Enterprise DNA
S Skills Claude Skills one click

Brand Guidelines Skill

by Anthropic

Teach Claude your brand voice, colours, typography, and forbidden phrases. Ships in the Anthropic skills repo.

BG

Skills

Brand Guidelines Skill

Added 17 May 2026

#claude-skill #brand #voice #guidelines #anthropic

Overview

The Brand Guidelines skill is the reference pattern for encoding a brand into a skill Claude loads on demand. Colour palette, typography rules, voice and tone, banned phrases, logo usage. Once installed, every relevant generation respects the brand without re-prompting.

Best for

Best for
Any team where on-brand output across writers and agents matters

Use cases

  • Keep all generated content in voice, no matter who in the team prompts
  • Stop the agent reaching for off-brand colours in slide and HTML output
  • Block forbidden phrases (legal-sensitive, retired product names) at the agent layer
  • Onboard a new brand to an existing agent stack in one commit

Notes

Why it matters

A skill is the right home for “things the agent should always know about our brand.” Better than a system prompt (auto-loaded) and better than a CLAUDE.md entry (composable with other skills).

How teams use it in production

Fork the reference skill, drop in your colours, typography, voice, and banned phrases. Commit. Every relevant agent run from then on respects the brand without anyone repeating it.

What to watch

Vendor-shipped brand skills are the obvious next step. A brand becomes something you install, not something you re-prompt.

Pros

  • Reference-quality pattern for any 'teach the agent our brand' problem
  • Composable with the document skills (DOCX, PPTX, PDF) for downstream output
  • Easy to fork for per-brand variants in a multi-brand org
  • Centralised brand state, no more 'remember our colours' in every prompt

Cons

  • Quality is bounded by how disciplined the source guidelines are
  • Visual brand still needs templates or shells to fully enforce
  • Trigger needs careful tuning to avoid firing on irrelevant prompts