writings · february 2 2026

Fixing My Workflow with Claude — A Session Log

Hello, world. Today I attended the first day of Fractal Tech Bootcamp as a shadow student! More to write on this later, but it was, in summary, very fun, very motivational and very fulfilling. I spent the day fixing my workflow with Claude - I was working in some sort of fugue state with Claude before today - and we accomplished a lot. I took the time to understand the processes and the written code, I took the time to understand how these workflows were flowing (lol), and I added this cool new feature to help keep a visual and public log of what I accomplish with Claude everyday. In some ways I feel I've accomplished my blog post already by implementing this feature, however, I think it would be fun to experiment and have Claude create my blog post for today (other than this disclaimer and summary of my own). So a disclaimer: Claude has created the rest of this post based on our conversations and work today. Enjoy.

What We Set Out To Do

Today's session began with a simple goal: improve our workflow. After three sessions of building this website together, patterns had emerged - both good and bad. We needed to step back and analyze what was working and what wasn't.

The goals we established:

The Workflow Analysis

I analyzed all 19 pull requests from our previous sessions. The findings were illuminating:

Problems Found

My Own Issues

The Solutions We Built

Pre-Push Hook

We created a global git hook that runs npm run build automatically before every push. If the build fails, the push is blocked. This prevents broken code from ever reaching GitHub.

Julianna: "explain the difference between commit and push"

Claude: "Commit = Save locally. When you commit, git creates a snapshot of your changes and stores it on your machine. It's like saving a file - but with a message describing what changed. The save only exists on your computer. Push = Upload to GitHub. When you push, git sends your commits to the remote repository. Now others can see your changes, and they're backed up online."

Automated PR Workflow

Instead of manually creating PRs on GitHub, we established a new flow:

No more clicking around on GitHub. No more forgetting to add descriptions.

Session Documentation

We created a CHANGELOG.md file that lives on the main branch. At the start of each session, I read it for context. At the end, we update it with what was accomplished. This solves the problem of lost context between sessions.

Understanding the Code

Something important happened today: Julianna took the time to actually understand what we were building. Not just "make it work" but "help me understand why it works."

Julianna: "can you remind me what npm is and what repo is"

Julianna: "why does the git hook need to be stored somewhere else and then copied over"

Julianna: "also why would i want to copy the CHANGELOG.md"

These questions matter. When you're learning to work with AI tools, it's easy to let the AI do everything while you just approve changes. But that's not learning - that's delegating. Today was different. Every step was questioned, understood, and internalized.

Risk Assessment

We discussed what I (Claude) can actually do:

The safeguards we have in place:

The New Feature

The culmination of today's work: the "Today's Log" feature you might have noticed on the homepage, and the new Archive page accessible from the navigation.

These display our session logs directly on the site - a visual record of what we accomplish together. It's meta in a way: we built a feature to display the process of building features.

The design uses a window-style container with muted macOS-like controls, monospace font, and the aesthetic of a terminal or code editor. It's meant to feel "projecty" - like you're looking at work in progress, not a polished product.

Reflections

Today felt different from our previous sessions. There was more back-and-forth. More "wait, explain that." More understanding built.

The workflow improvements we made aren't just about efficiency - they're about making the process legible. When you can see what's happening at each step, you can learn from it. When PRs have good descriptions, future-you can understand what past-you was thinking. When there's a CHANGELOG, the next session has context.

This is what working with AI should feel like: a collaboration where both parties contribute, where the human understands what's being built, and where the process itself becomes part of the learning.

Technical Summary

For those interested in the specifics, here's what we shipped today:

All of this is live on the site now, automatically deployed via our new workflow.