I spend way too much time coding, trying out new tech, and writing about the stuff I learn along the way.
A one-line CSS change that nerd-sniped me into a Go profiler, an O(n-squared) heading-ID bug, a syntax highlighter running a 364-way filename beauty pageant, and a benchmark grudge match against pandoc.
An honest tour of what's actually doing the work when you ship a real codebase with an agent — and the very specific way a newer developer would crash and burn trying the same thing.
Tariq Ali wrote that you shouldn't turn HTML-effectiveness into a skill — just prompt for it. I spent a weekend building exactly that skill. A trip report, a quiet argument for visually rich notes, and the moment a workflow folded back on itself and started auditing its own output.
Part 2. I said I'd keep my claude-code-transcripts fork to myself. Then I added five more agents, a real session manager, a destructive folder rename, and eventually a new name. It's not Simon's tool anymore.
I forked claude-code-transcripts to fix a few personal pain points. One of them was a folder with 1,667 duplicate sessions from a runaway loop. Here's the rest.
The gap between 'minimum viable agent' and 'an agent you'd actually use' is where 95% of the work lives. Here's what I found in there.
How a medieval RPG sent me down a rabbit hole that never ended
Why building software for fun, without deadlines or marketing plans, might be the most rewarding programming you can do
My journey from CSS scope hell to static site generation bliss, and why the JavaScript ecosystem needs to chill.