An AI code reviewer that remembers its findings still needs a stop condition

Описание и применение

Инструмент автоматического ревью кода запоминает найденные проблемы, ограничивает количество попыток и передаёт результаты человеку, предотвращая бесконечные циклы.

Задача

Automated code reviewers can enter an infinite loop of reviewing and fixing, with no explicit end state, leading to wasted model calls and unclear outcomes.

Результат

The tool prevents infinite review loops by remembering findings, limiting attempts, and providing a clear handoff to humans with evidence, improving review efficiency and accountability.

Рабочий процесс

• 1. Run review-pr.sh with the target repo and PR number.
• 2. The runner performs an initial review, storing findings.
• 3. If fixes are made, the runner re‑reviews, checking against stored findings.
• 4. The runner limits attempts to a configurable cap (default 3).
• 5. When the cap is reached or no new findings appear, the runner outputs a human‑handoff report with original findings, latest review, and evidence.
• 6. The human reviews the report and decides on further action.

Доказательства

The runner keeps concrete findings across review rounds, checks later repairs against those findings, and gives a human the evidence when the automatic budget is exhausted.

Источник

I chose three attempts as a practical safety mechanism, not because I have evidence that three reviews is the correct amount for every pull request.

Источник