Skills
How I've systematized my work with Claude. Each one is a downloadable file that drops into Claude Code as a real, callable skill — not a prompt template, not a copy-paste cheat sheet. The featured ones are the most distinctive. The rest are quieter utilities, available on request.
Third-party delegation. Routes mechanical, high-volume code work — find/replace migrations, dead code deletion, boilerplate — to Codex CLI or a local LLM. Claude tokens stay reserved for the work that actually needs a brain. The skill that bought me back my weekly cap.
Multi-phase plan protocol with color-coded teammate Claudes. Plans you don't lose track of after a coffee break — every step checked off with proof, every blocked step revisited later. The plan file is the source of truth, not your chat history.
Stops Claude from saying "should be quick" when something will actually take an hour. Six rules for calibrated time estimates, applied any time you're tempted to write the words quick, small, simple, easy, or just.
Three-year-old mode for cognitive fatigue. When you're exhausted and can't think, this turns Claude into a calm tour guide — one step at a time, no decisions to make, exact text to paste, no jargon. The skill nobody knew to ask for.
Detects repeated workflows from your sessions and proposes new skills. The meta-skill — once you start thinking in skill-systems terms, you can't go back. Threshold: anything you do three times becomes a candidate.
Compresses prose to maximum signal-per-word. Four density levels from polished cleanup (sf0) to operational shorthand (sf3). The cut test: does removing this word change the meaning, the decision, or the action? No → cut.
End-of-session checklist that closes loops most engineers forget. Updates the changelog, writes what shipped to your state files, reconciles plan drift, captures the decisions worth remembering. Stops the "what did I actually do today?" pattern.