Skip to content
CSA Loom — the Microsoft Fabric experience for Azure tenants where Fabric isn't yet available: lakehouses, warehouses, notebooks, semantic models, Activator rules, Data Agents, across Commercial, GCC, GCC-High, and DoD IL5

Copilot Studio Topic Editor — Fabric-parity spec

Captured 2026-05-26 by catalog agent from Microsoft Learn (Copilot Studio: authoring-create-edit-topics, authoring-triggers, authoring-variables, authoring-variables-about, authoring-system-topics, authoring-ask-a-question, advanced-entities-slot-filling, nlu-authoring, guidance/topic-authoring-best-practices, guidance/triggering-topics) + inspection of apps/fiab-console/lib/editors/copilot-studio-editors.tsx::CopilotTopicEditor and listTopics / upsertTopic in apps/fiab-console/lib/azure/copilot-studio-client.ts.

Overview

Topics are the conversation primitives in Copilot Studio (classic mode) and the structured fallback in generative mode. Each topic is a dialog flow: a directed graph of nodes that runs when a trigger fires. Triggers come from the NLU model (user utterance matched against trigger phrases), explicit Redirect calls from other topics, or system events (Conversation Start, Fallback, Escalate, On Error, custom event, inactivity). The orchestrator in generative mode selects topics dynamically using the topic name + description as routing hints. Topics support input/output variables for reuse across the agent.

Copilot Studio UX

Topics page chrome

  • Tabs: Custom topics · System topics
  • List columns: name · status (On/Off) · trigger phrases (preview) · last-modified · who-modified
  • Row actions: Open · Turn off · Copy · Delete · Export YAML
  • Top-right: Add a topic dropdown — Create from blank · Create from description with Copilot · Suggested topics (AI-generated from agent description)

Authoring canvas

  • Free-form canvas with draggable nodes connected by anchor lines
  • Trigger node at top (auto-created) — kind: Phrases (user utterance) · Event (Conversation Start / Inactivity / Custom event / Activity received) · On unknown intent (Fallback only)
  • Add node (+) opens a menu of node types (below)
  • Right-side Properties pane edits the selected node
  • Save + Test your agent chat panel (always uses latest unsaved draft)

Node types

Node Configuration
Send a message Rich-text editor with bold/italic/lists/hyperlinks · variable insertion ({x} picker) · adaptive cards · suggested replies (chips) · speech SSML override
Ask a question Question text · Identify: prebuilt entity (Number, Date and time, Person name, Email, Phone number, City, Country/region, Money, etc.) / closed list / regex / multiple-choice / dynamic / one-of-multiple / user's entire response · Options for user (multiple-choice buttons) · variable name · Include metadata (literal + value) · Additional entity validation (Power Fx) · No valid entity found behavior (Escalate / Set default / Set empty) · Interruption (allow switching to another topic — restrictable to specific topics) · Reprompt count
Add a condition Branch based on Power Fx expression over variables · Branch for all other conditions (else) · multiple branches
Call an action / Call a tool Pick Power Automate flow · prebuilt connector action · custom connector action · AI prompt (Foundry-deployed) · authentication node · Bot Framework skill
Redirect to another topic Pick destination topic · pass input variables to it · returns to current node when subtopic completes (unless redirecting to End/Escalate/Goodbye system topics, which end the conversation)
Variable management Set variable value · Parse value (string → record) · Clear variable values (specific / all / conversation history)
Show typing indicator Adaptive delay
End the conversation End with survey (CSAT) · Transfer to agent (Omnichannel handoff, optional private message) · Goodbye
Generative answers Knowledge sources subset · per-node knowledge override · suppress citations · prompt modification
Adaptive Card Author or paste card JSON; bind to variables

Variables

  • Scope: Topic (current topic) · Global (agent-level, persists across topics) · User (per-user, persists across sessions) · System (read-only: User.DisplayName, Conversation.Id, Conversation.LocalTimeZone, Activity.From.Id, UnrecognizedTriggerPhrase)
  • Data types: String · Number · Boolean · Date and time · Record · Table · Choice (enum) · File
  • Variable properties pane: name · type · Receive values from other topics (input parameter) · Send to other topics (output parameter) · default value · sensitive (PII flag)

Slot filling + entity extraction

  • When a Question node uses a prebuilt entity, the NLU model can pre-fill the answer from the initial utterance (e.g. user says "block my credit card" → CardType=Credit and Operation=Block are extracted before the Question runs)
  • Multiple-choice options auto-generate conditional branches per option

Trigger phrases panel

  • Inline list of user utterances (5–10 recommended) that should fire the topic
  • AI suggests more phrases based on existing ones (Suggest phrases)
  • Confidence / overlap warning if phrases collide with other topics
  • Description (orchestrator-routing hint in generative mode) — separate from trigger phrases

System topics (built-in, can't delete)

Conversation Start · Conversational boosting · End of Conversation · Escalate · Fallback · Multiple Topics Matched · On Error · Reset Conversation · Sign in · Start over · Thank you · Goodbye · Confirmed Success · Confirmed Failure

