Fabric IQ items
Learn's IQ workload holds six items ("What is Fabric IQ?", read
2026-10-12). Four have a reference here, in the table below; Power BI
semantic models are pbi-items', and Plan has none. Each reference
holds the depth, and this file picks the item and carries only what
crosses items. Take the steps in order: the wrong item sends every later
step to the wrong file.
1. Pick the item
By the job asked for, or by the folder of the file in hand:
| The job |
Folder |
Read |
| A shared business vocabulary bound to data: entity types, properties, relationships |
<name>.Ontology/ |
ontology.md |
| GQL over a labeled property graph built from OneLake tables |
<name>.GraphModel/ |
graph.md |
| Natural-language Q&A over a domain's data, read-only, asked by a user |
<name>.DataAgent/ |
data-agent.md |
| Watching live data and acting on it, unattended |
<name>.OperationsAgent/ |
operations-agent.md |
operations-agent.md § 7 sets the two
agents side by side. Near these items, and not them:
- KQL's graph operators in an Eventhouse (
make-graph, graph-match):
rti-items, a different engine from the GraphModel.
- Rule-based alerting with no LLM:
rti-items' Activator.
- The semantic model an ontology is generated from, or a data agent
queries:
pbi-items.
2. What crosses items
- An ontology grounds both agents. A data agent takes one as a
source, with known issues of its own
(data-agent.md § What a Data
Agent is). An operations agent takes one as its single source, in the
agent's own workspace
(operations-agent-reference.md
§ Data source constraints). Model the ontology in ontology.md, and
configure each agent in its own file.
- An ontology can generate a graph: a child
GraphModel, made only
on opt-in through Manage graph, and only from eligible entity types
(ontology.md, its opening and § The
constraints that produce silent or confusing failures). GQL itself is
graph.md's.
- Two MCP endpoints, two shapes. An ontology is an MCP server
(ontology-mcp.md), and so
is a published data agent
(data-agent-mcp.md).
The URLs differ in shape, so neither is built from the other.
- Each item has its own tenant settings; check them first when an
item won't create. The ontology's are in
ontology.md § Before you start, and, with
both agents', in
ontology-reference.md § 7. The data
agent's are on the Learn page its § Reference links, and the
operations agent's in
operations-agent-reference.md
§ Runtime, identity and availability. graph.md names none.
3. Read the item's reference first
Read the file step 1 names before answering or editing: limits,
definition shapes and preview status live there, dated. Each sends its
long detail one file further:
ontology-reference.md,
graph-reference.md,
operations-agent-reference.md,
and the data agent's own references/
and assets/. A task that spans items, an
ontology-grounded agent for one, reads each item's file.
An item's definition, REST calls and MCP server have files of their own
where written. Read one first for its job; past it, the REST spec's
folder, then Learn, by microsoft-learn-mcp, or with no server, curl -sL -H 'Accept: text/markdown' <page URL> (undocumented, read 2026-11-07):
| The job |
Read first |
| A data agent's definition parts, or a REST call past the generic item set |
data-agent-api.md |
| A data agent's MCP server: endpoint, token, tool discovery |
data-agent-mcp.md |
| An ontology's definition parts (TMDL, and the old JSON layout), or its REST calls (the generic seven only) |
ontology-reference.md § 1 |
| An ontology's MCP server: endpoint, prerequisites, tool names |
ontology-mcp.md |
A graph model's definition parts, or a REST call past the generic item set (executeQuery, getQueryableGraphType, the refresh job) |
graph-model-api.md |
A graph queryset's definition (exportedDefinition.json live, named by neither Learn nor the spec) or its REST calls (the generic seven only) |
graph-queryset-api.md |
An operations agent's definition (Configurations.json) or its REST calls (the generic seven only) |
operations-agent.md |
| An operations agent's MCP server: its configure and query endpoints, token, tool discovery |
operations-agent-mcp.md |
A live page that contradicts a line here wins: say so, act on the page,
and route the fix by capture-learning, bar a dated measured line that
names the page it overrules and the call that measured it, which a page
changed since its date sends back to re-measure. An observed claim is cited
or marked pending a doc check, and stands, marked observed, only where no
page covers it.
The data agent has two scripts, each run by uv run and its docstring
its usage: ${CLAUDE_SKILL_DIR}/scripts/data-agent-eval.py grades the
agent's replies against its semantic model's figures, and
${CLAUDE_SKILL_DIR}/scripts/data-agent-settings.py reads and sets the
draft's instructions and runtime, and publishes it
(evaluation.md).
