You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
- You can test workflows immediately after building — deployment is only needed for external access (API, chat, MCP).
146
-
- All workflow-scoped copilot tools require \`workflowId\`.
147
+
- Tools that operate on a specific workflow such as \`sim_workflow\`, \`sim_test\`, \`sim_deploy\`, and workflow-scoped \`sim_info\` requests require \`workflowId\`.
147
148
- If the user reports errors, route through \`sim_workflow\` and ask it to reproduce, inspect logs, and fix the issue end to end.
148
149
- Variable syntax: \`<blockname.field>\` for block outputs, \`{{ENV_VAR}}\` for env vars.
149
150
`
@@ -667,10 +668,10 @@ async function handleDirectToolCall(
667
668
}
668
669
669
670
/**
670
-
* Build mode uses the main chat orchestrator with the 'fast' command instead of
671
-
* the subagent endpoint. In Go, 'workflow' is not a registered subagent — it's a mode
672
-
* (ModeFast) on the main chat processor that bypasses subagent orchestration and
673
-
* executes all tools directly.
671
+
* Build mode uses the main /api/mcp orchestrator instead of /api/subagent/workflow.
672
+
* The main agent still delegates workflow work to the workflow subagent inside Go;
673
+
* this helper simply uses the full headless lifecycle so build requests behave like
674
+
* the primary MCP chat flow.
674
675
*/
675
676
asyncfunctionhandleBuildToolCall(
676
677
args: Record<string,unknown>,
@@ -680,6 +681,8 @@ async function handleBuildToolCall(
0 commit comments