I Built a Skill to Auto-Publish to My Blog from Claude Code

Using Claude Code and a cheap LLM subscription to compensate for ADHD while I wait for treatment.

I have ADHD. I want to do so much, but I get lost in the weeds constantly. Half-finished projects everywhere. Ideas that never make it out of my head because by the time I sit down to actually do the thing, I've forgotten what the thing was or I'm overwhelmed by all the steps between "idea" and "done."

Right now I'm waiting for treatment. In the meantime, I've been using Claude Code with a cheap subscription to z.ai's GLM 4.7 model as a sort of external brain. It's been helping me re-jig my entire infrastructure so I can replicate my setup anywhere with just a GitHub API token and an LLM API token—like having a dedicated sysadmin who lives in my terminal.

This blog skill is part of that.

The Actual Problem

My blog is a static Python site. To publish a post, I need to:

  1. Navigate to the blog repo
  2. Create a markdown file in posts/
  3. Add YAML frontmatter (title, date, category, slug, description)
  4. Write the actual content
  5. Run python3 build.py to validate
  6. git add, git commit, git push
  7. Wait for Cloudflare Pages

It's like seven steps. Seven places where my brain goes "that's too much effort actually" and I never publish the thing.

The Solution

Claude Code has a skills system—reusable workflows you can define. I used Gemini 3 Pro to write one that handles my entire blog pipeline.

Now I just type:

> "write a post about X and publish it"

And it does the whole thing.

How It Works

The skill is basically a markdown file that tells Claude Code how to work with my repo:

  1. Sync: Clones or pulls the blog repo
  2. Create: Writes the post with the right frontmatter format
  3. Validate: Runs python3 build.py so I don't break the pipeline
  4. Deploy: Adds, commits, pushes to main

Cloudflare Pages handles the rest.

Why Gemini Wrote It

Could I have written the skill myself? Sure. But honestly, I tried and got distracted halfway through. Gemini 3 Pro just did it.

There's something funny about using a skill written by one model (Gemini) to help me use another model (Claude) to publish posts while I wait for ADHD treatment. The LLM ecosystem is weird like that.

This Post

You're reading the result. I drafted an outline, handed it off to Claude Code with the skill, and it handled the entire workflow. No context switching, no remembering git commands, no "wait did I run the build?"

Friction removed.