all 35 comments

sorted by: hot top controversial new old
[–] 0 points 9 hours ago (1 child)

I write my own shell scripts and ruby on rails, but I ask Claude to create branches, manage commits, do all the force pushed to update it all, keep my local copy up to date, do all the things with the git origin etc. So Claude shows up as a contributor but didn’t do any of the work.

I’m fine with that though because I’ve destroyed git history so many times, or messed up commits and broken everything, or needed to do some complex invert paths thing. It’s just better to leave all that to Claude.

  • source
  • hideshow 1 child comment
  • [–] 47 points 1 day ago (13 children)

    It's such a nice treat when I don't see it on a new project, especially on Microsoft's GitHub.

  • source
  • hideshow 13 child comments
  • [–] 64 points 1 day ago* (12 children)

    All it takes to remove Claude attribution is something like this

    git filter-branch -f --msg-filter 'sed "/Claude/d"'
    

    That's before realising things like DeepSeek don't add the trailer

    Be pleased that people are generally not trying to hide it, but also don't trust its absence, look at the commit patterns

  • source
  • parent
  • hideshow 12 child comments
  • [–] 4 points 1 day ago (9 children)

    How does Claude attribution appear in the first place? I've written code with it before and it never shows up in GitHub.

  • source
  • parent
  • hideshow 9 child comments
  • [–] 4 points 1 day ago (7 children)

    Do you let Claude write your commit messages?

    The reason I do is because it makes the agents more capable of working with the code in the future, they put more info into the commit messages than I do and when they look back the commit messages help them navigate the change history better.

    Also, if you see a commit message that just looks whacked, completely missing the actual objective - that might be a Claude hallucination.

  • source
  • parent
  • hideshow 7 child comments
  • [–] [S] 7 points 1 day ago (5 children)

    do you need clauds help in the bathroom too?

    maybe it needs that data about you too to understand you better

  • source
  • parent
  • hideshow 5 child comments
  • [–] -3 points 16 hours ago (4 children)

    You must not be working professionally.

  • source
  • parent
  • hideshow 4 child comments
  • [–] 5 points 7 hours ago* (3 children)

    Working professionals are, if at all competent to be committing code to a professional code base, entirely capable of writing their own commit messages.

    If they can't write the commit message, they can't sufficiently explain in writing the commit they just made to anyone. So, they should not be working professionally on a code base.

    If they are capable of writing the commit message with their own understanding in their own words, but they choose not to communicate their understanding in the commit message? Also disqualifying, they should not be working professionally on a code base. Human communication is essential to this work.

    That this is a controversial position, is a clear sign of the Slopocalypse.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 0 points 6 hours ago (2 children)

    Now I definitely know you haven't seen a real professional codebase. Professional git repos are full of one sentence commit messages with little to no explanation of what's in the change set.

    Humans suck at communication, especially when there's pressure to get product features out the door. AI driven repos are full of beautifully summarized commits with clear explanations of architectural choices, change motivation, etc.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 2 points 2 hours ago* (last edited 1 hour ago)

    Now I definitely know you haven’t seen a real professional codebase.

    Extremely wrong. So much for your assumptions.

    Professional git repos are full of one sentence commit messages with little to no explanation of what’s in the change set.

    Yes, humans can communicate poorly, of course. But people letting the LLMs do it do not understand what is in the change, and the LLM certainly doesn't. There's no one at all to ask because no one understands what was changed or why.

    I'll take “sometimes lazy but they wrote it and their understanding of it can be interrogated”, all week.

  • source
  • parent
  • [–] 2 points 1 day ago

    If I ask an LLM to do something, the commit message is usually just the prompt. If it's for code I wrote myself, my process is to take the message, read it over, then edit appropriately before running git commit myself. It seems that this last step is what does it.

  • source
  • parent
  • [–] 16 points 1 day ago (4 children)

    I'm seeing Claude as a contributor all the time in random repos, but like, how does that even happen? In my experience with "agentic" coding, the agent does not get credit for commits, even if all diffs were made by them. What kind of workflows do people have for Claude to credit itself in a commit?

