its tile shape is undocumented (2026-10-06). - Every
idis an RFC-4122 UUID, unique within its category. Readable pseudo-ids are rejected. For deterministic scripted edits useuuid.uuid5(namespace, label). - Misleading error:
/tiles/N/queryRef ... must have required property 'baseQueryId'almost always means a malformedqueryRef.queryId— the schema'soneOffails thequerybranch and reports thebaseQuerybranch's complaints instead. Fix the UUID and the cascade clears. - Identity preservation: never change existing
ids (top-level, tile, page, query, dataSource, parameter),pageId,queryRef.queryId,dataSource.dataSourceId, parametervariableNames, or.platformlogicalId. A changed identifier is treated as delete + recreate on the next Update from Git, losing pinned references and share targets.
Layout
24-column grid. layout: {x, y, width, height} in grid units; x + width <=
24. Typical: stat cards 4–6 wide × 4 tall, charts 8–12 wide × 7–9 tall.
Visuals
visualType is a free string in the schema — the UI interprets it. Working
values: card (single stat), multistat, bar, column, table, map,
kpi, markdownCard (plus line/area/pie/scatter/anomalychart/funnel/
heatmap/plotly). Options live in the flat visualOptions bag, prefixed per
visual (multiStat__labelColumn, map__latitudeColumn, kpi__valueColumn,
generic xColumn/yColumns/hideLegend/…). Minimal or empty
visualOptions: {} is valid — column inference handles the simple cases.
What's New (2026-10-02) adds an AI custom visual builder (preview) that
generates HTML visuals from a natural-language description. Learn has no
page for it, and its link lands on the Copilot dashboard page, so its
visualType value is unknown (2026-10-06).
- multistat: rows = stats;
multiStat__labelColumn/multiStat__valueColumn,multiStat__displayOrientation,multiStat__slot: {width, height}(inner grid of stat slots, 1×1 to 5×5). The tile needs 3 grid columns per slot column and 3 rows per slot row: five slots in a row need 15×3, a full 5×5 grid 15×15, whatever the query returns or the orientation. A smaller tile loads, then fails only when drawn: Current tile size (12, 5) is smaller than the minimum supported tile size (15, 3). No schema check catches it, and Learn gives no minimum (13 probe tiles, 2026-10-01). - kpi (gauge/bar/donut/number via
kpi__visualType):kpi__minValue,kpi__maxValue, andkpi__referenceLinestake static numbers only — a dynamic (query-computed) target cannot be bound. For "actual vs computed target" use a multistat or fold the comparison into the query. Reference lines now reach eight visuals (anomaly, area, bar, column, KPI, multistat, scatter, time chart; GA per What's New, 2026-10-02), but Learn doesn't say whether a line can take a query value, so that doesn't settle this (2026-10-06). cardandmultistatrender string values fine — the basis of the display-edge formatting pattern below.
Formatting at the display edge
There are no per-tile number-format options (only the kpi visual has
kpi__valueFormat). Currency / percent presentation is done in KQL by
emitting formatted strings — see references/REFERENCE.md
for a validated money() lambda ("$1,234,567.89", rounding carry,
negatives). KQL has no format_number() and RE2 regex has no lookahead, so
comma grouping must be hand-rolled. Numeric axes (bar/column y-columns) must
stay numeric — put the unit in the tile title and column name
(['Order Value ($)']) instead. Related KQL pitfalls (round() returns
real, toint() nulls on decimal strings): rules/coding-kql.md.
Auto refresh
"autoRefresh": {"enabled": true, "defaultInterval": "5m", "minInterval": "30s"}
— intervals from the enum 1s|10s|30s|1m|5m|15m|30m|1h|2h|1d. minInterval
caps what viewers may select.
Live refresh is a second object, changeDetection, which the portal writes
between autoRefresh and tiles:
{"kind": "liveUpdates", "minRefreshRate": "30s", "fallbackRefreshRate": "5m"},
or {"kind": "manualUpdates"}. By the v81 schema's descriptions the two rates
are Learn's Refresh rate limit and Fallback refresh interval, the latter
for visuals that cannot detect ingestion, each with its own enum. Learn turns
Live refresh on for new dashboards, so a generated file without the object
shows as drift at the next portal commit (2026-10-12).
Live JSON Schema
https://dataexplorer.azure.com/static/d/schema/{schema_version}/dashboard.json
plus sibling tile.json, query.json, baseQuery.json, parameter.json.
The $schema URL embedded in Microsoft's own sample files
(pbiadx.powerbi.com/static/d/schema/...) returns 404 — swap the host.
What does not exist
- No image/render REST API. The Power BI
exportToFileAPI does not cover KQL dashboards; the RTDB REST surface is item CRUD + the JSON definition, and the portal's "export" is the JSON file. Visual verification means a browser (screenshot or automation with an authenticated session); data-level verification means running the tile queries directly against the cluster. - No cross-tile dynamic references — a tile query sees base queries and parameter variables, nothing from other tiles.
Editing workflow
Round-trip options: Git sync (Update from Git), or portal Manage →
Replace with file for immediate feedback. Test tile queries before shipping
by composing them exactly as the dashboard does: prepend
let <variableName> = (<base query text>); for each entry in
usedVariables, then run against the database. Serialization is CRLF with
no final newline (JSON item part) — see rules/fabric-git-serialization.md.
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.

