A Side Effect I Didn't Expect: I Started Sharing My Work
Part 2 of 2: How automating commit messages changed my relationship with sharing
I've been building PiPiece, a Raspberry Pi camera controller with a Vue.js touchscreen UI. A few weeks ago I spent an afternoon encoding my dev workflow into markdown files that Claude Code and GitHub Copilot both read as instructions. One of those files is a /commit skill: when I'm ready to commit, I invoke it and the AI reviews the diff, writes the commit message, and drafts a social post.
Part 1 covers the problem and the six skills. This part is about something I didn't set out to build.
I'm an introvert. I tend not to share what I'm working on until it feels done, which in practice means I don't share a lot. There's always something else to polish.
The Friction Was Small, but It Was Enough
Writing a good commit message is necessary work, but it's slow. Stopping mid-flow to figure out what you just did and why, then doing it again for a social post — that overhead quietly becomes "I'll do it later," which becomes never.
The /commit skill collapses all of that into one step. It's only been running on my last few commits, but already I notice the difference. More usefully, it signals whether something is worth sharing at all. If the AI drafts a post with some energy behind it, that's a hint the change might be interesting to other people.
It doesn't post automatically. I wouldn't want that. The post gets drafted, I read it, I decide. The friction is just lower now.
AI Drafts Need Editing
The /commit skill suggested this series. After I set up the workflows, the commit summary flagged it as "worth a LinkedIn article." So I asked for a draft.
That first draft was good. Better than starting from nothing. But I want to be specific about what I mean by "good": it needed real editing. Not just a pass for typos. It had invented facts.
In one version, it wrote that my wife had picked out her favorite photos from the bird detection runs. She hasn't. It had me coming downstairs to find the camera had already logged six detections. I live in a one-story house. It described birds feeding at the camera at night — my camera isn't sensitive enough for that, and birds don't do that anyway.
Three fabrications in one draft. Plausible-sounding. All wrong.
That's the part most people skip over when they talk about AI writing tools. The output sounds confident and reads well. That's exactly why you have to edit carefully. The AI isn't lying; it just fills in blanks with whatever fits the narrative. If you don't know your own story well enough to catch the gaps, you'll publish someone else's version of it.
The Takeaway from Both Parts
The workflow files took an afternoon. The real work was being able to describe how I develop software clearly enough that an AI would follow it every time. That kind of clarity doesn't just help the AI. It's what you'd put in a CONTRIBUTING.md or explain to someone new on a project.
The tools are capable. They just need to know what you consider done.
This series was inspired by "5 Claude Code skills I use every single day" by Matt Pocock.
PiPiece is open-source. The skill files are in the repo.