Case Study · Content Strategy Tooling

AI Skill Library

A searchable catalog that turned one-off prompt engineering into a shared, reusable practice for a content strategy team.

8
Reusable prompts cataloged
3
Skill categories organized
5-min
Data refresh via Google Sheets cache

Context: As the Help Center content strategy team took on more AI-assisted work — evaluating model responses, drafting AI-assisted content, running gap analyses — every strategist was writing prompts from scratch, with no shared reference for what already worked.

Challenge: Effective prompts existed only in scattered notes, chat threads, and individual memory. There was no way to search by use case, no shared signal for which prompts actually worked well, and no consistent format for the ones that did exist.

Approach: Build a searchable, filterable catalog so the team could find, reuse, and rate proven prompts instead of reinventing them each time.

AI Skill Library catalog view

Users can filter by category or difficulty, or search by title, description, or tag. See the fully functional, sanitized mock-up →


Ready-to-use templates

Every prompt, ready to customize

Each entry includes a full prompt template with variable placeholders highlighted, so a teammate can drop in their own help center name, time range, or volume threshold and get a working prompt in seconds — no rewriting from scratch.

Prompt template detail view

Catalog Interface Design

Designed and built a searchable, filterable frontend catalog with category and difficulty filters and a featured-skills panel to surface the most useful prompts first.

Reusable Content Modeling

Structured a reusable content model for prompt entries — title, category, difficulty, use case, and tags — so the catalog could scale without a rigid database schema.

Lightweight Data Architecture

Built the catalog on a Google Sheets data layer with five-minute caching, letting non-technical teammates add or edit prompts without needing engineering support.

Feedback-Driven Curation

Implemented a CSAT-style helpfulness vote on every prompt, surfacing the most-trusted entries to the top and standardizing how the team approached prompt engineering.

This is a sanitized, publicly shareable recreation of an internal Meta tool. The catalog structure and interactions are real; the prompt templates shown are illustrative reconstructions of the pattern the library held, written fresh for this public version rather than reproducing internal content.