Brand Skills

Brand Skills are procedural playbooks (SKILL.md files) that teach Cuppa how to perform tasks for your brand. Where Knowledge Sources give the AI facts to reference, Skills give it repeatable methods to follow: your SEO audit checklist, your social posting formula, your editorial review process.

Why Brand Skills Matter

A skill is a set of step-by-step instructions the AI follows when a task matches it. Skills are how power users encode their expertise once and have it applied consistently across every piece of content.

Knowledge Sources answer “what does the brand know?” Skills answer “how does the brand work?”

Brand Skills are available on the Power plan and above.

What a Skill Looks Like

A skill is a markdown file named SKILL.md with a YAML frontmatter block and a markdown body:

---
name: SEO Audit
description: Use when auditing a page or diagnosing why content isn't ranking.
---
 
# SEO Audit Playbook
 
When auditing a page, work through these checks in order:
 
1. Confirm the target keyword appears in the title, H1, and first 100 words.
2. Check heading hierarchy (one H1, logical H2/H3 nesting).
3. ...

• name: A short, human-readable title (max 120 characters).
• description: One line that tells the AI when to use the skill. This is the most important field for auto-matching, so be specific about the trigger.
• body: The instructions themselves. Write them as you’d brief a capable teammate.

A skill can also bundle reference files (templates, checklists, examples) alongside the SKILL.md. The AI loads those on demand when it needs them.

How Skills Are Used

When you generate content, Cuppa matches your active skills against the task and injects the most relevant ones into the prompt. This happens across every surface:

• Article generation (all article types, recipes, local news)
• Social posts, newsletters, and carousels
• Page generation
• Editor AI and Workbench
• Agentic Chat

Auto-matching

For each generation, Cuppa compares the task (keyword, topic, content type) against each skill’s name and description and includes the strongest matches. A clear, specific description dramatically improves matching, so write it as a trigger:

Good: “Use when writing a competitor comparison or alternative page.” Bad: “Comparison stuff.”

Explicit attachment

You can also attach specific skills to an AI Instructions preset so they’re always used for content generated with that preset, regardless of auto-matching. See AI Instructions.

In Agentic Chat

Chat uses progressive disclosure. It first sees a list of your installed skills (slug + description only). When a task matches, it loads the full skill on demand, then follows the steps. This keeps the context lean while still giving the agent access to everything you’ve installed.

Adding Skills

Navigate to Brand Settings > Skills, then choose one of three methods.

Import from GitHub

Paste a GitHub repository URL or owner/repo reference. Cuppa recursively discovers every SKILL.md in the repo, shows you a preview, and lets you select which skills to install.

Each imported skill is pinned to the exact commit it was imported from. The content is stored as an immutable snapshot, so a later change (or compromise) of the upstream repo never silently alters what Cuppa runs. To pick up upstream updates, re-import explicitly.

https://github.com/anthropics/skills
anthropics/skills
anthropics/skills/tree/main/document-skills

Upload a Zip

Upload a .zip containing one or more skill folders. Each folder with a SKILL.md becomes a skill; bundled reference files travel with it. Zips are scanned for safety (size limits, path traversal, and zip-bomb protection) before anything is stored.

Create Manually

Enter a name, a description, and the markdown body directly in Cuppa. Best for short, brand-specific procedures you don’t want to manage in a repo.

Security

Skills are instructions the AI follows, so Cuppa treats them as untrusted input and applies several safeguards:

• Immutable snapshots: Imported content is stored as a point-in-time snapshot. Upstream changes never affect installed skills until you re-import.
• Commit pinning: GitHub imports record the exact commit SHA.
• Prompt-injection linting: Skill content is scanned for patterns that attempt to override system instructions, exfiltrate data, or escalate privileges. Suspicious content is flagged on import.
• Content delimiting: When a skill is injected into a prompt, its body is wrapped in delimiters and labeled as reference material that complements but never overrides Cuppa’s system instructions.
• Bundled-file limits: Allowed file types and maximum sizes are enforced for bundled resources.
• Admin-only imports: Only team admins can import or create skills.

What to avoid putting in a skill:

• Secrets, API keys, or credentials
• Instructions that try to disable safety rules or impersonate the system
• Conflicting procedures that contradict your other skills

Enabling, Disabling, and Deleting

Each skill has an enable toggle. Disable a skill to keep it installed but exclude it from generation. Delete removes it entirely.

Best Practices

Write descriptions as triggers

The description is what auto-matching reads. State the exact situation the skill applies to (“Use when…”) rather than describing the skill abstractly.

Keep skills focused

One skill, one procedure. A focused “Schema Markup” skill matches and applies more reliably than a sprawling “Everything SEO” skill.

Use numbered steps

The AI follows ordered steps more faithfully than prose. Structure the body as a checklist or sequence.

Bundle examples

Include a reference file with a strong example output. The agent reads it on demand and matches its quality and format.

Review after upstream changes

Imported skills are pinned. When the source repo improves a skill, re-import to pick up the new version.

Availability Across the Product

Brand Skills are available everywhere you create content, plus the API, MCP server, and CLI:

• API: /v1/skills (list, get, create, import from GitHub)
• MCP: list_skills, get_skill, create_skill, import_skills_github
• CLI: cuppa skill list|get|create|import-github|enable|disable|delete

See API reference, MCP Server, and CLI.

Troubleshooting

”My skill isn’t being applied”

• Confirm the skill is enabled.
• Make the description more specific to the task (it drives matching).
• Attach the skill to your AI Instructions preset to force its use.
• Check your plan: Skills require Power and above.

”Import found no skills”

Cuppa looks for files named SKILL.md. Confirm the repo or zip contains at least one, and that the path you provided contains them.

”Import was flagged”

The content tripped the prompt-injection linter. Review the skill body for instructions that try to override system behavior, then re-import a clean version.

Related Features

• Knowledge Sources: Facts and reference material for RAG
• AI Instructions: Generation settings and skill attachment
• Agentic Chat: Chat that loads and applies your skills