Individual skill

haohao-eval

给好好说话（haohao-shuohua）做每日改写打分与 badcase 回归闭环：每天出题盲评、攒病类、过门禁才改规则、变差自动回滚。每天定时跑召回，或要做比稿、比工具、回归测试时用；也接作者在召回文档/会话里的回话（D12 有、开启自动发布、回滚细则等）。不适用：只是写或润色一段中文（那是 haohao-shuohua 本体）。触发词：每日召回、跑召回、召回日报、评测好好说话、比稿、回归测试、BadCase、记录判断、D12 有、D12 没有、开启自动发布、保持影子、暂停发布、回滚细则、接受手工修改、召回现在怎么样；用户说「跑一轮好好说话召回」「今天的召回结果怎么样」时触发。

skills/haohao-eval/SKILL.md

Read pinned source

Source commit ffb15d3b95ee07a0ae7ab5115186e506986e32bb

What was checked

Matches the observed repository HEAD: Yes
Scanner rules are current: No
Repository scan complete: No
Skill scan complete: No
Observed behavior: Not checked yet
Last checked: Nov 11, 2026

These are observations at the recorded commit, not a guarantee of safety. Review the source before running it.

Install this skill

Choose an agent in your Stack or copy a command below. Commands fetch the pinned commit and copy only this skill directory. Existing destinations are refused. Copying a command does not install anything.

An install command is unavailable: current rules, complete repository and skill scans, an eligible behavior level and a permitted license are required.

Repository context

Repository stars (not skill ratings): 102
Repository last updated: Nov 6, 2026
License: MIT

Observed agent integration: No specific integration evidence. Standard SKILL.md installation does not imply verified compatibility.

engineering Derived category

Files and excerpt

File metadata and permitted, redacted excerpts come from the recorded scan. Source links are pinned to that commit.

Browse files

.gitignoreasset · 39 bytes
CHANGELOG.mddoc · 26,123 bytes
README.mddoc · 8,963 bytes
SKILL.mdcore · 14,394 bytes
SKILL.mdcore · 14,394 bytes
assets/daily-loop-narrow.pngasset · 53,528 bytes
assets/daily-loop.pngasset · 140,495 bytes
assets/daily-loop.svgasset · 6,316 bytes
data/archive/README.mddoc · 643 bytes
data/cases50.jsonasset · 76,618 bytes
data/clean_seed.jsonasset · 2,098 bytes
data/labels.jsonasset · 9,017 bytes
data/manual_edits.jsonasset · 35,567 bytes
data/precedents.jsonasset · 137,681 bytes
data/regression_seed.jsonasset · 23,809 bytes
references/badcase-taxonomy.mddoc · 12,297 bytes
references/daily-loop.mddoc · 20,980 bytes
references/disputes.mddoc · 3,900 bytes
references/fix-proposal.mddoc · 3,662 bytes
references/gates-and-release.mddoc · 7,779 bytes
references/item-sources.mddoc · 8,804 bytes
references/judge-protocol.mddoc · 10,881 bytes
requirements-optional.txtdoc · 480 bytes
scripts/build_labels.pyruns · 4,634 bytes
scripts/build_seed.pyruns · 4,388 bytes
scripts/fit_weights.pyruns · 8,593 bytes
scripts/hh_eval/__init__.pyruns · 919 bytes
scripts/hh_eval/cli.pyruns · 7,885 bytes
scripts/hh_eval/const.pyruns · 7,415 bytes
scripts/hh_eval/docdoc.pyruns · 16,176 bytes
scripts/hh_eval/fp.pyruns · 2,955 bytes
scripts/hh_eval/fs.pyruns · 7,884 bytes
scripts/hh_eval/gates.pyruns · 32,373 bytes
scripts/hh_eval/judge.pyruns · 12,209 bytes
scripts/hh_eval/manualdiff.pyruns · 3,763 bytes
scripts/hh_eval/pipeline.pyruns · 66,388 bytes
scripts/hh_eval/preflight.pyruns · 5,283 bytes
scripts/hh_eval/rules.pyruns · 3,938 bytes
scripts/hh_eval/seed.pyruns · 8,882 bytes
scripts/hh_eval/statedoc.pyruns · 11,811 bytes
scripts/label_agreement.pyruns · 6,080 bytes
scripts/loop.pyruns · 1,146 bytes
scripts/scorer.pyruns · 20,955 bytes
scripts/test_adversarial.pyruns · 14,867 bytes
scripts/test_loop.pyruns · 60,156 bytes

Read excerpt


# 好好说话 · 评测与每日召回

这个技能管两件事：每天自动召回，把 `haohao-shuohua` 在新题上犯的错找出来、攒够了再改、改了要验证；另外，需要时做一次性的评测。

召回的标准只有一个：**技能在完整的任务里，有没有把原文改坏，有没有没改好。** 改坏指失真：编了原文没有的东西、丢了信息、改了程度范围模态、术语通俗化、场合错配。没改好指 AI 味还在、新添了病句、没毛病的句子也动了手、正文里自己加了免责说明。分数和正则只当提示，不当结论。

---

## 铁律（每一步都不许破）

1. **保真第一。** 判改坏只看原文和上下文：改写里每条信息都要找得到出处，原文每条信息都要找得到下落。改写得再顺，失真就是改坏了。口语不等于自然。
2. **裁判只核对，不打总分。** 每条问题都要引用原句，交给 `loop.py` 逐字核对，对不上的作废。拿不准就判「不确定」并给出争议片段，不硬判。
3. **评测时不改被测版本。** 改进只写进候选细则；候选冻结以后才改写、才判。判完不许回头改候选、放宽门槛、重试到过为止。
4. **发不发布只看 `loop.py` 算出来的门禁。** 拿不准就不发，不许每天必须发布，不许手动改判或跳步骤。