  • source
  • hideshow 4 child comments
  • [–] 26 points 1 day ago

    Claude has established precedent that Claude credits itself on commits unless you direct it not to. I don't fight it, I would rather people looking back know... Yes, I used this tool - no, it's not "responsible" for the code, I am.

  • source
  • parent
  • [–] 6 points 1 day ago

    Same for me, but that's because I have a rule saying to the claude to not interact with git unless instructed, and I commit manually after reviewing every line of code changed. The bot runs in its own container, isolated from all credentials, so it can't push commits and can't access production. I also hate that the super long commit messages, so I just want to write it myself.

    Basically if you are surrenderinf your entire PC to AI, then you'll not get those commit messages authored by claude.

    All my colleagues instead gave it access to the whole work pc including ssh keys, and the bot can do whatever the fuck it wants with the whole repo on github. It co-authors every single commit. Our master branch pushed straight to prod, and is not protected, and all our team is repo admin anyway. It's complete insanity, bordering criminal negligence.

  • source
  • parent
  • [–] 21 points 1 day ago* (last edited 1 day ago) (10 children)

    It is with great pride that I can say that a GCP-HPC pure-bash application I've been working is completely AI free.

    For the interested, it's something I was working on while bored at work has been blessed as good use of time for the benefit of our dry lab. I'm hoping to be allowed to open source it. It's basically a git-like interface between GCP head nodes and GCP buckets

  • source
  • hideshow 10 child comments
  • [–] 4 points 22 hours ago

    I've been working on a fully open source Obsidian Notes clone for over a year, and now that the decompilation AI tool just dropped I'm not sure people will believe me when I say I didn't use AI

  • source
  • parent
  • [+] -11 points 1 day ago (8 children)

    Congrats, we all had AI free apps 2 years ago, and for the 50 years before that... Keep calm and carry on, what worked then, still works now.

    Or, freak out - you just might consider having an agent like Opus 5.5 review your code and write you a report of its findings... if you don't, others will - that's the new landscape. If you do get "findings" in the report, I'd be curious to hear if any of them were both valid and concerning. I'm sure you won't care about a lot of what it finds, but you might just be surprised by some things it does find.

  • source
  • parent
  • hideshow 8 child comments
  • [–] 8 points 1 day ago (4 children)

    I'll pass on the AI review. I'll QA my work myself, thanks.

  • source
  • parent
  • hideshow 4 child comments
  • [–] -3 points 1 day ago (3 children)

    Honestly, passing on any code review is a pretty thick headed thing to do.

  • source
  • parent
  • hideshow 3 child comments
  • [–] 4 points 1 day ago (2 children)

    I only want the code reviewed by people. That's not the same thing as passing on a review.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 0 points 4 hours ago (1 child)

    That's fine. It is probably worth noting human written code that has been reviewed by 1000s of humans are having valid critical bugs found in them by AI. No one is forcing you to use it but humans aren't magically perfect just because you don't like AI.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 2 points 4 hours ago* (last edited 2 hours ago)

    Any critical security vulnerabilities will be the fault of Google. This app is a glorified filepusher manipulating bash strings into standard path structures for google buckets.

    Edit: to describe more precisely, it's a bunch of bash functions that manipulate a bash string for the commands gcloud storage ls and gcloud storage storage cp. It has some useful functionality beyond that, such as tracking projects and project versions for collaborative purposes, but this is a mask on top of existing commands.

  • source
  • parent
  • [–] 4 points 1 day ago (1 child)

    Opus reviewing an entire codebase is pretty impressive. I use it a lot for some hobbyist stuff and its great at finding bugs, security issues, and just basic stuff like spelling/grammer mistakes.

    But of course, on Lemmy you'll get downvoted because "ai bad"

  • source
  • parent
  • hideshow 1 child comment
  • [–] 4 points 1 day ago

    of course, on Lemmy you’ll get downvoted because “ai bad”

    There was a time that was 100% true, I'm seeing a bit more from different points of view lately.

    24 months ago I wasn't saying "ai bad" I was saying "ai nearly useless."

  • source
  • parent
  • [–] 4 points 1 day ago (2 children)