Working notes on Claude.
Things I learned, mistakes I made, and patterns that held up across real projects.
From interactive session to scheduled non-interactive automation — the actual steps, and what breaks first.
Read →Why most prompt libraries get abandoned, and the four-field format that keeps them alive.
Read →A plain-text NOW/NEXT/WAITING file that Claude can reason about — no task app required.
Read →Morning open, work block, afternoon review, end-of-day close — what the loop really looks like.
Read →Three categories, one prompt, and how I tuned it after Claude flagged too much.
Read →Hooks let code run automatically around Claude’s tool calls — before a commit, at session start, after a write.
Read →Claude forgets everything between sessions — here’s the small-file system I landed on after the obvious approach stopped working.
Read →
Patriola's Guide to Claude