The commit message was the whole instrument panel
I asked an agent to diagnose a failing pipeline and leave it broken. It fixed it, then wrote a commit message about something it hadn't done, which bothered me a lot more than the fix.
Categories
89 pages
go-tool-base
The underpinnings for any Go application that ships as a binary: command-line tools to start with, and (as more than half of what is built on it now proves) long-running services and local web UIs too. It standardises the boilerplate and wiring you end up writing in every Go project, and puts config, logging, self-update, a DI container and the rest straight at your fingertips. Highly opinionated, and it makes some calls people will disagree with, but it works, and it works really well. The CLI it ships with is a big part of that: keeping a Go app built, updated and maintained becomes almost trivial.
Rotating a signing key without breaking self-update
Signing releases is the easy day. Replacing the key you sign with, while every installed copy is still checking against the old one, is a job I went looking for a guide to and couldn't find.
A comment is for a different reader
The tool had comments on under ten per cent of its lines and ninety pages of documentation built into the binary, and I'd argue that was one decision made for two very different readers.
Eighty-one megabytes of hello world
How big is a tool that does nothing? For go-tool-base it was 81 MB, and a comment in the code said it shouldn't have been.
The library that appointed itself signal handler
A commit landed a month before I noticed it. By the time somebody told me, two of my own tools had already written workarounds for it without mentioning it.
What happens when an AI burns out
A Claude session ran for two days, stopped listening and got my hand-written code backwards. So I took it off the job, and asked for a proper handover first.
The bill for fifty-one modules
Change one line in the tls module and you've signed up for twelve releases. That's not a bug in the architecture, that's the architecture.
The extraction playbook only got real when chat left
I wrote the process document before doing the hard one. Then the hard one went through it and rewrote a step.
Prove the bug before you fix it
A model handed a bug will hand you back a plausible fix and a green test suite in about a minute. That tells you almost nothing.
The packages that didn't want to leave
Everyone writes up the modules that came out. The decisions that mattered were the four packages I decided to leave exactly where they were.
