Show HN: Build the habit of writing meaningful commit messages

Conventional Commits and metadata

• Strong disagreement over enforcing Conventional Commits (feat/fix/chore, etc.).
• Critics: the type prefix is low‑value noise that occupies the most important part of the subject line; they care more about a natural “what/why” sentence, scopes already appear organically, and bug‑hunting is better done with blame/bisect or issue IDs.
• Supporters: the type/scope conventions aid scanning, filtering, enforcing atomic commits, and building changelogs (including with LLMs). They argue trailers are under‑surfaced in common UIs, so prefixes are more visible.
• Some dislike specific labels (e.g., “chore” as value‑judging work) or the spec’s MUST/SHOULD tone, but others treat it as a flexible convention to adapt.

Value and role of commit messages

• One group sees detailed commit messages as pedantic, preferring to optimize for coding speed, squash merges, WIP messages, or just ticket numbers; many say they almost never read history.
• Another group relies heavily on history (git blame, editor integrations) to understand intent years later, arguing that even if only ~2% of commits are re‑read, the payoff is huge.
• There’s tension between documenting in commit messages vs. in code comments, ADRs, or issue trackers; some advocate linking commits to tickets as a mutable context store.
• Several emphasize that commit messages should explain “why” more than “what”, and that good habits around atomic commits make messages simpler and more useful.

AI‑generated commit messages and this tool

• The tool is praised for caring about commit quality and for asking the developer questions about “why” instead of blindly summarizing diffs.
• However, example commits from the repo drew strong criticism: overly verbose, generic, marketing‑style language; repetition of what’s obvious from the diff; weak or even incorrect rationales; and missed opportunities to split changes into smaller commits.
• Concern: providing a long AI draft biases people to accept “good enough” fluff rather than think carefully; some would prefer a terse human one‑liner to paragraphs of AI text.
• Suggestions include: use AI to critique and tighten human‑written messages, aggressively prompt against filler/weasel words, and focus on helping people learn to write, not avoid writing.

Broader concerns and resources

• Some worry that delegating commit writing will erode developers’ communication skills and detach commit history from human reasoning, making both human and future AI understanding worse.
• Others view commit writing as a chore that LLMs are “very good” at and are happy to offload.
• Multiple commenters link to guidance on good messages (Google and Zulip commit/CL description guides, essays on theory‑building and signs of AI writing) and exemplary real‑world commits as better models than LLM‑style prose.