Skip to content

Makes

Creative brief

Turns an ad angle, a post or a theme into a brief a designer can build, with the format, the concept, the copy and the proof behind each claim.

Other tools

Add this skill to Claude

Claude takes skills as a ZIP file. Your download has started; if it didn't, use the button below.

  1. 1In Claude, open Customize, then Skills.
  2. 2Choose +, then Upload a skill, and pick creative-brief.zip.
  3. 3Ask for it by name, or let Claude use it when a task matches.

Skills need code execution turned on in Claude's settings. On a Team or Enterprise plan, an admin can share it with everyone.

Download the folder

Try it now, without installing

See the skill work on a made-up input before you add it. Nothing is installed, and you can swap in your own input.

Open in the Claude app

Opens a new chat in the Claude desktop app with the prompt filled in. Nothing is sent until you press enter.

Includes the whole SKILL.md and the sample input. Paste it into Claude, ChatGPT or any chat.

Influenced by
3 copy tests
Needs
The source, an ad angle, a post or a theme · The channel and placement
Works with
Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md
Version
1.0.0 · Last checked · License MIT
Maintained by
Sabih AhmedCreative
SkillCreative and conversion

Creative brief

Example output

Brief one piece
FieldBriefCheck
FormatSingle image, one pointPass
ProofMedian 1-day turnaround, source on the imagePass
SizesFrom the platform's spec pageConfirm

Turns an ad angle, a post or a theme into a brief a designer can build, with the format, the concept, the copy and the proof behind each claim.

The skill

The whole skill, in plain text.

Nothing is hidden. Read it, change it to fit your team, or take the whole folder.

Scripts
None. It's instructions only.
Network
Opens web pages or your tools if your agent can. Works from pasted data if not.

