Skip to content

gooseworks-ai/goose-skills / skills/video/write-video-ad-script

write-video-ad-script

Verified

Write or revise a short-form ad from a sourced angle and reference. Plan connected words, product proof and sound; expose production gaps before script review.

references/pacing.md

View on GitHub
# Human version

**Summary.** Plan the pace for the particular ad, with a separate budget for each spoken
beat. A brisk creator reference can guide delivery without proving the selected engine
can reproduce it. Keep recipe limits, proof and the CTA intact. The finished read decides
whether the plan works.

Choose the target from explicit brief direction or observed reference audio, then the
recipe default. Use a labelled fallback only when those are absent. A demonstration,
testimonial and quiet product story can need different delivery, even on the same engine.
Do not treat a recipe's estimated rate as a measured engine ceiling.

---

# Agent version

## Resolve the actual ad's plan

1. Read the full recipe and its generation capability. Preserve explicit word limits,
   per-call duration caps, supported assembly and the allowed voice path.
2. Determine delivery style from the brief and chosen reference. A transcript establishes
   wording, not cadence. Measure heard words against the audible speech window, record
   pauses separately, and cite an observed audio/video record. Do not infer timing from
   transcript length or total video runtime.
3. Write the desired words-per-second and pacing source into the run's shape. Allow a
   beat override for a slower proof, a quick reaction or a different speaker. Retain the
   recipe's estimated or observed rendered rate as delivery guidance when it differs.
4. Allocate speech time within each beat. Subtract planned pauses or specify a smaller
   speech window for a visual action. A visual with simultaneous narration is a spoken
   beat with a visual plan; a silent demonstration is a visual beat. Silent end cards
   cannot subsidize spoken copy elsewhere. Do not shorten pauses solely to make lint pass.
5. Run strict lint. Its timing rows give word count, speech window, required rate, target
   budget and provenance, including repeated speaker slots. A target above delivery
   guidance is an experiment warning; an explicit recipe limit still applies. Resolve
   material warnings in review, and retain untested targets as unverified.

The files reference defines these fields. Old scalar-rate shapes remain readable.
New timing plans use the strict target budget; legacy estimated budgets retain their
existing tolerance. Rounded word counts are planning aids, especially for languages,
numbers and pronunciation that the tokenizer cannot model accurately.

## Repair and carry the plan into production

Keep the promise, feasible proof and CTA. Tighten wording first without changing supported
claims. If the message still conflicts with the recipe, present a supported longer plan
or a compatible execution through the runtime's existing review. Do not silently delete
the proof, change the template, inflate a tested limit or invent faster-engine evidence.

Give production the exact lines, delivery style, per-beat speech windows, pauses and
unverified targets alongside the native script and visual plan. Separate voiceover can
establish its actual cadence before visuals and lip-sync are aligned. Native-audio video
uses connected natural thoughts, with take boundaries preserving the sentence and
delivery direction supported by its capability;
do not invent a speech-speed parameter. Optional finishing retimes native picture and
audio together, then aligns captions to that finished timeline. It is not the default
solution, and a small speed-up cannot establish a large increase in supported word count.

## Verify the finished read

Record actual moving-picture and listening coverage separately from frames/transcripts.
Unavailable perception stays incomplete; an audio file returned by a tool does not prove
hearing. Preserve one accepted speech source under B-roll and matching source time/rate
when the speaking face returns. No-speech coverage still has a separate soundtrack plan.

Inspect the complete audio and video: every intended word is audible, proof and CTA are
present, pronunciations and word endings are intact, delivery sounds natural, and timing,
captions and lip-sync agree. Record actual speech duration and defects. Update only the
run's result; a successful sample is not proof of a universal engine limit. Recipe
calibration, cost changes and paid comparison runs are separate authorized work.
receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementationStars: 322kUpdated Claude CodeCodexVerifiedInstall
requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirementsStars: 322kUpdated Claude CodeCodexVerifiedInstall
using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallbackStars: 322kUpdated Claude CodeCodexVerifiedInstall
verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions alwaysStars: 322kUpdated Claude CodeCodexVerifiedInstall