An ask-cue turns a vague sentence into the clarifying question a coding assistant would have asked you anyway, and offers the answers as options you cycle inline. You get the round-trip before you send, instead of after.
What it looks like
You type "add caching to the API". Instead of sending that and waiting to be asked, the sentence carries a note: which layer? - with in-memory, Redis and HTTP cache headers as options. Cycle to the one you meant and, where the option carries a concrete change, the sentence is rewritten to say it. Options that are purely advisory just tell you something and leave your text alone.
Where the question comes from
Not from a template. Coding assistants already have a well-defined way of asking you to choose between options, and an ask-cue is populated from that same prompt, so the questions look like the ones the tool itself would raise. It is grounded in whatever the surface knows: on the terminal assistants that means the session so far, and in the browser it means the page and the field you are typing in. A vague sentence gets a relevant question either way.
It stays quiet most of the time
Silence is the default. A sentence that is already clear gets no question, because a needless prompt is worse than none. This was tuned deliberately: an early version had no restraint and asked about everything, which is exhausting rather than helpful.
No new interface to learn
An ask-cue reuses what is already there: the question is the note under the sentence, and the answers are the alternatives you cycle with the same keys you use everywhere else. Nothing changes in your text until you choose.
Turning it on
It is off by default. Add ask-cues-mode: on to ~/.cues/OPENCUES.md and it hot-reloads in a couple of seconds. It works on every surface, and shares a single call with session-contradiction cues when you have both on, so running the pair costs about the same as running one.
Related: Contradict an earlier decision · Sentence cues · What is OpenCues costing me? · GitHub repo