3 test cases

creative-brief/SKILL.md

In the folder

  • creative-brief/SKILL.md
  • creative-brief/references/formats.md
---name: creative-briefdescription: >-  Turns a theme, a post or an ad angle into a creative brief a designer can build without a meeting: the one  message, the format and why, the visual concept and panel count, the on-image copy line by line, the text around  it, and the proof each claim rests on. Writes 2 executions that really differ, then checks its own copy with  voice-check and claim-check before handing over. Use it after choosing an ad angle or a post to turn into a  visual. Not for choosing the angle (use ad-themes), writing ad variants at scale (use ad-variants) or reviewing  finished ads (use ad-creative-critic).license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Reads the platform's current ad specs  if the agent can browse; otherwise marks sizes and limits to confirm. Never generates the image itself.metadata:  author: SHAKEUP  version: "1.0.0"  stage: make  plugin: creative-conversion  homepage: https://www.shakeuphq.com/skills/creative-brief--- # Creative brief Specs one piece of creative so the designer builds the right thing the first time, with every claim on it backed. ## Purpose - One job: one brief for one piece (a social image, a multi-panel post, a video, a static or video ad, a page hero), with 2 executions.- Stage: make. It ends with its own check.- It never makes the asset, never picks colours, type or illustration style (that is the brand guide's job), never rewrites a hook another skill already chose without saying why, and never puts a figure on a piece without its source. ## Use when / not | Situation | Use || --- | --- || An ad angle, a post or a theme needs a visual piece | this skill || A founder post should get an image or a multi-panel version | this skill || You don't yet know which angle to run | `ad-themes`, then this skill || You need many copy variants within platform limits | `ad-variants` || The ads exist and need checking | `ad-creative-critic` || You need the buyer's words first | `message-mining` | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || The source: a theme, a post, or an ad angle (with its hook if one was chosen) | yes | an `ad-themes`, `hook-writer` or `founder-post` handoff; the post or angle pasted | ask once; with none, `BLOCKED` || Channel and placement (a LinkedIn or Meta feed, Stories, display, a page hero) | yes | the handoff; the reader's note | ask once; with no answer, `BLOCKED` || The goal and the next step (awareness, a click to a page, a reply, a download) | no | the handoff; the reader's note | infer from the channel and source; say so || Placement specs (sizes, text limits) | no | the platform's ad specs page, read today; specs the reader pastes | write `confirm` beside each size and limit || Brand guide (colours, type, logo rules, illustration style) | no | a file or link the reader shares | the brief says "style: brand guide" and leaves style to the designer || Buyer phrases | no | a `message-mining` handoff | use the profile's buying group and messaging house wording | ## Profile read From `company-profile.md`: `## Company` (brand terms), `### Buying group` (who the piece speaks to and their objection), `## Buying moments`, `## Messaging house` (the pillar and its proof), `## Voice`, `## Claims and proof` and `### Never say`, `## Competitors` (mention policy), and every `## Learnings` entry for `creative-brief` with status `noted`, `fact` or `accepted` (for example a format or length that worked for this company). No profile: ask in one message for who the piece is for, what the product does, and the figures the company can use in public with their sources. Say that `company-profile` saves this next time. Without approved claims, the piece carries no figures until the reader sources them. ## Steps 1. **Apply learnings.** A learning about formats, lengths or concepts for this company overrides the defaults in `references/formats.md`. Name it when it fires. 2. **Read the source and write the one message.** From the theme, post or angle, write: the one message in one sentence (if you need 2 sentences, there are 2 pieces); the buyer role it speaks to; the buying moment it meets; the messaging house pillar it serves. If the source carries a hook from `hook-writer` or `ad-themes`, keep it word for word; change it only to fix a failed check, and say which. 3. **Match the proof.** List every claim the piece would make. For each, find the matching line in `## Claims and proof`.    | The claim | Do this |   | --- | --- |   | Matches an approved claim (same figure, same thing, same period) | use it; the source line goes on the piece |   | Has no match, but a fact about the product could carry it | rewrite it as the mechanism ("reads your CRM stages, never rewrites them"), no figure |   | Has no match, and the message stands without it | keep it off the piece; list it in open items with the source it needs |   | Has no match and the message depends on it | stop: `NEEDS DECISION` |   | Is on `### Never say`, or breaks the competitor mention policy | cut it; if the message depends on it, `NEEDS DECISION` | 4. **Pick the format.** Use the format table in `references/formats.md`, first match wins. A poll or a video needs the reason written in the brief. 5. **Pick the visual concept and the panel count.** Use the concept table in `references/formats.md` by the shape of the content. Panels equal the distinct points; never pad or cram. Literal and labelled before metaphor. 6. **Write the copy.**   - On-image copy: one line per element in viewing order, never a paragraph. The title stands alone. No label printed twice. Any figure carries its source line on the image.   - Text around the piece: by placement, from the text table in `references/formats.md`, within the limits from today's spec (or marked `confirm`).   - Words: buyer phrases from `message-mining` where they exist, the profile's words to use, nothing from words to avoid.   - Labels on a diagram are claims too: every step, day, stage or figure on a timeline, flowchart or scorecard comes from the source or the profile. Never fill a diagram with steps nobody gave you. 7. **Write the second execution.