Azure Skill: Detecting Flow Groups
Purpose: Authoritative rules for how an AI agent should group discovered integration artifacts into logical flow groups. The agent MUST follow these rules exactly.
1. Grouping Strategy (Shared Orchestration)
- HIGHEST PRIORITY: If multiple Receive Locations feed into the SAME Orchestration (directly or transitively), they MUST be in the SAME group — the orchestration is the unifying element.
- Orchestration Call Chains: The
connectionGraphincludesorchestration-callsedges when one orchestration calls/starts another. Orchestrations linked by these edges (directly or transitively) MUST be in the SAME group. Example: If Orch-A calls Orch-B and Orch-B calls Orch-C, all three belong in ONE group. - Only create separate groups for Receive Locations that feed into DIFFERENT orchestrations with no transitive connection.
- Transitively connected artifacts belong in the SAME group.
- Name each group by business purpose.
2. Fallback Rules (When receiveLocationCount = 0)
Do NOT return empty groups. Use this fallback order:
- Orchestration-root grouping — one group per CONNECTED orchestration cluster. Follow
orchestration-callsedges to find clusters. Do NOT create one group per individual orchestration when they call each other. - Trigger-capability grouping — file/queue/http/timer/db-poller entry-capable artifacts.
- Connected-component grouping — from the graph edges.
Only leave artifacts in ungroupedArtifactIds if they are truly isolated with no meaningful edges or entry semantics.
For fallback groups, set entryPoint to orchestration/trigger/internal-entry (not receive-location).
3. Required Output Fields
Each group MUST have:
| Field | Type | Description |
|---|---|---|
id |
string | Unique group ID (e.g. flow-1-order-processing) |
name |
string | Human-readable name by business purpose |
description |
string | What this flow does |
category |
string | e.g. message-flow, shared-infrastructure |
artifactIds |
string[] | MUST NOT be empty |
entryPoint |
object | Entry point (type, name, messageType) |
exitPoints |
array | Exit points (type, name, messageType) |
4. Procedure
- Call
migration_detectFlowGroupsto get the artifact connection graph and summaries. - If the connectionGraph has few or no
orchestration-callsedges, the call-chain data may be incomplete. In that case: a. Callmigration_listArtifactswithcategory="custom-code"to find DLLs. b. Decompile relevant DLLs per skilldependency-and-decompilation-analysis§2 to discover orchestration call relationships hidden in compiled code. c. Read orchestration source files (migration_readSourceFile) to findCall Orchestration/Start Orchestrationshapes and their targets. d. Use the discovered call chains to merge groups that should be together. - Determine logical flow groups using the shared-orchestration strategy above, incorporating both graph edges AND any call chains discovered in step 2.
- Call
migration_discovery_storeFlowGroupswith the groups array,ungroupedArtifactIds, and explanation.
5. What NOT to Do
- Do NOT create empty
artifactIdsarrays. - Do NOT split orchestrations linked by call chains into separate groups.
- Do NOT skip decompilation/source reading when the connectionGraph has missing call-chain edges — incomplete grouping causes downstream analysis failures.