Set up Claude Code hooks for formatting, tests and protected files

Instructions in CLAUDE.md are advisory; hooks are deterministic. One user on blocking a folder: "a hook that exits with an error when Claude tries to write there will stop it reliably. Much cleaner than hoping your instructions hold."

In .claude/settings.json, add a PostToolUse hook with matcher Edit|Write that runs Prettier on the edited file; the docs' command is jq -r '.tool_input.file_path' | xargs npx prettier --write. One user's reason: "Claude's style drifts in long sessions."

Add a fast test command as another post-edit hook. "Claude makes a change, tests run automatically, Claude sees the output and adjusts. It closes the feedback loop."

Add a script in .claude/hooks/ that reads the file path from stdin and exits with code 2 for patterns such as .env, package-lock.json or .git/, and register it as a PreToolUse hook on Edit|Write. Exit 2 blocks the action and sends your message back to Claude. "Only pre tool use can actually stop Claude."

Add a Notification hook (or one on Stop) that fires a desktop notification. One user includes the exit status "so I know immediately if it finished clean or hit an error."

Run /hooks to confirm each hook is listed under the right event. .claude/settings.json can be committed so the team shares it; .claude/settings.local.json stays personal.