My Git history was a mess of 'update' and 'fix' – so I made AI clean it up

Original Article ↗ Hacker News Discussion ↗

Tool’s Intended Use vs Perceived Misuse

  • Author positions the tool as a one-off “rescue” for chaotic, private or early-stage branches full of “update/fix” commits, not for polishing serious main branches.
  • Several commenters fear it will be used to cosmetically “fake” good history, misleading others into thinking a project was well maintained.
  • Some argue that if intent was never captured, AI can only reconstruct “what” changed, not “why,” so it cannot truly restore meaning.

Value and Purpose of Commit Messages

  • One camp: commit messages should capture the author’s intent and reasoning at the time; AI cannot know this and may hallucinate intent, reducing trust.
  • Others: many people barely read old messages; diffs plus a decent natural-language summary are already a big improvement over “fix” and “wip.”
  • Disagreement over primary audience: some write for future self, others for collaborators, some claim they never re-read personal commit messages.

Professional Standards vs Side-Project Freedom

  • Some argue “chaotic side projects” are no excuse; good commit hygiene is a habit that benefits both solo and professional work.
  • Others say side projects are for fun, unpaid, and shouldn’t be burdened with company-style rigor; if you want to write “did stuff,” that’s fine.
  • There’s pushback against moralizing: focusing on pristine history over building things is seen by some as missing the point of hacking.

History Immutability and Integrity

  • Several insist git history, especially on shared branches, should be treated as immutable; rewriting messages risks confusion and undermines archaeology.
  • Suggestions: use git notes to add explanations post hoc, enforce server-side rules against force-push on important branches.
  • Some view bad messages as honest “truth” about how development happened; rewriting them retroactively obscures that signal.

Alternative AI Uses and Improvements

  • Popular alternative: don’t rewrite history; use an explain-commit-style command that generates summaries on demand, benefiting from newer models over time.
  • Other ideas: pre-commit hooks or UI integrations that propose messages from diffs for humans to edit; hybrid workflows where the AI asks clarification questions and suggests splitting incoherent commits.
  • Several want AI-generated messages explicitly tagged (e.g., [LLM]) so readers can interpret them accordingly and distinguish them from human intent.