Roadmap
A public learning path. Updated as I go.
Semantic markup, Flexbox, Grid, responsive design
Closures, async/await, modules, DOM manipulation
Hooks, context, component patterns, state management
Types, interfaces, generics, utility types
Component architecture, directives, two-way data binding, dependency injection
Composition API, reactivity system, Vue Router, component patterns
App Router, RSC, SSR, SSG, API routes
Signals, virtual DOM, and how frameworks handle change — contrasting React, Vue, Svelte, and SolidJS to understand the trade-offs behind each approach
Vitest, React Testing Library, Playwright
Core Web Vitals, code splitting, caching strategies
Syntax, data structures, OOP, scripting, and backend frameworks (FastAPI/Django)
HTTP, middleware, routing, REST API design
JWT, OAuth, session management, security fundamentals
PostgreSQL, SQL fundamentals, ORMs (Prisma/Drizzle)
Branching strategies, PRs, rebasing, Git workflows
Scrum, sprints, standups, retrospectives, iterative delivery
Containers, Docker Compose, multi-stage builds
GitHub Actions, automated testing, deployments
Slash commands, keyboard shortcuts, permission modes, and the basic interaction model
How to write project guidance that shapes Claude's behavior — commands, architecture, conventions, and what to leave out
Understanding the context window, when to start a new session, using /clear, and keeping context focused
Using plan mode for complex tasks, reviewing plans before execution, and knowing when to let Claude run autonomously vs. step-by-step
Configuring pre/post tool-use hooks in settings.json to enforce patterns, run formatters, or gate actions automatically
Writing your own SKILL.md files and plugins to encode repeatable workflows and share them across projects
Dispatching parallel subagents for independent tasks, background agents, and orchestrating complex multi-step automation
Deliberately shaping what goes into the context window — structuring system prompts, injecting relevant state, and managing information so the model has exactly what it needs to reason well
The middleware layer between LLMs and your application — defining tool boundaries, permission rules, and structured prompting so agents behave predictably within a controlled environment
Vector embeddings and similarity search, text chunking and ingestion pipelines, combining retrieval with LLM generation, vector databases (Pinecone), and grounding AI responses in real data
Writing precise, scoped prompts; providing the right context; iterating on AI output effectively without over-correcting
Model Context Protocol — extending Claude with tools for GitHub, databases, browsers, and custom integrations
Critical review patterns, catching hallucinations, maintaining code ownership, and knowing when not to use AI