Azure Kusto Irql Graph

IRQL Graph Functions – Query Results to Visualization

Apply the IRQL graph function family to tabular results. Given a KQL or IRQL query and the user’s graph description, generate a Lift_To_Graph mapping and compose only the stored graph functions needed to visualize, fold, extract, or enrich the graph in Kusto Explorer. The source query does not need to use IRQL.

Scope and Routing

If a request mixes visualization and native graph analysis, use this skill for the lift/render portion and azure-kusto-graph for operator semantics. Do not replace graph-lift functions with a hand-built edges-first graph unless the user asks for native graph operators.

Input Contract

• Preferred input: a working KQL/IRQL query that produces tabular results, plus a natural-language description of the desired nodes, edges, labels, icons, extracts, enrichments, or folds.
• This skill is not a natural-language-to-KQL or NL-to-IRQL converter. It transforms existing query results into graph visualizations. For general NL-to-KQL or NL-to-IRQL conversion, use a dedicated query-generation skill (available separately).
• Preserve the supplied query’s retrieval, joins, filters, and aggregations. Add only projections or synthetic IDs required by the graph mapping.
• A basic natural-language source request is supported only when it maps directly to one known table or IRQL Get_* selector with obvious columns and simple filters. State the assumed source, and do not invent joins, schema, or investigation logic.
• For non-trivial query construction, use a separate Kusto/IRQL query-generation skill first, then apply this skill to its output.
• If no query or output schema is available and the source is not trivial, request the KQL query or its result columns before generating a mapping.

Activation Triggers

Use this skill when the user:

• Supplies KQL/IRQL results and asks for an IRQL graph visualization or mapping
• Mentions Lift_To_Graph, Graph_Render_View, or Graph_Fold_By_Property
• Asks for icon-decorated node/edge mappings in Kusto Explorer
• Wants to fold/collapse nodes by a shared property
• Requests graph extraction or enrichment through Extract_Node_*, Enrich_Node_*, or Enrich_Graph_*

Do not activate this skill solely for graph-match, graph paths/components, persistent graphs, or generic make-graph construction; those belong to azure-kusto-graph.

Not a natural-language-to-KQL/IRQL converter. The input should generally be a working KQL or IRQL query whose results need graph visualization. Basic NL source requests work only for trivial single-table/selector cases. For general NL-to-KQL or NL-to-IRQL, use a dedicated query-generation skill (available separately).

Environment

• Cluster: https://kc7001.eastus.kusto.windows.net
• Databases: ValdyTimes, JoJosHospital (graph functions pre-deployed)
• Rendering: Kusto Explorer desktop app (make-graph visualization window)
• Tool: kusto_query (via Azure MCP Server)

Function Preflight

Lift_To_Graph and Graph_Render_View are stored functions, not built-in Kusto operators. Before generating or running a lift pipeline against a target database, check what is deployed:

.show functions
| where Name in~ ("Lift_To_Graph", "Graph_Render_View", "Graph_Fold_By_Property")
| project Name

• Lift_To_Graph and Graph_Render_View are required.
• Graph_Fold_By_Property is required only when folding is requested.
• Check any Extract_Node_*, Enrich_Node_*, or Enrich_Graph_* function before using it; omit optional enrichment when unavailable unless the user wants it deployed.
• If a required function is missing and you have permission to alter the database, ask the user for confirmation before deploying. Then use the .create-or-alter function definitions in references/DEPLOY_IRQL_FUNCTIONS.md (opens in a new tab). Run the relevant .create-or-alter block, then rerun the preflight check to confirm.
• If you do not have alter permissions, tell the user which functions are missing and point them to references/DEPLOY_IRQL_FUNCTIONS.md for manual deployment.

IRQL Graph Function Family

Lift_To_Graph(T, mappingJson)

Transforms any tabular KQL result into a unified node + edge table.

Input: Any table T + a JSON mapping string. Output: Rows with EntityType = "node" or "edge", ready for make-graph.

Graph_Render_View(T)

Takes Lift_To_Graph output, splits nodes/edges, and calls make-graph to open Kusto Explorer’s graph window.

Graph_Fold_By_Property(T, NodeType, PropertyName)

Collapses nodes of a given type sharing a property value into a single node. Rewires edges automatically.

Graph Extraction and Enrichment Functions

These are additional stored functions that must already be deployed on the target database. They are not bundled in references/DEPLOY_IRQL_FUNCTIONS.md. Use .show functions to verify availability before including in a pipeline.

