How I use Claude to automatically create new blog post ideas based on last week commits

· 3 min read

Figuring out what my next blog post will be is always a hassle, and mostly because of my inability to remember to take notes of my work, of what I learn, of what I do, basically if something happens and I don’t write a post about it almost instantly, I never will.

These past few months I’ve been working on 5 personal apps (3 already in teh App Store) and luckily I’ve set them all up to be very similar in repo structure, a todo file that has the high level overview of the task list, then a tasks/ folder, where each big task has its own file with description and things I need to keep in mind, and a lessons section in the Claude.md where Claude automatically write lessons when something crashes, debugging is taking to long, or I call out why something shouldn’t be done in a specific way.

These additions have helped me with the first part, taking notes about what I do, and they’ve been live for months now, but sitting there, unused.

Over the weekend, I had the idea of copying all of the lessons into claude and generate a list of possible posts, so that I could pick one, and then write about my experience with it.

It worked well, which left me with another question, can this be automated? And yup, it sure can, I created a skill where now, whenever I remember, I can just type in any claude session /newsletter-ideas and it will do these steps in order:

It scans my apps/ folder, where my real apps live, not the tests, experiments, or old projects, just the currently working apps. From there it looks at the lessons, tasks, and claude files to see what’s changed since the previous scan.

In priority order it grabs the lessons first, since each of those is a debugging story, or a gotcha moment, then TODO/tasks to review completed milestones and open gaps (which helps me identify tradeoffs) and the recent git log, to catch anything that didn’t make it onto those files.

It uses subagents to scan the apps, while the main session stays clean to review what the agents bring, which is really two things:

First, verifies it actually happened, and it’s not just a write topic (I can’t write about stuff I don’t do, it feels icky).

And second, it verifies that it fits the content I wanna write about: solo shipping around my day job, AI workflows, App Store and Play Sore gotchas (As a side note, I don’t know when it became this fucking hard to release on Google Play, but more on that on another post), stuff like generic listicles are explicitly declared worthles.

Once all of that is done, it goes into my blog project, and writes to the backlog file, so that whenever my scheduled time to write an article comes, I now have a list of things I can pick from, instead of trying to figure out, hey, what was it that I wanted to write about.

There are a bunch of other things, and rules it has that ight be cool to some of you, so you can check the actual skill, I published it on Github.