Real capture of the public interface · October 30, 2026
The problem
Across different projects, selecting instructions, design references and review criteria requires context and repetition. Mixing inspiration links with executable instructions also creates ambiguity about what is being applied.
What I built
I created the Orchestrator to bring the working toolkit into a searchable catalog, with selection, confirmation and destination. As an MCP App it brings confirmed instructions into ChatGPT; in the browser it prepares text for the project.
Impact across front and back
Front-end
A catalog with search, filters, selection and staged confirmation. Skills and external references have distinct labels.
Next.js
MCP Apps
Interface de seleção
Back-end
An MCP endpoint and proposal flow with input and origin validation, Turnstile and submission limits. The GitHub App creates a branch and Pull Request for review.
MCP
GitHub App
Turnstile
Validação
Decisions behind the solution
A catalog instead of scattered instructions
Brings design, UX, engineering, review, motion, SEO and accessibility together. The catalog contained 48 options across skills and external references.
Authorship and original resources
Includes my PageSpeed & SEO Optimizer, Anti-AI-Slop directive and Website Image Prompts alongside community references.
Explicit separation of types
Personalizations are reference resources, not automatically activated skills. Tests check this distinction in the generated prompt.
Evolution through proposals and review
Proposals create a branch and Pull Request; input validation, origin checks, limits and Turnstile protect the submission flow.
Automated catalog, interface and submission-protection scenarios.
Results in context
The documented benefit is centralizing and reusing the production process. There is no comparable measurement of time or cost savings. The 48 options include third-party resources; each reference retains its authorship.
