# Standby, Go

> A cue sheet for when the tech fails mid-talk. Type the running order of a talk, class presentation, pitch or demo - one segment per line: what | minutes | what it leans on - and see it in a stage manager's prompt book: the clock against your slot, where the slot runs out, a cue light on every segment (green, amber, red) and a board of everything that can fail. One metered run calls the show: a verdict, and for every segment a standby check, a fallback that works without the thing that failed, and a line to say while you recover - plus what to cut, the pre-show checks and what to watch out for.

URL: https://standby-go.skillsafe.ai/
API tutorial: https://standby-go.skillsafe.ai/api.html

## Free, in the browser
- Reading the running order: minutes like 3, 1.5, 90s or 1:30; needs as comma-separated words (slides, wifi, live demo, login, video call, video, sound, presenter notes, cable, mic, clicker, power, props, volunteer - or your own words). A .txt or .csv file can be dropped in.
- The prompt book: each segment's start time and length, each as tall as its time, a red rule where the slot runs out, the total against the slot.
- The cue lights: internet, live software, a login and a video call count 3; video, sound, presenter notes and a cable count 2; slides, a mic, a clicker, power, props, another person and anything else count 1. Five or more is red, 2 to 4 amber, 0 or 1 green. These weights are a rough rule of thumb, an estimate of fragility, not a measurement.
- The cue-light board: one lamp per need, the time that leans on it, the segments, and a generic pre-show tip.
- The verdict rule, checked in order: rework (more than 10% over the slot, or red segments carry more than half the talk), go (fits the slot with nothing red), hold (everything else).
- Exports: a cue sheet (.csv, one row per segment), Markdown and JSON; kept cue sheets (in the browser, or in your account when signed in).

## Metered (gpt-terra), one task: cues
The model gets every figure from the page and does no arithmetic of its own. It returns the verdict by the rule above, one cue per segment in running order (standby, fallback, say), 0-3 cuts (at least one whenever the talk runs over; none on a go), 2-6 pre-show checks and 0-3 watch-outs. It names no product or brand and adds no person, topic, venue, date or backstory the input does not state. The page re-checks every reply: the verdict against its own rule, a cue for every segment, every cut a real segment, and no number, capitalised name, he/she, field name or brand the input does not hold.

## Sources
Built on the SkillSafe skill @louisblythe/presentation-skills - https://skillsafe.ai/skill/@louisblythe/presentation-skills - (help people deliver clear, engaging demos and pitches: presenting to a room, giving demos, public speaking), narrowed to one question: when something fails mid-talk, what do you do and say next.
The trend note on the page is from the Hacker News front page in October 2026 (a story that a tablet software update broke a presentation feature speakers rely on).