About this template
One safe fix a night. One draft PR. You merge.
If you want your codebase a little cleaner every morning without letting an agent loose on critical code, this template is built to your exact level of paranoia.
The design principle is deliberate restraint. Each run, this skill hunts for exactly one antipattern with a cheap check behind it: a lint error, a type error, an unused import. It fixes that one thing and opens one small draft PR. If nothing clears the safety bar, you get a "nothing safe" report instead of a risky guess.
It never touches auth, billing, or migrations. Those paths sit on a vault denylist, and you control the list.
How the overnight code cleanup works
- Fork the template, sign in to Claude Code, and export a
GH_TOKEN. - Paste:
Find one lint fix. Draft a PR. Do not merge. - Review the tiny PR with your coffee. Merge it or close it.
Run it while you watch first. Schedule it nightly only after you trust the denylist on your own repo.
What you get when you fork
- The
nightly-one-antipatternskill, capped at one fix per run - A vault denylist pre-seeded with the high-risk paths: auth, billing, migrations
- A demo prompt to run under supervision before you schedule anything
- Your own copy in an isolated Firecracker microVM on your subdomain, working nights whether your laptop is open or not
What you connect
Bring your own keys; they stay inside your VM:
- Claude login or
ANTHROPIC_API_KEY - A
GH_TOKENso it can open draft PRs
Secrets are scrubbed from the template. You supply your own.
Who it's for
You, if you want compounding code quality from small, boring, reviewable PRs.
Not you, if you want an agent refactoring critical surfaces unattended. This one is intentionally small and intentionally slow.
Fork it
$25+/mo, 30-day money-back, no free trial. Fork it - your copy boots in about 3 minutes, and its first tiny draft PR is one supervised run away.
Part of the Software Factory collection.