Here will be centralized all the old and current posts and thoughts, with tips and some solutions that I have implemented throughout my career.
How to combine the terminal, browser automation, and computer use to prepare releases, fill in store metadata, and submit apps for review without exposing sensitive data.
How to connect Claude or Codex to a simulator, turn requirements into executable scenarios, and build a mobile QA loop with evidence, logs, and regression tests.
A deep look at the architecture of a Minecraft server built in Elixir, covering Java and Bedrock, OTP processes, RakNet, binary codecs, C terrain generation, and mutation-based persistence.
The line between specification and implementation is disappearing. The developer who writes the best specification will ship faster than the one who types the fastest code.
After 30 days and 141 Claude Code sessions, the data is clear: custom skills are what separate vibe coding from shipping real products. Here's what the numbers say.
A single document that works as an RFC for engineers and as a prompt for agents. It's not a compromise - it's the better version of both.
RFCs assume the reader knows the project. Agents know nothing. Externalizing implicit context is the difference between perfect code in the wrong architecture and code that actually fits.
RFCs are static documents. Prompts are interactive conversations. When you combine the discipline of one with the flexibility of the other, that's where the magic happens.
Without clear acceptance criteria, the agent doesn't know when to stop. The result is gold-plating or incomplete delivery - the same problems humans have, amplified.
Without explicit contracts, each agent invents its own interface. The result? Systems that work alone but break together.