Aleksandar Velichkovikj

• GRIND

  An interval timer for workouts and focus sessions. Installable to the home screen, and genuinely private — no accounts, no analytics, no network requests once it has loaded. Everything you build stays on your device, with a backup file if you want it somewhere safer.

  Next.js · TypeScript · PWA

  Live

• Toolshed

  Single-page tools for questions that keep coming up: is this website real, where does a link actually go, what is hidden in this text, could someone forge mail from this domain — plus formatting JSON and decoding tokens. No build step, no backend, no accounts.

  HTML · CSS · JavaScript · RDAP · DNS-over-HTTPS · Unicode · JWT

  Open

• Crawler Policy

  Paste a domain and see which AI crawlers its robots.txt actually allows, resolved the way a crawler resolves it rather than by searching the file for a name.

  Next.js · TypeScript · RFC 9309

  Live

• Intercept

  A word game in the Star Wars alphabet. Decode a transmission written in Aurebesh — a real letter-for-letter substitution — and the file on whoever sent it opens. You need to know nothing about Star Wars to play, and you can read Aurebesh by the end of it.

  Next.js · TypeScript · Aurebesh

  Play

Working effectively with agents is a setup problem, not a prompting one. I make production codebases that agents can work in autonomously — documented patterns, rules enforced by tooling, and verification that runs before anything reaches me. On any project of real size, that system is the first thing I build.

Documented patterns, not prompts

Every pattern in the codebase has a written reference — paging, caching, async operations, JSON-driven forms, localisation — plus a feature index tracking each shipped feature to its branch and commit. An agent reads the pattern instead of inventing a fifth way to do the same thing. The same documents onboard a human developer faster, for exactly the same reason.

Rules enforced, not requested

A pre-write hook blocks edits to protected paths. A post-write hook flags any comment an agent adds, so a no-comments rule actually holds. Custom skills wrap the operations that are easy to get wrong — adding a localisation key can never invent a translation or hardcode a string. Guardrails in the tooling hold when a prompt doesn't.

Agents verify their own work

Nothing is reported done on an agent's say-so. A verify skill runs analysis, tests and build and reports the real output. For UI work the agent drives the iOS simulator itself — gestures, screenshots, hot reload — and confirms the change on screen before claiming it works.

Reviewed before I even look

Substantial features go through a multi-agent review pass — architecture, behavioural regressions, financial correctness, security, accessibility — before they reach me. Human review is still required before merge, and the accountability for what ships stays with me.

Mobile

FlutterDartGetX state managementNavigation architectureLocal persistencePush notificationsPerformance profilingWidget and integration testing

Web

Next.jsReactTypeScriptJavaScriptServer-side renderingStatic site generationProgressive web appsInternal dashboardsAdmin panels

Delivery

iOS and Android release management · Apple App Store and Google Play submission · Cross-device testing · Defect triage · Continuous integration basics · Git

Practice

Requirements gathering · Stakeholder communication · Technical documentation · Remote and asynchronous collaboration · Agent-assisted development workflows · Claude Code · MCP integrations · Custom agent skills and hooks · Automated verification and multi-agent code review