What Loom has today

From apps/fiab-console/lib/editors/copilot-studio-editors.tsx::CopilotTopicEditor and app/api/items/copilot-studio-topic/**: - Env picker + agent picker (shared with agent editor) - List topics (GET /msdyn_botcomponents?$filter=componenttype eq 9 and _msdyn_copilotid_value eq {agentId}) - Per-topic row: name · trigger-phrase count · modified-on - Edit topic drawer with two fields: trigger phrases (textarea, newline-separated) + flow YAML (Monaco editor, raw) - POST/PATCH to /msdyn_botcomponents writing data={triggerPhrases, flowYaml} + content=flowYaml - Delete topic - MessageBar for Copilot-Studio-not-enabled 503

Gaps for parity

  1. Visual canvas — Loom has raw YAML only; no node graph, no drag-and-drop, no node-add menu, no anchor-line editing
  2. Node-type forms — no typed Properties pane for Message · Question · Condition · Call action · Redirect · End · Generative answers · Adaptive Card · Variable management · Show typing
  3. Question node entity picker — no dropdown of prebuilt entities (Number, Date and time, Person name, etc.); no closed-list / regex / multiple-choice authoring
  4. Slot filling — Loom can't model Include metadata / Additional entity validation (Power Fx) / No valid entity found behavior
  5. Variables UI — no variable browser with scope/type/Receive from other topics/Send to other topics; no {x} insertion picker inside Message text
  6. Trigger configuration — Loom shows phrases only; no event-trigger kind (Conversation Start / Inactivity / Custom event / Activity received), no On unknown intent
  7. Description field — Loom doesn't expose the topic description used by the orchestrator for generative routing
  8. Trigger-phrase suggestions + overlap detection — no AI-suggest, no collision warning
  9. System-topics tab — Loom only lists custom topics (filtered by componenttype=9); system topics + their overrides not exposed
  10. Topic from description with Copilot — no natural-language topic generation entry point
  11. Suggested topics — no list of AI-generated topics seeded from agent description
  12. Per-topic Test panel — no in-context chat to test the topic in isolation
  13. Topic on/off toggle — statecode not surfaced; only delete
  14. Topic export/import — Loom has flowYaml in/out but no .zip export with related variables/skills
  15. Adaptive Card visual designer — no card-builder UI
  16. Generative answers node — no per-node knowledge-source picker
  17. Power Automate flow picker — Loom can't browse the env's flows from inside a Call-action node
  18. Redirect target picker — no dropdown of topics; raw YAML only
  19. Input/output variable binding on Redirect — not visualized
  20. Validation — no YAML schema-validation on save (broken topic flows currently save and fail at runtime)

Backend mapping

Dataverse Web API: - Topic CRUD — msdyn_botcomponents (id · name · componenttype=9 for topics · data JSON · content YAML · _msdyn_copilotid_value · modifiedon) - Topic flow YAML conforms to the Bot Framework Composer / Copilot Studio dialog schema (kind: OnUnknownIntent / OnConvUpdateActivity / etc.; actions: Microsoft.SendActivity / Microsoft.TextInput / Microsoft.IfCondition / Microsoft.BeginDialog / etc.) — but Copilot Studio uses its own simplified YAML schema documented at microsoft-copilot-studio/authoring-yaml (kind: SendActivity / Question / ConditionGroup / InvokeFlowAction / Redirect / EndDialog) - System topic overrides — msdyn_botcomponents with componenttype=9 + name matching the system-topic key (ConversationStart, Fallback, Escalate, etc.) - Variables — declared inline in the topic YAML (inputs:, outputs:, variables:); global variables live in msdyn_botvariables (separate table) - Trigger phrases — stored in topic YAML under triggers: section, and also indexed in msdyn_botcomponents.data.triggerPhrases for NLU model training - NLU model training — implicit; happens via bound action msdyn_TrainCopilot when topics change (Loom doesn't call this — manual publish triggers it)

Required Azure resources / tenant settings

  • All Agent-editor prerequisites
  • Dataverse search enabled (for NLU model training)
  • Copilot Studio Maker security role (or higher) on the SP for msdyn_botcomponents CRUD
  • For Call-action nodes: target Power Automate flow must exist in the env (msdyn_flow table) and the SP must have prvReadProcess privilege
  • For Generative answers nodes: Foundry-deployed model or default GPT-4o family must be set as the agent's model deployment
  • No new Azure resources beyond the agent's prerequisites

Estimated effort

5–6 sessions — this is the largest editor by far. Visual canvas (node graph + drag-drop + anchor lines + add-node menu) is ~2 sessions (React Flow or equivalent + custom node-renderer per node type). Typed Properties pane for the seven core node types (Message / Question / Condition / Call action / Redirect / End / Generative answers) is ~2 sessions. Variable browser + {x} insertion picker + trigger config (event triggers + system topics + description) is ~1 session. AI-assist (Copilot natural-language topic generation + suggested topics + trigger-phrase suggestions) is ~1 session. Adaptive Card visual designer can be a separate ~2-session follow-on if scope tight.