Agent UI: status line keeps showing "Calling <tool>..." after the tool result has arrived

[Ron-Brouwer]

Which component is this issue related to?

Umbraco.AI.Agent (Umbraco.AI.Agent.UI - chat run controller / uai-agent-status)

Which Umbraco AI version are you using?

Umbraco.AI.Agent.UI 17.1.0-rc.3 (Umbraco.AI 17.4.0-rc.3, Umbraco.AI.Agent 17.2.0-rc.3, Umbraco.AI.Agent.Copilot 17.1.0-rc.3, Umbraco CMS 17.6.2). The same code is on v18/dev.

Bug summary

After a backend tool call has returned its result, the status line under the chat keeps showing Calling <tool>... for as long as the model is generating its next turn. The tool call itself is already rendered with a completed tick, so the two indicators contradict each other: the tick says "done", the caption says "still calling". The caption only changes when the next tool call starts or the run finishes.

On a long step (the model reading a large get_umbraco_content or get_context_resource result and then writing a long answer, or preparing a big set_value payload) this lasts a minute or more, during which editors believe the tool is hanging and reach for Cancel.

Specifics

Umbraco.AI.Agent.UI/src/Umbraco.AI.Agent.UI/Client/src/chat/services/run.controller.ts (line numbers for tag Umbraco.AI.Agent.UI@17.1.0-rc.3; v18/dev in brackets):

• onToolCallStart sets this.#agentState.next({ status: "executing", currentStep: \Calling ${info.name}...` })` - lines 326 and 353 [275, 302].
• onToolCallResult → #handleServerToolResult (line 418 [367]) marks the tool call completed and appends the tool message, but never touches #agentState.
• onTextStart / onTextDelta / onTextEnd (lines 249–294 [198–]) never touch #agentState either.
• #agentState is only replaced by the next onToolCallStart, by onStateSnapshot, or cleared in #handleRunFinished.

chat/components/agent-status.element.ts renders state.currentStep ?? defaultLabel. While the state object from the last onToolCallStart is still current, the caption stays Calling <tool>... even though the tool call's own status in the message list is already completed.

Steps to reproduce

1. Open the Copilot sidebar on a document with an agent that has the built-in content-read and search scopes (default tools search_umbraco, get_umbraco_content, get_content_type_schema, plus the system tools list_context_resources / get_context_resource).
2. Ask: "Search for pages about , read the three most relevant ones and write a detailed comparison of their content."
3. Watch the status line. search_umbraco and the get_umbraco_content calls each flip to a completed tick; the caption keeps saying Calling get_umbraco_content... for the whole time the model writes the comparison, until the run finishes.

The same happens with a single tool: "Retrieve context resource X (via get_context_resource) and summarise it in detail" shows Calling get_context_resource... throughout the summary generation.

Expected result / actual result

Expected: once the tool result is in and no other tool call is pending or executing, the status line falls back to Thinking... (the model is generating). Calling <tool>... is shown only while a tool is actually running.

Actual: Calling <tool>... remains until the next onToolCallStart or onRunFinished, i.e. for the whole model-generation phase after the tool result.

Suggested fix

In #handleServerToolResult, after updating #currentToolCalls:

const stillRunning = this.#currentToolCalls.some((tc) => tc.status === "pending" || tc.status === "executing");
if (!stillRunning && this.#agentState.value?.status === "executing") {
    this.#agentState.next({ status: "thinking" });
}

and in onTextStart, set { status: "thinking" } (or a dedicated streaming state if you prefer a separate label) so text generation after a tool call is never labelled with the previous tool's name. The awaiting_approval state for frontend / HITL tools should keep precedence, so the fallback should only replace an executing state.

Dependencies

Umbraco CMS 17.6.2, Umbraco.AI.Anthropic 17.1.0 (provider does not matter; the caption is set client-side).

[mattbrailsford]

This fix is now released in Umbraco.AI.Agent.UI 18.0.5 (v18) and 17.0.5 (v17), both available on NuGet and npm.