Edit Action

Edit an existing action (kind: TaskDialog) in a Copilot Studio agent. Supports both regular connector actions (InvokeConnectorTaskAction) and MCP server actions (InvokeExternalAgentTaskAction). Uses connector definitions to understand the full operation schema (inputs, outputs, types) and action templates as structural reference.

Connector Lookup

Help the user find the right connector and operation before they go to the UI. Use the connector lookup tool. Important: The connector lookup script only covers a subset of connectors (run list to see which ones). If the user's requested connector is not in the list, tell the user they need to find and add the connector action entirely through the Copilot Studio portal, then ask them to pull again the files locally. Once pulled, /copilot-studio:edit-action can still be used to customize the YAML.

When the connector IS available, use the lookup tool to help the user before they go to the UI:

node ${CLAUDE_SKILL_DIR}/../../scripts/connector-lookup.bundle.js list                              # List connectors
node ${CLAUDE_SKILL_DIR}/../../scripts/connector-lookup.bundle.js operations <connector>            # List operations
node ${CLAUDE_SKILL_DIR}/../../scripts/connector-lookup.bundle.js operation <connector> <opId>      # Full input/output details
node ${CLAUDE_SKILL_DIR}/../../scripts/connector-lookup.bundle.js search <keyword>                  # Search operations

Use schema lookup for structural properties:

node ${CLAUDE_SKILL_DIR}/../../scripts/schema-lookup.bundle.js summary TaskDialog
node ${CLAUDE_SKILL_DIR}/../../scripts/schema-lookup.bundle.js summary InvokeConnectorTaskAction
node ${CLAUDE_SKILL_DIR}/../../scripts/schema-lookup.bundle.js summary InvokeExternalAgentTaskAction
node ${CLAUDE_SKILL_DIR}/../../scripts/schema-lookup.bundle.js summary ModelContextProtocolMetadata

Instructions

1. Auto-discover the agent directory:

   Glob: **/agent.mcs.yml

   If multiple agents found, ask which one.
2. Find the action file to edit:

   Glob: <agent-dir>/actions/*.mcs.yml

   Read the action file. If the user's request is ambiguous and multiple actions exist, list them and ask which one.
3. Identify the connector and operation:
   • Read the action YAML to find action.operationId and the connector API name from connectionreferences.mcs.yml
   • Look up the full operation details (if the connector is available in the lookup script):

     node ${CLAUDE_SKILL_DIR}/../../scripts/connector-lookup.bundle.js operation <connector> <operationId>

   • This gives you the complete list of available inputs and outputs for the operation
   • If the connector is not found, try broader terms. If still not found, inform the user and proceed with edits based on the existing action YAML and schema-lookup only
   • If the connector is SharePoint (shared_sharepointonline), read the SharePoint-specific reference before making any edits:

     Read: ${CLAUDE_SKILL_DIR}/sharepoint-actions.md

     This covers OData filter syntax, the critical "'$filter'" quoting pattern, which inputs should be Manual vs Automatic, and dynamic output handling.
4. Determine the action type from the YAML:
   • If action.kind is InvokeConnectorTaskAction → regular connector action
   • If action.kind is InvokeExternalAgentTaskAction → MCP server action
5. Read the appropriate template for structural reference:
   • Connector actions:

     Read: ${CLAUDE_SKILL_DIR}/../../templates/actions/connector-action.mcs.yml

   • MCP actions:

     Read: ${CLAUDE_SKILL_DIR}/../../templates/actions/mcp-action.mcs.yml

   Use this alongside the connector-lookup output from step 3 to understand the YAML structure and available inputs/outputs.

6. Make the requested edits using the Edit tool. Common modifications:

   Modify Input Descriptions

   Update description on AutomaticTaskInput entries to improve how the orchestrator fills them:

   inputs:
     - kind: AutomaticTaskInput
       propertyName: body
       description: The email body content in HTML format   # <-- edit this

   Add or Remove Inputs

   Cross-reference the connector definition to see all available input properties. Add new ones or remove unnecessary ones:

   # Add a new input from the connector definition
   - kind: AutomaticTaskInput
     propertyName: importance
     description: The importance level of the email (Low, Normal, High)

   Switch Between AutomaticTaskInput and ManualTaskInput

   • AutomaticTaskInput: The orchestrator fills this dynamically based on the conversation
   • ManualTaskInput: A hardcoded value (string only)

   # Change from automatic to manual (hardcode a value)
   - kind: ManualTaskInput
     propertyName: timeZone
     value: "(UTC+01:00) Amsterdam, Berlin, Bern, Rome, Stockholm, Vienna"

   Modify Outputs

   Add or remove output properties. Cross-reference the connector definition for available outputs:

   outputs:
     - propertyName: id
       description: The unique ID of the created item

   Change modelDisplayName / modelDescription

   These control how the orchestrator understands and routes to this action:

   modelDisplayName: Send a Teams message
   modelDescription: Posts a message to a Teams chat or channel on behalf of the user

   Change Connection Mode

   action:
     connectionProperties:
       mode: Invoker    # or Maker

   • Maker: Uses the developer's credentials (shared connection)
   • Invoker: Each end user authenticates with their own credentials

   MCP Actions — Editable Fields

   MCP actions (InvokeExternalAgentTaskAction with ModelContextProtocolMetadata) have a simpler structure than connector actions. The MCP protocol handles tool discovery, so there are no explicit inputs/outputs.

   Safe to edit:

   modelDisplayName: Microsoft Learn Docs MCP Server   # display name for the orchestrator
   modelDescription: MCP Server to search Microsoft Learn content   # routing description
   action:
     connectionProperties:
       mode: Invoker    # or Maker

   MCP actions may also have ManualTaskInput entries for passing context to the MCP server (e.g., user identity as a header):

   inputs:
     - kind: ManualTaskInput
       propertyName: userid
       value: =System.User.Email

   This is fine — ManualTaskInput with Power Fx expressions can pass user context. But do NOT add AutomaticTaskInput entries — MCP handles tool parameter discovery dynamically.

   Do NOT edit on MCP actions:

   • action.operationDetails.operationId — identifies the MCP operation
   • action.connectionReference — links to the authenticated connection
   • Do NOT add AutomaticTaskInput entries — MCP handles tool discovery dynamically
   • Be cautious with modelDescription — keep it on a single line; multi-line descriptions have been reported to break MCP tool registration after push (see #91)
7. Validate the edited file:

   node ${CLAUDE_SKILL_DIR}/../../scripts/schema-lookup.bundle.js validate <action-file-path>

8. Inform the user about next steps:
   • The action has been edited
   • Push changes via the Copilot Studio VS Code Extension
   • If connection mode was changed, the user may need to reconfigure the connection in the portal

Important Rules

• Never change action.operationId (connector) or action.operationDetails.operationId (MCP) — this identifies which operation runs. Changing it breaks the action.
• Never change action.connectionReference — this links to the authenticated connection. Changing it breaks the action.
• Property names must match the connector definition — use connector-lookup operation to verify exact property names.
• ManualTaskInput values are strings only — if the value needs to be a number, enum, or complex type, warn the user that it may need UI configuration.
• Output propertyName values must match the connector definition's output schema — use connector-lookup operation to see available output properties.
• MCP actions must not have AutomaticTaskInput entries — the MCP protocol handles tool parameter discovery dynamically. ManualTaskInput entries are OK for passing context (e.g., user identity via Power Fx expressions like =System.User.Email).
• MCP modelDescription should be single-line — multi-line descriptions have been reported to break MCP tool registration after push.