AI Stuff

Hard-won lessons from months of AI-assisted development

Model Context Protocol — the open standard that lets AI assistants use external tools and data

Understanding the three ways to extend AI coding agents: slash commands, MCP servers, and skills

My Growatt guest account kept getting rate-limit-locked by the Home Assistant cloud integration, so I built a local RS485/Modbus feed with an ESP32. Claude did the research, read the manual, wrote the ESPHome YAML, and helped me wire it. Then I forgot which wire color went where.

A growing list of things AI said in my sessions that made me stop and screenshot. Newest at the top.

A growing log of upstream OSS contributions I made with Claude Code that I could not have made without it

A Rust CLI proxy that filters noisy tool output before it hits the context window, saving 60-90% of wasted tokens

Running multiple Claude Code agents as a coordinated team via a native /agent-team skill inspired by Viktor Farcic's dot-agent-deck, which I've tried and am watching as it matures

Patrick Debois (Tessl) argues that context, not code, is the raw fuel of AI coding agents, and lays out a Context Development Lifecycle to engineer it with the same discipline we give code.

Forking Anthropic's Claude Code GitHub Action to work on Forgejo, the adapter pattern, and keeping it synced with upstream

Controlling Arc Browser from Claude Code using the Chrome DevTools Protocol MCP server

A nanoclaw-based chat agent that turns two-letter messages into orchestrated actions across the homelab

Viktor Farcic's AI-powered Kubernetes operations tool — what it does and how I use it for custom prompts

Quick one-liner prompts for AI conversations

Custom AI prompts that encode domain expertise for Kubernetes, CI/CD, and infrastructure tasks

How I use CLAUDE.md files to give Claude Code persistent project context and instructions

How I migrated 191 scattered permission rules to a single, readable config with guidance messages

A two-line status bar for Claude Code with per-session colors, session topics, context/quota bars, and Kubernetes context

AI tools, projects, and articles I'm tracking — some tested, some not yet