← skills directory● official

Skill · Anthropic

brand-guidelines.

If you have ever written the same brand brief into three different prompts in one day, this is the fix. A skill that teaches Claude your voice, your colors, and your forbidden words once.

$ /plugin install brand-guidelines@anthropic-agent-skills

Most brand consistency problems with AI come from the same root. Nobody wants to re-paste a brand guide into every prompt, so people just do not, and the output drifts. brand-guidelines is the skill that absorbs the guide once and applies it whenever it is relevant, without anyone having to remember to invoke it.

The skill itself is less interesting than the pattern it represents, which is encoding a team's implicit conventions as an installable artifact. Brand is the obvious case. Design system tokens, code style guides, content tone, and legal boilerplate all want similar treatment. Once the pattern is in your head, every recurring "I keep prompting the same context" moment becomes a candidate for the same approach.

§01What it captures

The skill stores the layers of brand decisions that normally live in a Notion page or a PDF nobody opens:

  • Voice and tone. Whether the brand says "you" or "you are," whether em-dashes are okay, whether the register is dry, warm, or somewhere in between.
  • Color tokens. Primary, secondary, accent, and neutrals, with the hex values the team actually uses in production.
  • Typography. Display, body, and mono fonts, with weights and pairings.
  • Forbidden words. The terms the brand absolutely does not use, which matters most for legal-adjacent products and regulated industries.
  • Examples. On-brand and off-brand samples, because rules-only descriptions do not generalize as well as paired examples do.

§02Where it earns its keep

Marketing teams running content through Claude. Designers asking for variations on a theme. Anyone doing landing pages, social copy, or product UI where the brand voice is supposed to carry through but the model keeps drifting toward generic. The pattern of "I want fifteen variations and I want them all to sound like us" is exactly what this skill collapses.

§03Setup

/plugin install brand-guidelines@anthropic-agent-skills

# the skill prompts you to populate:
#   skills/brand-guidelines/
#     voice.md
#     colors.md
#     typography.md
#     forbidden.md
#     examples/
#       on-brand-1.md
#       off-brand-1.md

§04Caveats

  • Examples beat rules. Two well-chosen on-brand and off-brand pairs do more work than five paragraphs of guidelines. Lean into the examples.
  • The skill biases output, it does not enforce it. Brand-critical copy still wants a human review pass before it ships.
  • Skill files leak in the same way any project file does. If the brand guide is sensitive, the skill repo wants the same access controls the original guide does.