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 Action Editor — Fabric-parity spec

Captured 2026-05-26 by catalog agent from Microsoft Learn (Copilot Studio: advanced-flow, advanced-flow-input-output, advanced-connectors, authoring-connections, add-tools-custom-agent, guidance/agent-tools, nlu-prompt-node, advanced-generative-actions, advanced-use-skills) + inspection of apps/fiab-console/lib/editors/copilot-studio-editors.tsx::CopilotActionEditor and listActions / bindAction in apps/fiab-console/lib/azure/copilot-studio-client.ts.

Overview

Actions (also called Tools in current Copilot Studio terminology) let an agent take action on external systems — call REST APIs, run business logic, invoke AI prompts, query data. They're the agent's hands. Each action declares typed inputs (collected from the conversation) and outputs (returned to the agent for use in subsequent nodes). The orchestrator in generative mode picks tools dynamically based on the user's intent and the tool's description; in classic mode, topics explicitly call tools via a Call an action node. Tools come in five flavors: Power Automate flow, prebuilt connector action, custom connector action, AI prompt, and Bot Framework skill.

Copilot Studio UX

Tools page chrome

  • Left rail entry: Tools (formerly Actions / Plugins)
  • List columns: name · type icon · description · used-by-N-topics · last-modified
  • Row actions: Edit · Test · Delete · Enable/Disable · View runs
  • Top-right: Add a tool dropdown — New action (Power Automate flow) · Prebuilt connector · Custom connector · Prompt (Foundry/GPT) · Skill (Bot Framework manifest URL) · Conversational tool (from another agent)

Tool authoring

Power Automate flow tool

  • Pick existing flow from env, or Create new → opens Power Automate authoring canvas pre-seeded with a "When Power Virtual Agents calls a flow" trigger
  • Inputs — typed parameters declared on the flow trigger (Number, String, Boolean — list/object/date types NOT supported as inputs from the agent)
  • Outputs — typed return values from the "Return value(s) to Power Virtual Agents" action at the end of the flow
  • Description — used by the orchestrator to decide when to call this tool (mandatory for generative mode)
  • Authentication — flow runs as the maker (default) or as the end user (per-user connections)

Prebuilt connector action

  • Browse ~1500 Power Platform connectors (Salesforce · ServiceNow · Jira · SharePoint · Outlook · Teams · Dataverse · Microsoft Graph · OpenAI · etc.)
  • Pick an operation (e.g. SharePoint → Create item, Outlook → Send email)
  • Connection reference — bind to an env-scoped connection (which authenticates with shared creds or per-user OBO)
  • Map agent variables to connector operation parameters (typed)
  • Map connector response fields to agent output variables
  • Description — orchestrator routing hint

Custom connector action

  • Pick a custom connector defined in the env (Microsoft.PowerApps/connectors)
  • Same flow as prebuilt connector
  • Custom connectors are defined separately in Power Apps maker portal (OpenAPI spec or "from blank" with HTTP triggers)
  • Auth modes: OAuth 2.0 (Entra preferred) · API key · Basic · No auth · Windows

AI prompt tool

  • Author a prompt with typed inputs (variables) referenced as {{var}}
  • Pick Model — GPT-4o / GPT-4o-mini / Foundry-deployed model
  • Optional Grounding sources: Dataverse table · uploaded file · image input
  • Optional Code interpreter for structured data
  • Output format: text · JSON (with declared schema)
  • Temperature · Max tokens
  • Tested in Prompt Builder before binding

Bot Framework skill

  • Manifest URL — points to a Bot Framework skill manifest (V3) describing actions + activities
  • Skill registration tokens are managed by Azure Bot Service
  • Agent calls skill actions like sub-conversations

Connections page (linked from Tools)

  • List all env connections used by Tools
  • Per connection: connector name · auth mode · owner · status (Connected · Expired · Failed)
  • Test connection · Reconnect · Delete

Tool runs / activity

  • Per-tool runs list with status (Success · Failed · Pending) · inputs · outputs · timestamp · cost (for AI prompts)
  • Drill-down to per-run trace (Power Automate run details / connector call response / prompt completion text)

What Loom has today

From apps/fiab-console/lib/editors/copilot-studio-editors.tsx::CopilotActionEditor and app/api/items/copilot-studio-action/**: - Env picker + agent picker (shared) - List actions (GET /msdyn_bot_actions?$filter=_msdyn_copilotid_value eq {agentId}) - Per-action row: name · type · connectorId · flowId · enabled (statecode==0) - Bind action dialog with a type dropdown (power-automate-flow | custom-connector | prebuilt) + free-text name, connectorId, flowId - POST to /msdyn_bot_actions with msdyn_copilotid@odata.bind - Delete action - MessageBar for Copilot-Studio-not-enabled 503