** It must change the visual concept, the structure or the title's angle. If the same on-image copy fits both, start again. 8. **Spec the build.** For each execution: format, size or ratio per placement (from the spec or `confirm`), panel count, what each panel shows, the source line for each figure, alt text (what the image shows, in one sentence, for screen readers), and the style line (from the brand guide, or "style: brand guide"). 9. **Check.** Run the loop below. ## Check and verdict The brief's own checks, each pass or fail: | Check | Passes when || --- | --- || B1 One message | The message is one sentence and every panel serves it || B2 Proof attached | Every claim maps to `## Claims and proof` or is a mechanism; every figure has its source line on the piece || B3 Format earned | The format matches the table; an exception has its reason written || B4 Panels match points | Panel count equals distinct points (plus title and close where shown one at a time) || B5 Executions differ | The 2 executions differ in concept, structure or angle || B6 Copy rules | Title stands alone; no label printed twice; no paragraph on the image; nothing that blames the buyer || B7 Voice | `voice-check` returns PASS or PASS WITH NOTES on all copy || B8 Claims | `claim-check` returns PASS or PASS WITH NOTES on all copy || B9 Specs | Every size and limit is from today's spec or marked `confirm` || B10 Style left to the brand | No colour, type or mood picked unless the brand guide was given | Without `voice-check` and `claim-check` installed, run B7 and B8 inline: no dashes, nothing from `### Words to avoid`, the 3 copy tests (can I picture it, can it be proved wrong, could no one else say it), and every figure traced to a line in `## Claims and proof`. | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No source, or no channel and placement after asking || `NEEDS DECISION` | The message depends on a claim with no approved source, or on one the profile forbids || `READY WITH OPEN ITEMS` | All checks pass except B9 (sizes to confirm), or a check still fails after 2 rounds and is labelled || `READY` | Every check passes | Named stops: `BLOCKED` asks for the missing source or placement. `NEEDS DECISION` shows 2 options and stops: (a) the same piece built on the mechanism instead of the figure, written out; (b) keep the figure once the reader adds its source to `## Claims and proof`. Never invent or borrow a figure to fill the gap. No other stops. ## Loop 1. Run B1 to B10.2. Fix each fail: rewrite the line, merge or split panels, change the second execution, or replace a figure with its mechanism.3. Run the checks again. At most 2 rounds. Anything still failing goes to open items with its check id, and the verdict is `READY WITH OPEN ITEMS`. ## Learn Append to `## Learnings` in the profile: - The reader reports how a piece did (saves, click-through rate, cost per lead, replies, from their platform data): a `fact` with the format, the concept, the result and the source.- The reader overrides a format or concept choice ("our buyers ignore polls"): a `proposal` to change the default for this company.- A designer sends a brief back as unclear: a `fact` naming the field that was missing. Example: `- 2026-10-24 | creative-brief | fact | 5-panel document posts drew 3x the saves of single images in the company page's feed | Why: the platform's page analytics, Jul to Sep 2026, 12 posts | Status: noted` ## Output ```Verdict: <READY | READY WITH OPEN ITEMS | NEEDS DECISION | BLOCKED> · Coverage: <complete | partial (what's missing)>Learnings applied: <list or none> The one message: <one sentence>Buyer: <role> · Moment: <buying moment> · Pillar: <pillar> · Goal: <goal> · Channel and placement: <...>Hook: "<hook>" (<carried from ... | written here>) Proof| Claim on the piece | Source (from Claims and proof) | Shown as | Execution 1: <name>Format: <format> (<why>) · Concept: <concept> · Panels: <n> · Size or ratio: <from spec | confirm>Summary: <the visual idea and its layout in 2 sentences, no colour or style>On-image copy:- Title: <line>- <element>: <line>Text around it: <hook, beats, close or headline, intro text, button>Alt text: <one sentence>Style: <from the brand guide | style: brand guide> Execution 2: <name>(same fields) Checks: B1 <pass/fail> ... B10 · voice-check <verdict> · claim-check <verdict> (<or: inline>)Open: <items with check ids>``` Then the handoff block. ## Hand off ```handoffskill: creative-briefversion: 1.0.0verdict: READYcoverage: partial (sizes to confirm on the platform's spec page)for: <campaign, post or theme>outputs:  message: <one sentence>  format: <format>  concept: <concept>  executions: [<name 1>, <name 2>]  claims: [<approved claim used>]  landing_page: <URL or none>next: [ad-creative-critic]open: [confirm sizes]``` `ad-creative-critic` reads `message`, `claims` and `landing_page` to check the finished ads against the brief. For `NEEDS DECISION`, `next` is `[company-profile]` (to add the source) and `open` names the claim. ## Failure modes | Problem | Do this || --- | --- || The source has 2 or more messages | Write a brief for the strongest one; list the others as separate briefs to make || The reader asks for "something eye-catching" with no message | Ask for the one thing the buyer should remember; until then `BLOCKED` || The source includes figures from a blog or a report | They are external claims: `claim-check` must open the source; until it does, keep them off the piece || 2 rounds of designer feedback miss the same way | Stop guessing: ask the reader to describe what they picture, then rewrite the summary || The brand guide asks for a mascot or a style the rules here avoid | The brand guide wins on style; the copy rules still apply || The placement isn't known yet | Brief for the format and concept; mark every size and limit `confirm`; verdict `READY WITH OPEN ITEMS` | ## Influenced by - SHAKEUP's social pipeline: one message per piece, a default format with earned exceptions, panels that match the points, literal labelled diagrams before metaphor, 2 executions that really differ, and no style choices in the brief.- Harry Dry's 3 copy tests, used in the copy check: https://www.demandcurve.com/newsletters/growth-newsletter-202

