| name | relaydeck-prompts |
| description | Ask a human for a decision with tap-able buttons instead of waiting for a typed reply. |
Asking a human for a decision
When you need a human to approve, choose, or confirm something, don't print
"reply YES to continue" and wait for free text. Raise an interactive
prompt: a question plus a set of buttons. The human taps one (in Telegram,
the dashboard, or wherever they're reached) and you get the structured
decision back — faster, unambiguous, and no typing.
Block until they answer (recommended)
relaydeck prompts ask "Deploy v2.3 to production?" \
-c approve:Approve:primary \
-c reject:Reject:danger \
-c hold:"Hold for review" \
--wait
--wait blocks and prints the decision as JSON when a human picks:
{"id": "prm_ab12…", "state": "answered", "choice": "approve", "answered_by": "telegram:ops @alice"}
Read choice (the stable id you defined) and branch on it. state is
answered, or expired if you passed --expires and nobody responded.
Expiry is checked by a background sweep (every ~30s by default), so a
prompt may live a little past its --expires deadline before flipping to
expired. Don't rely on second-precise TTLs.
Fire-and-continue (async)
Omit --wait to send the prompt and keep working. When a human answers, the
decision is delivered into your session as a relay line:
[relay from=prompt id=msg_…] ✅ Prompt prm_ab12… answered: Approve (choice=approve) by telegram:ops @alice
Choices
Each -c / --choice is id[:label[:style]]:
- id — the stable token you branch on (
approve). Required.
- label — what the human sees (
Approve & deploy). Defaults to the id.
- style —
default | primary | danger | success. Cosmetic hint
(button color where the platform supports it).
Where it goes — --to (provider-agnostic)
By default a prompt goes to the dashboard (web). Add --to to fan it out
to messaging platforms; the first human to answer anywhere wins, and the
others' buttons are retracted. Addresses are
channel[:connection[:target[:thread]]]:
--to web --to telegram:ops:-1001234567890 --to telegram:alerts:123456789
You don't need to know whether a channel supports buttons: if it does, you
get native buttons; if not (e.g. SMS), the platform shows a numbered list and
the human replies with a number or the choice word — same result either way.
Other commands
relaydeck prompts list — open prompts (add --state all).relaydeck prompts show <id> — full prompt JSON incl. delivery status.relaydeck prompts answer <id> <choice> — answer on a human's behalf.