1---2name: iq-items3description: "Use for Microsoft Fabric IQ items, in Git or by REST, `fab` or a script: an ontology (`.Ontology`, `/ontologies`: entity types, properties, relationships, data binding, semantic enrichment, its MCP server), a graph model or graph queryset (`.GraphModel`, `/graphModels`, `.GraphQuerySet`, `/graphQuerySets`: GQL over a labeled property graph), a data agent (`.DataAgent`, `/dataAgents`: natural-language Q&A over Fabric data, its MCP server, evaluation) or an operations agent (`.OperationsAgent`, `/operationsAgents`, `Configurations.json`: watches live data and acts on it). Picks the item, then reads its reference."4when_to_use: "Use before a Read, edit or REST call on an item of type Ontology, GraphModel, GraphQuerySet, DataAgent or OperationsAgent, or before writing GQL for a Fabric graph model."5# model: inherit # any model: value blocks Copilot slash invocation6# effort: medium # unset = inherit session effort; there is no 'effort: inherit'7disable-model-invocation: false8metadata:9 source: https://skillmd.com/skills/wardawgmalvicious/iq-items10---1112# Fabric IQ items1314Learn's IQ workload holds six items ("What is Fabric IQ?", read152026-10-12). Four have a reference here, in the table below; Power BI16semantic models are `pbi-items`', and Plan has none. Each reference17holds the depth, and this file picks the item and carries only what18crosses items. Take the steps in order: the wrong item sends every later19step to the wrong file.2021## 1. Pick the item2223By the job asked for, or by the folder of the file in hand:2425| The job | Folder | Read |26| --- | --- | --- |27| A shared business vocabulary bound to data: entity types, properties, relationships | `<name>.Ontology/` | [ontology.md](references/ontology.md) |28| GQL over a labeled property graph built from OneLake tables | `<name>.GraphModel/` | [graph.md](references/graph.md) |29| Natural-language Q&A over a domain's data, read-only, asked by a user | `<name>.DataAgent/` | [data-agent.md](references/data-agent/data-agent.md) |30| Watching live data and acting on it, unattended | `<name>.OperationsAgent/` | [operations-agent.md](references/operations-agent.md) |3132[operations-agent.md](references/operations-agent.md) § 7 sets the two33agents side by side. Near these items, and not them:3435- KQL's graph operators in an Eventhouse (`make-graph`, `graph-match`):36 `rti-items`, a different engine from the GraphModel.37- Rule-based alerting with no LLM: `rti-items`' Activator.38- The semantic model an ontology is generated from, or a data agent39 queries: `pbi-items`.4041## 2. What crosses items4243- **An ontology grounds both agents.** A data agent takes one as a44 source, with known issues of its own45 ([data-agent.md](references/data-agent/data-agent.md) § What a Data46 Agent is). An operations agent takes one as its single source, in the47 agent's own workspace48 ([operations-agent-reference.md](references/operations-agent-reference.md)49 § Data source constraints). Model the ontology in ontology.md, and50 configure each agent in its own file.51- **An ontology can generate a graph**: a child `GraphModel`, made only52 on opt-in through *Manage graph*, and only from eligible entity types53 ([ontology.md](references/ontology.md), its opening and § The54 constraints that produce silent or confusing failures). GQL itself is55 [graph.md](references/graph.md)'s.56- **Two MCP endpoints, two shapes.** An ontology is an MCP server57 ([ontology-mcp.md](references/ontology-mcp.md)), and so58 is a published data agent59 ([data-agent-mcp.md](references/data-agent/references/data-agent-mcp.md)).60 The URLs differ in shape, so neither is built from the other.61- **Each item has its own tenant settings**; check them first when an62 item won't create. The ontology's are in63 [ontology.md](references/ontology.md) § Before you start, and, with64 both agents', in65 [ontology-reference.md](references/ontology-reference.md) § 7. The data66 agent's are on the Learn page its § Reference links, and the67 operations agent's in68 [operations-agent-reference.md](references/operations-agent-reference.md)69 § Runtime, identity and availability. graph.md names none.7071## 3. Read the item's reference first7273Read the file step 1 names before answering or editing: limits,74definition shapes and preview status live there, dated. Each sends its75long detail one file further:76[ontology-reference.md](references/ontology-reference.md),77[graph-reference.md](references/graph-reference.md),78[operations-agent-reference.md](references/operations-agent-reference.md),79and the data agent's own [references/](references/data-agent/references/)80and [assets/](references/data-agent/assets/). A task that spans items, an81ontology-grounded agent for one, reads each item's file.8283An item's definition, REST calls and MCP server have files of their own84where written. Read one first for its job; past it, the REST spec's85folder, then Learn, by `microsoft-learn-mcp`, or with no server, `curl -sL -H 'Accept: text/markdown' <page URL>` (undocumented, read 2026-11-07):8687| The job | Read first |88| --- | --- |89| A data agent's definition parts, or a REST call past the generic item set | [data-agent-api.md](references/data-agent/references/data-agent-api.md) |90| A data agent's MCP server: endpoint, token, tool discovery | [data-agent-mcp.md](references/data-agent/references/data-agent-mcp.md) |91| An ontology's definition parts (TMDL, and the old JSON layout), or its REST calls (the generic seven only) | [ontology-reference.md](references/ontology-reference.md) § 1 |92| An ontology's MCP server: endpoint, prerequisites, tool names | [ontology-mcp.md](references/ontology-mcp.md) |93| A graph model's definition parts, or a REST call past the generic item set (`executeQuery`, `getQueryableGraphType`, the refresh job) | [graph-model-api.md](references/graph-model-api.md) |94| A graph queryset's definition (`exportedDefinition.json` live, named by neither Learn nor the spec) or its REST calls (the generic seven only) | [graph-queryset-api.md](references/graph-queryset-api.md) |95| An operations agent's definition (`Configurations.json`) or its REST calls (the generic seven only) | [operations-agent.md](references/operations-agent.md) |96| An operations agent's MCP server: its configure and query endpoints, token, tool discovery | [operations-agent-mcp.md](references/operations-agent-mcp.md) |9798A live page that contradicts a line here wins: say so, act on the page,99and route the fix by `capture-learning`, bar a dated measured line that100names the page it overrules and the call that measured it, which a page101changed since its date sends back to re-measure. An observed claim is cited102or marked pending a doc check, and stands, marked observed, only where no103page covers it.104105The data agent has two scripts, each run by `uv run` and its docstring106its usage: `${CLAUDE_SKILL_DIR}/scripts/data-agent-eval.py` grades the107agent's replies against its semantic model's figures, and108`${CLAUDE_SKILL_DIR}/scripts/data-agent-settings.py` reads and sets the109draft's instructions and runtime, and publishes it110([evaluation.md](references/data-agent/references/evaluation.md)).