The Receiving Inspection Workshop: A Gate for AI-Written Code
说明与用途
该工作坊创建了一个接收门,记录每个 AI 生成候选者的 JSON 收据,确保只有经过验证的代码才能进入 trunk。
问题
AI‑written patches can be merged without explicit verification, leading to failures like a nightly seed job crash.
结果
The workshop produced a gate that records a JSON receipt for each candidate, enabling traceability and preventing unverified code from merging.
工作流
- Define invariant: nothing reaches trunk without a receipt
- Create fixture with three test cases for the task
- Build gate that parses, runs tests, records verdict, and stops
- Deliberately break gate to test quarantine logic
- Add property test to catch fake passes
- Yield accounting and transport swap to produce receipts.jsonl
证据
Three agent‑written patches landed on a Friday afternoon. Two were nine‑line diffs, one touched a helper nobody remembered owning, and all three were reviewed in the ten minutes between a standup and a release.
A receipt is one JSON line. It names the task, the model that produced the candidate, the verdict, a machine‑readable reason, and a timestamp.
- 记录日期: 2026-10-16T05:56:04.682Z