Day 3: Tool-Centric Development

Day 2 taught you to build applications. Day 3 teaches you the tools and patterns that turn a single prompt into a reliable development system.

A prompt can build a feature. But a project that grows across days, across sessions, across commits needs something more: it needs the agent to remember what you decided yesterday so it does not contradict it today, it needs specifications that keep the agent on track instead of letting it wander, it needs a file structure the agent can navigate without getting lost, and it needs repeatable workflows you trigger with a slash command instead of retyping paragraphs. These are not nice-to-haves. They are what separate a toy project from a real one when the code is being written by natural language.

By the end of today, you will have the four patterns that make an agent reliable across time: persistent project memory, spec-driven development, custom command workflows, and architecture that scales with your project. Each one is a single lesson, each one is immediately useful on its own, and together they form the backbone of every vibe-coded project that survives past the first weekend.

What you will cover

• Agent Memory & Context Management -- how CLAUDE.md carries project state, decisions, and conventions across sessions so your agent remembers what you built yesterday. What goes in the file vs what stays in your prompt. How to update it as your project grows.
• Custom Commands for Project Management -- build reusable slash commands that encode your project management workflow. /spec to generate a spec from a description, /review to audit code against the spec, /deploy to ship changes. Each command becomes a tool you build once and use forever.
• Spec-Driven Development -- how to write a specification document that an LLM actually follows. The spec format that works: what it builds, what it does NOT build, the contract it must satisfy, the edge cases it must handle. A real spec, a real agent following it, and the difference between a good spec and a bad one.
• Multi-File Project Architecture -- how to structure a project so the agent can work on it without getting lost. File organization patterns that LLMs understand, separation of concerns, when to split vs keep together, and how to describe your architecture so the agent respects it. Real before/after: a monolith that confused the agent, restructured into components it navigates reliably.

These four patterns are the infrastructure of vibe coding. Not heavy process — lightweight conventions that you set once and the agent follows automatically. When they are in place, your project stops being a sequence of isolated prompts and starts being a system that grows coherently.

Next: Day 4 -- Systems Design & Project Management