StickerGiant Brand Playbook

The PDF was not the system

StickerGiant already had brand guidance. The problem was where it lived and who could use it. A PDF worked for a designer looking up a logo rule, but it could not answer an agent, validate an asset, or keep an automated workflow from drifting off brand.

I rebuilt the guidance as structured data. The website, command-line tools, APIs, and automated reviews now read the same rules instead of maintaining separate interpretations.

One set of rules, three ways to use it

The Brand Playbook has three layers:

• For people: A web interface for browsing guidance, previewing components, and editing rules through a CMS.
• For agents: A small Python CLI that fetches rules by category, checks assets, and returns structured responses.
• For automation: JSON manifests, API endpoints, an OpenAPI specification, and an MCP server that other tools can call directly.

Every rule starts in JSON. The interfaces around it can change without creating another copy of the brand guidance.

What the playbook covers

The foundation includes color, type, logo usage, voice, accessibility requirements, and rules for deciding when AI can act on its own. Channel guides add constraints for social, email, print, packaging, events, advertising, and web. Shared resources connect the data to Figma tokens, downloadable assets, and component documentation.

Deciding what AI is allowed to do

Not every brand task carries the same risk. The playbook separates work into four levels:

• AI-final: Deterministic formatting, tags, and clearly decorative images that require empty alt text.
• AI-assisted: An agent drafts; a person reviews and approves. This includes alt text for meaningful images.
• Concept-only: An agent can open up options, but a person chooses and rewrites the direction.
• Human-only: Crisis response, executive communication, and other brand-defining statements.

Those boundaries are data, not a paragraph someone has to remember. An agent can check the rule before it acts.

How automated review works

When content enters the review queue, the workflow identifies the channel and content type, fetches the relevant rules, checks the submission, and returns a decision with the reasons attached. Clear passes can continue. Uncertain or sensitive work goes to a person.

The point is not to remove judgment. It is to stop spending that judgment on checks the system can make reliably.

Why I kept the stack small

The public interface uses plain HTML, CSS, and JavaScript. Decap CMS handles editing. Netlify hosts the site and API functions. The Python CLI uses only the standard library.

That choice keeps the Brand Playbook easy to deploy and easy for other tools to consume. There is no build framework between the rule and the system asking for it.

My role

I designed the information model, wrote the brand and AI-governance rules, built the site and integrations, and connected the playbook to the workflows that use it.