Category
Four cheaper models auditioned for a seat on my review panel. Twenty-line diffs could not tell them apart. Three-hundred-line diffs could, and also found the bugs in my benchmark.
I was burning Claude session quota too fast, so I added another account. claude-swap auto-switches at 90% while the session keeps running.
Every merge request I open gets read by three AI models that debate each other's findings with code evidence. A deterministic rule, not a model, decides whether it merges.
I wanted everything Claude Code writes for developers to follow the Google developer documentation style guide. A forked skill, a few adaptations, and a local snapshot of all 70 guide pages got me there.
I kept running the same four commands to start work on an issue, so I wrote a script that does it. claude-work gives every issue its own git worktree, branch and tmux session running Claude Code.