Gaps for parity

  1. Tool-type coverage — Loom supports the metadata for flow / custom-connector / prebuilt, but is missing AI prompt and Bot Framework skill tool types
  2. Flow picker — Loom takes a free-text flowId; no list of env-scoped Power Automate flows with name/owner/last-run
  3. Create new flow — no entry point to open Power Automate authoring canvas pre-seeded with the PVA trigger
  4. Connector browser — no UI to browse the 1500+ connectors with category filters + auth-mode info + premium-badge surfacing
  5. Connector operation picker — no operation list per connector (e.g. SharePoint → Create item, Get items, Update item)
  6. Typed input/output mapping — Loom binds the action shell only; the per-input variable mapping (Number | String | Boolean) is not authored, and no validation against the connector's OpenAPI schema
  7. Description field — msdyn_description not exposed; orchestrator generative-routing breaks without it
  8. Connection reference picker — Loom can't pick or create env connections; no connectionreferences table integration
  9. AI prompt authoring — no Prompt Builder embed for model + temperature + grounding sources + output schema
  10. Bot Framework skill manifest URL — not exposed
  11. Tool runs / activity feed — no per-run inputs/outputs/status drill-down
  12. Test action — Loom has no way to fire a tool with synthetic inputs and inspect the output
  13. Enabled/Disabled toggle — statecode is read-only in Loom (mapped to enabled boolean but not writable)
  14. Used-by-N-topics — Loom doesn't compute the reverse lookup (which topics call this tool)
  15. Per-user vs shared auth — connection auth mode not exposed
  16. DLP impact — env DLP policy may block a connector; Loom doesn't surface why a tool fails to bind
  17. Custom-connector OpenAPI import — no embed of the custom-connector authoring flow (defined in Power Apps maker portal)
  18. List/object/date input parameters — Microsoft Learn explicitly calls out these are NOT supported as flow inputs from the agent; Loom should validate and block

Backend mapping

Dataverse Web API: - Action CRUD — msdyn_bot_actions (id · name · type · msdyn_connectorid · msdyn_flowid · _msdyn_copilotid_value · statecode · description · inputs JSON · outputs JSON) - Tools/Plugins (new framework) — recent Copilot Studio versions are migrating to msdyn_plugin + msdyn_pluginaction tables (declarative-agent tool framework); Loom currently targets the older msdyn_bot_actions table — for forward parity, also support msdyn_plugin rows - Power Automate flow lookup — workflow table (Dataverse) category=5 (modern flow); flow's PVA trigger is identified by clientdata JSON containing "connectorName":"shared_powervirtualagents" - Custom connector lookup — connector table (Dataverse legacy) or Microsoft.PowerApps/connectors ARM - Prebuilt connector metadata — Power Platform REST https://api.powerapps.com/providers/Microsoft.PowerApps/apis?api-version=2020-06-01 returns the full connector list + per-connector OpenAPI - Connections — connectionreferences table (Dataverse); each tool binds via msdyn_connectionreferenceid - AI prompts — msdyn_aimodel + msdyn_aimodelschema tables; prompt versions in msdyn_aigeneration - Skills — botcomponent rows of skill kind referencing manifest URL - Run history — Power Automate run records in flowrun table (env-scoped) + cross-ref to bot conversation log - Train + publish — msdyn_PublishCopilot bound action regenerates the orchestrator's tool registry from msdyn_bot_actions / msdyn_plugin rows

Required Azure resources / tenant settings

  • All Agent-editor prerequisites
  • Power Automate license for the SP/env (Per User / Per Flow / included with Copilot Studio license)
  • Premium connector licensing for connectors flagged premium (Salesforce / ServiceNow / SQL / etc.) — Power Apps per-user or per-app plan
  • Connection references for each connector in scope — SP needs prvReadconnectionreference + prvWriteconnectionreference
  • For AI prompts: Azure OpenAI / Foundry model deployment with quota; alternatively the default GPT-4o family included with Copilot Studio messages
  • For Bot Framework skills: Azure Bot Service resource (Microsoft.BotService/botServices) registered as the skill endpoint
  • DLP policy review — env DLP policy must permit the connectors the agent will use

Estimated effort

3–4 sessions. Flow picker + connector browser + connector operation list + typed input/output mapping is ~1.5 sessions (Power Platform REST integration + OpenAPI rendering). AI prompt builder embed + description field + enabled/disabled toggle + used-by lookup is ~1 session. Tool runs / activity feed + Test-action panel + DLP impact viewer is ~1 session. Bot Framework skill + custom-connector OpenAPI import + msdyn_plugin forward-compat layer can fold into the third session, or carry into a follow-on if scope tight.