Hash the Sampler Output Before You Freeze an Agent-Patch Failure

说明与用途

一份自动化配方，先对采样器输出进行哈希，存入固定文件，然后对代理补丁运行进行分类，以正确判断何时允许冻结。

问题

Agent‑patch failures are hard to distinguish from flake when the sampler output is not hashed and compared, leading to false freeze decisions.

结果

The recipe ensures that only genuine harness flakes trigger freeze requests, preventing accidental freezes due to property misses or fixture drift.

工作流

• Persist sampler bytes before any assertion and hash them with SHA‑256.
• Store the digest beside the run ID and confirm the pin with the standard library.
• Point the property oracle at the fixture file path, avoiding live calls.
• Record transport status in explicit fields (transport_ok, status code, timeout flag).
• Load two run records, classify the pair, and allow a freeze only for harness_flake.
• Leave expiry, merge rights, and other policy decisions outside the module.

证据

Persist sampler bytes before any assertion and hash them with SHA‑256.

来源

The oracle takes a path. It does not call the model, and it does not open a socket to the generator host.

来源

Keep the transport record small and explicit. Suggested fields are transport_ok, a status code, and a timeout flag.

来源

Load two run records. A missing digest stops at unclassified. Different digests stop at fixture_drift and should open a fixture-refresh ticket, not a freeze.

来源