Improve commit messages

[Thanatermesis]

Issue

AI is very good for automate coding, but is very important "to know what he did", for that we normally read the code, but this is not always clever... I find it extremely important to mark the commits with a very good description detailing what he did, why, and how fixes the issue...

How to do that in aider? maybe we can have an extra parameter to describe commit messages, this can be accomplished with the weak model, what I use is basically paste "all that he did" (so, my request, + its thinking + its final patch modification) and ask to analize what happened to create an improved detailed commit message.

My (draft) prompt:

analize this aider (cli coder) output and write me a git commit message to add with "git commit --amend". So describe what was the issue or need, why this modification, and how this modification fixes or improves it.: [CONTENTS]

So this is an example of commit result:

Before:

    fix: fallback to Return-path header for unsubscribe recipient detection

After:

    Fix: Extract recipient from Return-path for unsubscribe emails without body address

    The unsubscribe handler was only looking for email addresses in the message body. For emails that use `X-Apple-Unsubscribe: true`, the body often contains no email address (only a confirmation message), so `find_email_addresses()` returns empty and `recipient` stays blank. This prevented the system from identifying which subscriber to unsubscribe, causing the request to be ignored.

    Now, when an unsubscribe is detected, we first try to get the recipient from the `Return-path` header (the envelope sender, who is the one requesting the unsubscribe), and fall back to scanning the body only if that is empty. This matches the approach used for autoresponses and ensures Apple-unsubscribe requests are processed correctly.

Now, the person who is going to read this commit (me, to understand it, to search a future introduced-bug, or to send this commit to a reviewer or the original author of a source code), is able to understand perfectly why this modification was made, how improves the code, etc.

NOTE: using my method consumes of course an extra AI call, with the amount of tokens of the patch + my request + its thinking, which is not much, but definitievely worth. OR, instead, maybe the actual commit message can be simply improved to do this, or maybe we can have a --long-commit-description parameter that includes a description/body in the commit message (and not only a title)

[Thanatermesis]

After to experiment a bit, seems like I can use the parameter "--commit-prompt", not sure if it includes (knows) all the data given like the "thinking" result or not but the result seems to be pretty decent.

I used this prompt for it:

--commit-prompt="You are an expert software engineer that generates detailed Git commit messages based on the provided diffs. Review the context and diffs carefully. Generate a commit message with a single-line title and a detailed description body separated by a blank line. The title must follow the format : using these types: fix, feat, build, chore, ci, docs, style, refactor, perf, test. Ensure the title starts with the appropriate prefix, is in imperative mood, and does not exceed 72 characters. After a blank line, provide a thorough description body that explains what changes were made, why they were necessary, how they improve the codebase (performance, readability, bug fix, new feature, etc.), and any non-obvious design decisions, trade-offs, or alternatives considered. If helpful, include ASCII art, code snippets, or diagrams to illustrate the change. Format the message properly: wrap lines at 72 characters, use blank lines between paragraphs, and write in clear developer-friendly English without markdown code fences. Do not use single or double quotes in the message. The message must be self-contained so future developers understand the commits purpose and impact. Reply only with the commit message (title, blank line, body) without any additional text before or after."

[dwash96]

Yeah, so it actually already works how you're describing. The commit history gets the file diffs of files edited in the session, the entire conversation history, and your updated prompt else the default prompt which is:

commit_system = """You are an expert software engineer that generates concise, \
one-line Git commit messages based on the provided diffs.
Review the provided context and diffs which are about to be committed to a git repo.
Review the diffs carefully.
Generate a one-line commit message for those changes.
The commit message should be structured as follows: <type>: <description>
Use these for <type>: fix, feat, build, chore, ci, docs, style, refactor, perf, test

Ensure the commit message:{language_instruction}
- Starts with the appropriate prefix.
- Is in the imperative mood (e.g., \"add feature\" not \"added feature\" or \"adding feature\").
- Does not exceed 72 characters.

Reply only with the one-line commit message, without any additional text, explanations, or line breaks.
"""

[dwash96]

I'm a bit remiss to change the default because your prompt does encourage some really verbose outputs depending on the model generating the message. Does --commit-prompt accomplish what you're going for given that it does include the conversation history as a part of the context of generating a commit?