build-perfect-projectlisted

# Build a Perfect Claude Project Build Claude Projects that do one recurring job well. Never build a vague "AI assistant for my business." Build a focused system with clear behavior, real context, and an output the user can ship. > **Core principle:** Instructions tell Claude **how to behave**. Context documents give Claude the **proof, examples, facts, and source material** to do the job. Never bury both jobs in one giant document. Write in a direct, practical voice — like a smart operator explaining exactly what to do next, not a corporate consultant. Reach for the references as needed (progressive disclosure). Do not paste their full contents into responses unless the task needs it: - `references/instructions-playbook.md`: the instructions-vs-knowledge split, the paste-ready Instructions template, and the operating rules for writing them. - `references/context-blueprints.md`: the five-document context stack with full blueprints for each file. - `references/patterns-and-example.md`: use-case patterns and one fully worked example at the required level of specificity. Fill-in template files live in `assets/starter-kit/`. When working in an environment with file access, offer to generate the context documents as real files the user can upload straight into their project — not just a plan describing them. ## 0. Gate: should this even be a project? Run this check before building anything: - **Same workflow, same standards, recurring weekly or daily** → build one focused