Anthropic / brand-guidelines
How do you stop Claude inventing a new brand every time?
By giving it the brand once, in a form it reloads on every task.
What to remember
- Prohibitions matter as much as permissions. Write down what the brand must never look like.
- It pays for itself across a team, less so for one person working alone.
- It pairs with theme-factory for the implementation layer and with pptx or docx for deliverables.
- Guidelines are a living file. Fix them when the output drifts rather than re-prompting each time.
01What it changes in practice
Without a brand skill, every session starts from zero and every colleague gets a slightly different interpretation of the brand. The drift is invisible one document at a time and obvious across twenty.
With the rules loaded once, the same constraints apply to a deck, a report and a landing page, which is the only way consistency survives a team that works quickly.
02Where it disappoints
Vague guidelines produce vague compliance. Modern and friendly is not a rule, while a hex value, a type scale and a list of forbidden treatments are.
Strict pixel level brand policing still needs a human reviewer, especially in decks where spacing drifts fastest.
Install and first run
- Install it from anthropics/skills under Apache 2.0.
- Record the real rules, including the ones about what never to do, not just the palette.
- Generate one document and one interface, then correct the guidelines where the output drifted.
Questions people actually ask
- What should go into the guidelines?
- Exact colours and type, tone of voice, and explicit prohibitions. The prohibitions are what stop drift.
- Does it apply to writing as well as visuals?
- Yes, tone rules carry into text output, which matters for customer facing documents.
- Is it worth it for a single person?
- Less so. The value grows with the number of people producing material under the same brand.
Sources
Machine readable versions of this page