// lab
I build my own tools.
Open-source tools and experiments. I use them on my own products first.
// is it proven?
trustbut
Your agent says it’s done. trustbut checks whether anything proves it.
When an agent says “all tests pass”, that’s the agent grading its own work. trustbut reruns the tests in a clean copy and checks that each new test fails without the change. If a test passes either way, it doesn’t prove anything. No AI decides the verdict.
Measured on 201 real merged pull requests, and run on DogBase’s own repos.
// did it break a rule?
Holdline
The rules your business runs on, held even when an agent writes the code.
Agents are good at keeping tests green, and good at quietly breaking a rule nobody wrote down: who may move a record between teams, which rows a user may read. Holdline finds those rules in your code, asks you to approve them in plain words, and checks them on every change.
It catches every rule broken on purpose in its test app.
// do you still understand it?
Apprentice
Keep understanding the code Claude changes, one question at a time.
Writing code is no longer the slow part. Understanding it is. Apprentice maps how much of your repo you still know, and after Claude edits something, asks one short question about it. Get it right and that part of the map lights up again.
It runs inside Claude Code, on your machine, with no account.
// also building
More from the lab
- n8n content & lead pipelineSelf-hosted n8n workflows that collect leads and content signals into daily digests, replacing per-seat SaaS.Available
- DogBase AI co-pilotAn agent for working-dog teams that answers questions from a unit’s own training and deployment records.In development