Mapping JSON Schema

The JSON mapping has two arrays: node_types and edges.

node_types[]

edges[]

Icon Repository

Use icons from https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/:

• IP: Public-IP-Addresses-(Classic).svg
• Host/VM: Virtual-Machine.svg
• User: Users.svg
• Email: Mailbox.svg (or azure-cds/command-1070-Mail.svg)
• Process: App-Services.svg
• File: Storage-Accounts.svg
• Alert: Activity-Log.svg
• Domain: DNS-Zones.svg

Mapping Generation Rules

Given the supplied query columns and the user’s graph description, generate the mapping JSON by:

1. Identify entities -> each distinct noun becomes a node_type
2. Identify relationships -> each verb/preposition becomes an edge
3. Map to columns -> use actual columns produced by the supplied query; never assume unavailable columns
4. Set direction -> source is the actor, target is the acted-upon
5. Add properties -> include columns relevant to investigation (timestamps, results, hashes)
6. Assign icons -> pick from the icon set above based on entity type

Column Reference (IRQL unified schema)

Function Selection

1. Start with the supplied KQL/IRQL tabular pipeline.
2. Use Lift_To_Graph(mapping) to create graph entities.
3. Add Extract_Node_*, Enrich_Node_*, or Enrich_Graph_* only when requested and compatible with the mapped keys.
4. Add Graph_Fold_By_Property() only when grouping/collapse is requested.
5. End visual output with Graph_Render_View().
6. Preflight the exact stored functions selected for the pipeline.

Pipeline Pattern

// 1. Preserve the supplied KQL or IRQL query
<input query>
// 2. Lift to graph
| invoke Lift_To_Graph(<mapping_json>)
// 3. Optionally extract or enrich graph entities
| invoke <Extract_Node_* | Enrich_Node_* | Enrich_Graph_*>()
// 4. Optionally fold nodes when requested
| invoke Graph_Fold_By_Property("<NodeType>", "<PropertyName>")
// 5. Render
| invoke Graph_Render_View()

Examples

For additional prompts and worked examples, see references/EXAMPLES.md (opens in a new tab).

Authentication graph: IP -> AuthEvent -> User -> Host

Input query: Get_Event_Authentication_All | where Result == "Failed Login" | take 200

Graph request: “Show IPs, authentication events, users, and hosts; fold events by result.”

let auth_mapping = '{"node_types":[{"type":"SrcIp","id":"SrcIp","key":"ClientIp","props":["ClientIp"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/Public-IP-Addresses-(Classic).svg"},{"type":"Host","id":"Host","key":"Hostname","props":["Hostname"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/Virtual-Machine.svg"},{"type":"User","id":"User","key":"Username","props":["Username"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/Users.svg"},{"type":"AuthEvent","id":"AuthEvent","key":"AuthEventId","props":["AuthEventId","EnvTime","UserAgent","Result","Description"],"defaults":{"Result":"unknown"},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/Activity-Log.svg"}],"edges":[{"type":"RequestsAuth","source":{"id":"SrcIp","type":"SrcIp"},"target":{"id":"AuthEvent","type":"AuthEvent"},"props":["EnvTime"]},{"type":"TargetsUser","source":{"id":"AuthEvent","type":"AuthEvent"},"target":{"id":"User","type":"User"},"props":["EnvTime"]},{"type":"AgainstHost","source":{"id":"AuthEvent","type":"AuthEvent"},"target":{"id":"Host","type":"Host"},"props":["EnvTime"]}]}';
Get_Event_Authentication_All
| extend AuthEventId = strcat(Username, "_", Hostname, "_", EnvTime)
| where Result == "Failed Login"
| take 200
| invoke Lift_To_Graph(auth_mapping)
| invoke Graph_Fold_By_Property("AuthEvent", "Result")
| invoke Graph_Render_View()

Email graph: Sender -> Message -> Recipient

Input query: Get_Email_All | take 400

Graph request: “Visualize sender-to-message-to-recipient flow and fold messages by verdict.”

let mail_mapping = '{"node_types":[{"type":"EmailMessage","id":"Message","key":"Subject","props":["EnvTime","Subject","Verdict"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/Media-File.svg"},{"type":"Sender","id":"Email","key":"EmailSender","props":["EmailSender"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-cds/command-1070-Mail.svg"},{"type":"Recipient","id":"Email","key":"EmailRecipient","props":["EmailRecipient"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-cds/command-1070-Mail.svg"}],"edges":[{"type":"SentBy","source":{"id":"Message","type":"EmailMessage"},"target":{"id":"Email","type":"Sender"},"props":["EnvTime","Verdict"]},{"type":"DeliveredTo","source":{"id":"Message","type":"EmailMessage"},"target":{"id":"Email","type":"Recipient"},"props":["EnvTime","Verdict"]}]}';
Get_Email_All
| take 400
| invoke Lift_To_Graph(mail_mapping)
| invoke Graph_Fold_By_Property("EmailMessage", "Verdict")
| invoke Graph_Render_View()

Suspicious domain investigation (end-to-end)

Basic source request: “Use outbound network events for these suspicious domains and graph IP-to-domain connections enriched with employee names.”