How it thinks

Inputs in, a verdict out, and a stop only where you decide.

What we built

A brief fails when it leaves the designer to guess or carries a claim nobody can back. This one fixes the message, the proof and the copy before anyone opens a design tool, and leaves style to your brand guide.

It reads your company profile if you have one, or asks for only what it needs. Then it works through its steps, checks its own work and ends in a verdict.

The steps

  1. 01

    Write the one message

    One sentence the piece must leave behind. If it takes 2 sentences, that's 2 pieces. A hook chosen earlier is kept word for word.

  2. 02

    Match every claim to its proof

    Each claim maps to an approved line in your profile, with its source printed on the piece. A claim with no source becomes the mechanism behind it, or the run stops for you.

  3. 03

    Pick the format, and earn the exceptions

    One point gets a single image. Ordered steps get one panel each. A poll or a video needs a written reason, not a wish for variety.

  4. 04

    Pick the concept by the shape of the content

    2 things side by side make a comparison, ordered stages a timeline or funnel, a process a flowchart. Literal and labelled comes before metaphor.

  5. 05

    Write the copy line by line

    On-image copy in viewing order, never a paragraph, with a title that makes sense on its own. The text around the piece follows the placement's own limits.

  6. 06

    Write a second execution that really differs

    It must change the concept, the structure or the angle. If the same copy fits both, it starts again.

  7. 07

    Check, fix, check again

    The voice check and the claim check run on all the copy, plus 10 brief checks. It fixes and rechecks twice at most, then labels what's still open.

What you need

  • Required

    The source, an ad angle, a post or a theme

    A handoff from ad themes, hook writer or founder post → The angle or post pasted

  • Required

    The channel and placement

    The handoff → Your note, such as a feed ad or a page hero

  • Optional

    The placement's sizes and text limits

    The platform's ad specs page, read on the day → Specs you paste

  • Optional

    Your brand guide

    A file or link you share

  • Optional

    Your buyers' own words

    A message mining handoff

What it can return

  • READY

    Every check passes.

  • READY WITH OPEN ITEMS

    Every check passes except sizes to confirm, or something still fails after 2 rounds and is labelled.

  • NEEDS DECISION

    The message depends on a claim you haven't sourced, or one your profile rules out.

  • BLOCKED

    There's no source to brief, or no channel and placement after asking.

Where it stops for you

  • When there's nothing to brief, or no placement
  • When the message depends on a claim with no source, it shows 2 options and waits

Example output

What you get back.

A run for a made-up company. Your output depends on your inputs.

Here's a run for a made-up B2B SaaS company that sells security questionnaire software. The source is an ad angle for a single-image ad in a paid social feed.

Verdict

READY WITH OPEN ITEMS

Coverage: partial (sizes to confirm on the platform's spec page)

The one message

questionnaires go back the same day, so deals stop waiting on security review.

Buyer

head of sales, at the moment a deal stalls in security review. Hook, carried from ad themes: "Your deal is waiting on a questionnaire."

Claim on the pieceSourceShown as
Median turnaround of 1 dayThe company's product data, Q3 2026A source line under the figure

Execution 1, side by side. A comparison in one panel. The left column shows the questionnaire sitting in a security inbox. On the right, answers are drafted from the approved library and sent the same day.

  • Title: Your deal is waiting on a questionnaire.
  • Left label: In the security inbox
  • Right label: Drafted from your approved answers
  • Figure: Median turnaround: 1 day
  • Source line: Product data, Q3 2026

Execution 2, the deal timeline. A timeline of a deal's stages, with the security review step shrunk to a single day. It changes the concept, so it tests a different idea.

Checks

the 10 brief checks pass except sizes, which are marked to confirm. The voice check and the claim check both pass.

The handoff block the next skill reads
skill: creative-brief
version: 1.0.0
verdict: READY WITH OPEN ITEMS
coverage: partial (sizes to confirm)
for: same-day questionnaires, paid social
outputs:
  message: questionnaires go back the same day, so deals stop waiting on security review
  format: single image
  concept: comparison
  executions: [side by side, the deal timeline]
  claims: [median turnaround of 1 day]
  landing_page: same-day questionnaires campaign page
next: [ad-creative-critic]
open: [confirm sizes]

What good looks like

  • One message in one sentence, and every panel serves it
  • Every figure on the piece has its source line beside it
  • Panels match the points in the source, with none padded or crammed
  • 2 executions that a reader can tell apart at a glance
  • Sizes from today's spec, or marked to confirm

Install

Use it in the tool you already have.

Pick your tool. Each tab has the shortest path we know of.

  1. 01
  2. 02
    In Claude, open Customize, then Skills.
  3. 03
    Choose +, then Upload a skill, and pick the ZIP. Code execution must be on in your settings.
  4. 04
    Ask for it by name, or let Claude use it when a task matches.

Questions

Before you add it.

Does it design the image?

No. It writes the spec a designer builds from, including the copy for every element and the alt text. Colours and style come from your brand guide.

What if our best angle has no proof?

It stops and shows 2 options. One is the same piece built on the mechanism. The other keeps the figure once you add its source to your profile.

Why 2 executions?

So you can test ideas, not layouts. The second one must change the concept, the structure or the angle, or it doesn't count.

Does it know each platform's sizes?

It reads the platform's spec page if your agent can browse. Otherwise it marks each size to confirm, because specs change and a remembered size is often wrong.