[JS] Prompts always appends user message on resume, breaking interrupt resume

[Ehesp]

Given the scenario; I've got a Genkit prompt which generates a poem. The prompt has a tool, which handles interruption (asking the user to review the poem).

I invoke this like so:

const response = await poemPrompt(input, {
  messages: input.resumable?.messages ?? [],
  resume: input.resumable ? {
    respond: input.resumable?.responses,
  } : undefined,
});

This works fine, and the tool is triggered. I collate the response(s), the messages which existed on the response at the time the interrupts happened.

I then run the prompt again with the resumable object, containing those messages and responses, as above.

I get an error:

GenkitError3: FAILED_PRECONDITION: Cannot 'resume' generation unless the previous message is a model message with at least one tool request.

Looking at the code:

When resuming an interrupted generation by calling a defined prompt with existing history and the resume payload (e.g. prompt(input, { messages: response.messages, resume: { respond: answers } })), the prompt pipeline still appends the template as a new user message in renderUserPrompt().

That makes the last message in messages a user message instead of the model message that contains the tool request, so the resume check in generate() fails with "Cannot 'resume' generation unless the previous message is a model message with at least one tool request." The resume option is correctly forwarded into the final generate options, but the message list is already wrong.

I think the prompt layer should skip appending the user prompt when renderOptions.resume is set (e.g. in renderUserPrompt(), only push the user message when !renderOptions?.resume), but I'm not 100% sure.

Runtime (please complete the following information):

• OS: MacOS
• Version 26.2

** Node version
v24.8.0

Additional context

Note, I'm not using dotprompt files, I'm defining the prompts in code but not sure this makes any difference.