The result is a maintained product
Tw93’s archived account of developing Mole focuses on keeping a growing codebase reliable: humans define architecture and boundaries, regressions become tests, causes inform rules, and repeated checks become Skills. The public project provides a concrete product to inspect. It does not establish which model wrote a given part of the code. Historical code and test counts are deliberately not presented here as current measurements.
Supporting documentation (opens in a new tab)Turn a complaint into a regression test
Our suggested experiment is to choose one bug in your own project that you can reproduce without customer data. Describe the visible wrong result and write a check that fails for that reason. Keep the check while fixing the behavior. Try the nearest related path as well: a repeated click, an empty input, a stale response or a narrower screen. We are proposing a small practice exercise, not claiming to have rerun Mole’s tests.
Put the lesson in the right place
A test should preserve an observable requirement. A short rule should explain the boundary or surprising cause. A reusable Skill should describe a check that is genuinely repeated. A work log should retain the particular change and evidence. Do not copy every chat message into all four places: contradictory instructions are another maintenance problem. This division is our editorial interpretation of the method.
A useful acceptance record
Keep the original failure, the revised output, the check that now passes, the adjacent path you inspected and what remains untested. Let a reviewer verify the result without replaying the whole conversation. If a check only searches for the exact code you just wrote, it may be measuring implementation shape instead of user behavior.
Boundaries before trying software
A maintenance lesson is not a recommendation to run a cleanup tool on important files. Review the current product documentation and permissions separately. This page does not distribute Mole, its screenshots or a branded Skill, and it does not claim that any model can safely maintain an arbitrary codebase without human review.