Blog
Technical guides, tips, and explainers for developers — tools, APIs, AI, and more.
Stop Approving Every Claude Code Command: Set Up the Built-In Sandbox
Claude Code's built-in Bash sandbox lets you run tests, installs, and builds without approving every command — enforced at the OS level, not by string matching.
Read more →Give Claude Code Its Own Mac: A Practical Guide to Isolated Agent Delegation
How to set up a spare Mac that Claude Code can fully control via SSH and computer-use, so risky --dangerously-skip-permissions work never touches your main machine.
Read more →Firefox Runs in WebAssembly Now — Three Lessons for Your Own Port
Gecko compiled to WASM is a 233MB curiosity. But the choices behind it — single-process, Emscripten, WebSocket tunnelling — are the same ones any desktop-to-browser port faces.
Read more →The LLM Cliché Highlighter Is an Editing Tool, Not an AI Detector
A browser tool highlights phrases that give away LLM-written text. Useful for editing your own drafts — and useless for catching anyone.
Read more →Mermaid to ASCII: Diagrams That Survive Plain Text
A Go library compiled to WebAssembly turns Mermaid syntax into ASCII box art in your browser. Here's why plain-text diagrams beat rendered ones in code.
Read more →The SQLite Setting That Took a Query From 5s to 0.05s
SQLite ships with no query planner statistics. Running ANALYZE made one search query 100x faster. Here's the short list of things to do before you ship SQLite.
Read more →The Human-in-the-Loop Is Tired: Practical Guardrails for Reviewing AI Agent Work
AI agents shifted the bottleneck from writing code to reviewing it. Here are concrete guardrails to cut supervision fatigue in Claude Code and agent workflows.
Read more →A $0 Local Coding Agent That Never Hits a Rate Limit: OpenCode + Qwen2.5-Coder 7B
You don't need a giant model to escape Claude Code's weekly limits. Here's the honest case for running OpenCode on Qwen2.5-Coder 7B locally — and where it falls short.
Read more →Claude Code Sends Up to 75,000 Tokens Before It Reads Your Prompt — Here's How to Cut That
A recent test found Claude Code sends 32,800–75,000 tokens of overhead before processing your instructions. Here's what causes it and how to reduce it.
Read more →