A finished story ad brief your editor can cut from

How a brief gets made

Your team's judgement goes in once. The skill reuses it on every brief, and the checker stops a brief that breaks your rules.

1. Brand profileVoice, avatars, the live offer, a proof bank of claims you can back, real reviews, and the words you never use.
2. Winning adsTranscripts of your best story ads with your team's note on why each one won. The skill copies patterns, never sentences.
3. The skill writesStory beats, three swappable hooks, a timed scene table, b-roll and shot lists, CTA, claims ledger, compliance and editor notes.
4. The checker proves itTiming, voiceover pace, hook length, every claim tied to proof, no invented quotes, no banned words, the exact CTA.
5. Your editor cutsOne body, three hooks, with what to film, what to source and what the brand must still supply.

The brief, as the editor receives it

Every section below is parsed from the brief file the skill wrote. The full Markdown file is in the file viewer further down.

Hooks

Scenes

Claims ledger

The checker, and proof that it fires

The skill runs check_brief.py on its own brief and hands it over only on a pass. It needs only the Python standard library, so it runs in Claude's code environment and on any laptop.

What the drafts taught the skill

Draft 1 passed every check, but read aloud its voiceover was thin: 177 words in 117 seconds, so the editor would have inherited long silences. A density check went into the checker and the voiceover was rewritten from the same facts. This is the first draft re-checked with the new rule:

Draft 1 also had a problem no script can see: the founder story says Dana flipped her pillow most nights, and the draft said every night. The final draft brought each line back to the source, and the skill now tells Claude to re-read every narrator line against the founder story, with a matching question in the review checklist. A checker proves structure and claims; a person still reads the story against the facts.

The skill's files

The same files are in the zip, laid out the way Claude loads a skill: a short instructions file that triggers on its description, reference files read only when needed, and a script that runs without entering the conversation.

What it needs from your team, and what it never does