Hacktoberfest 2026:维护者为十月标记出来的 issue,仍然开放、适合新手。 浏览 Hacktoberfest issue

fix(evals): give the methodology rewrite answer a template whose target slot needs a stated basis

未关闭 适合新手
#10,217 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
72/100
Issue 类型
文档
描述清晰度
描述清楚
活跃度
活跃
技术栈
python, shell

调研方向

阅读 plugins/evals/skills/methodology/SKILL.md,尤其是“Rewrite this criterion”的回答以及第42–52行附近现有的指导。将该回答替换为所要求的填空模板,在 .changes/evals/ 下添加一个片段,然后运行列出的两个验证命令;prompt 和 grader 文件应保持不变。basis: 检查通过、两个验证器均以退出码 0 结束,并且 CI 在 PR head 上为绿色,即表示完成。

由索引模型根据 Issue 内容生成。

描述

priority: medium status: ready work-class: scoped

Parent: #5973 (slice 1 of 2). Runs first; slice 2 waits on it.
Runs in: lane. Skill text and an offline grader check only; no model run.

Problem

In the measurable-criterion case (plugins/evals/evals/measurable-criterion/) the evals plugin passes the calibrated four-properties grader in 2 of 3 with-plugin runs, on two VALID runs (2026-10-13 and 2026-10-24). The failing reply sets "at least 99%" for red-flag escalation and then says the number comes from severity. The hub already says severity is no basis (plugins/evals/skills/methodology/SKILL.md:42-45) and that the rewrite carries a grounded target (:46-52), and a one-line wording addition tried on 10-24 did not move the result. The model states the rule and still keeps the severity-based number.

Decision (taken from the 10-24 data, not escalated): option (a) of that comment. The rewrite answer gets a template whose target slot can only be filled from a measured baseline, a prior result, a benchmark or expert review, or else the baseline-relative form, so a severity-only number has no slot. Basis: the failing reply named the rule and broke it, which points at the answer's shape, not missing knowledge (#5973 row of the backlog HITL batch).

Fix

In plugins/evals/skills/methodology/SKILL.md, replace the prose of the "Rewrite this criterion" answer with a short fill-in template: quality, number or scale, trial set and size, target, and a basis: slot whose allowed values are exactly the four sources above or the baseline-relative form ("at most half the current rate over N trials"). Say a target with no allowed basis is written in the baseline-relative form, never as a placeholder. Do not change the case prompt or the four-properties rubric. If the change makes any four-properties calibration sample stale, update the sample and say so in the PR (slice 2 recalibrates). Add a .changes/evals/ fragment.

Done when

  • grep -c 'basis:' plugins/evals/skills/methodology/SKILL.md prints 1 or more, inside the "Rewrite this criterion" answer
  • git diff --name-only origin/main...HEAD -- plugins/evals/evals/measurable-criterion/prompt.md plugins/evals/evals/measurable-criterion/graders/ prints nothing
  • python3 plugins/evals/skills/validate/scripts/validate-cases.py plugins/evals/evals exits 0
  • bash plugins/skill-quality/scripts/check-skill.sh plugins/evals/skills exits 0
  • ci-status is green on the PR head

Split by the backlog HITL session (2026-10-25, owner rule Q93: split large work before a run picks it up). This was generated by AI.

主要语言
Shell
星标
22
派生
2
平均合并
2 小时 48 分钟
30 天内合并 PR
939

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

melodic-software/claude-code-plugins 的其他 Issue

查看 melodic-software/claude-code-plugins 的全部 Issue

相似的 Issue

更多 Shell/Bash Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。