¬ just serendipity 🍀 (not just serendipity)

Problem statement: Claude Code likes to declare “done” without running tests. Said no overconfident LLM ever.

prove_it by Justin Searls is a config-driven hook framework that intercepts Claude’s lifecycle events (SessionStart, PreToolUse, Stop) and runs whatever tasks you configure — tests, linters, AI reviewers — blocking the Stop until they pass.

I like this concept in spirit.

Installation:

% brew install searlsco/tap/prove_it
% prove_it install                    # one-time, registers global hooks at ~/.claude/
% cd <git repo> && prove_it init      # per-repo config + script stubs + git hooks; comparable to `pre-commit install`
% prove_it doctor                     # comparable to `brew doctor`

I (finally!) tried it today and concluded that it’s not for me.

That said, I really like the ideas that Justin developed there. I’ll look into adopting some of them into my global CLAUDE.md.

See the README.md for documentation.