This is the limited fallback: one known selector, one extractor, and one direct filter.

let suspicious_domain_mapping = '{"node_types":[{"type":"IP","id":"IP","key":"ClientIp","props":["ClientIp"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/Public-IP-Addresses-(Classic).svg"},{"type":"Domain","id":"Domain","key":"DomainName","props":["DomainName"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/DNS-Zones.svg"}],"edges":[{"type":"ConnectsTo","source":{"id":"IP","type":"IP"},"target":{"id":"Domain","type":"Domain"},"props":["EnvTime"]}]}';
Get_Event_NetworkOutbound
| invoke Extract_Event_Network_Domain()
| where DomainName has_any ("raisinkanes.com", "nothing-to-see-here.net", "totally-legit-domain.com")
| invoke Lift_To_Graph(suspicious_domain_mapping)
| invoke Enrich_Node_Ip_Employee("Name")
| invoke Graph_Fold_By_Property("Domain", "DomainName")
| invoke Graph_Render_View()

Process execution graph: User -> Process -> ParentProcess

Input query: Get_Event_Process_All | where ProcessCommandLine has "powershell" | take 300

Graph request: “Visualize process, parent process, host, and user relationships.”

let proc_mapping = '{"node_types":[{"type":"Process","id":"Proc","key":"ProcessName","props":["ProcessName","ProcessCommandLine","ProcessHash"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/App-Services.svg"},{"type":"ParentProcess","id":"Proc","key":"ParentProcessName","props":["ParentProcessName","ParentProcessHash"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/App-Services.svg"},{"type":"Host","id":"Host","key":"Hostname","props":["Hostname"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/Virtual-Machine.svg"},{"type":"User","id":"User","key":"Username","props":["Username"],"defaults":{},"defIcon":"https://raw.githubusercontent.com/benc-uk/icon-collection/master/azure-icons/Users.svg"}],"edges":[{"type":"SpawnedBy","source":{"id":"Proc","type":"Process"},"target":{"id":"Proc","type":"ParentProcess"},"props":["EnvTime"]},{"type":"RanOn","source":{"id":"Proc","type":"Process"},"target":{"id":"Host","type":"Host"},"props":["EnvTime"]},{"type":"ExecutedBy","source":{"id":"Proc","type":"Process"},"target":{"id":"User","type":"User"},"props":["EnvTime"]}]}';
Get_Event_Process_All
| where ProcessCommandLine has "powershell"
| take 300
| invoke Lift_To_Graph(proc_mapping)
| invoke Graph_Render_View()

Query Results -> Mapping Translation

When the user supplies a query and describes the graph:

1. Inspect the query’s final output columns
2. Parse the entity nouns and relationship verbs
3. Generate the mapping JSON using only those columns
4. Preserve the supplied pipeline and append Lift_To_Graph()
5. Include Graph_Render_View() at the end
6. If the user mentions grouping/collapsing and the function exists, add Graph_Fold_By_Property()

Output the complete KQL – the supplied query plus mapping JSON inline as a string let binding – after the required-function preflight passes. Clearly mark unverified function dependencies when the target database cannot be checked.

Opening Queries in Kusto Explorer (Windows Only)

Optional convenience feature. The default workflow is to output the KQL in chat and let the user copy it into Kusto Explorer or the VS Code Kusto extension manually. Auto-launch is opt-in only.

Always output the complete KQL query in the chat response with Step 1 (connect) and Step 2 (query) clearly labeled:

// Step 1: Connect to your cluster (skip if already connected)
// Example: uncomment to connect to the KC7 training cluster
// #connect cluster('kc7001.eastus.kusto.windows.net').database('ValdyTimes')
// Or replace with your own cluster:
// #connect cluster('<YOUR_CLUSTER>').database('<YOUR_DATABASE>')

// Step 2: Run the query below
<KQL_QUERY>

If the user asks to save or open in Kusto Explorer, follow the procedure in references/KUSTO_EXPLORER_LAUNCH.md (opens in a new tab). Key rules:

• Use ask_user to confirm before writing files or launching executables
• Display file contents in chat so the user can review before opening
• Never use shell interpolation or here-strings — write files via Set-Content/Add-Content
• Never encode queries into browser URLs
• On macOS/Linux, save the .kql file and suggest the VS Code Kusto extension or ADX Web Explorer