AI Code Is Cheap. Your Pull Request Still Needs Proof.

Описание и применение

Код, генерируемый ИИ, может выглядеть корректным, но без реального тестирования на провайдерах он может содержать ошибки. FetchSandbox обеспечивает проверку, воспроизводя провайдерские сбои до и после патча, и выдаёт доказательство в виде receipt, которое можно прикрепить к PR.

Задача

AI coding agents generate plausible pull requests, but the integration bugs they fix may not survive real provider failures, leading to unreliable merges.

Результат

Pull requests now include a verification block with a receipt URL, turning the merge decision from “agent says it works” to “evidence shows it works.”

Рабочий процесс

• 1. Agent identifies bug with find_bugs
• 2. Agent proposes diff with fix_bug
• 3. Agent calls prove_fix before applying diff
• 4. FetchSandbox materializes pre‑patch project
• 5. Apply diff to a separate copy
• 6. Run identical provider failure scenario on both versions
• 7. Generate receipt with green_allowed, state, reproduced, verified, receipt_url

Доказательства

FetchSandbox acts as the integration verification layer between the coding agent and the pull request.

Источник

The receipt proves the behavior shown on the artifact: the named scenario ran, the bug reproduced on the pre‑patch code, and the measured invariant held or did not hold after the patch.

Источник