test(user-interface): run the write-css three-arm rule eval and settle which rules stay in the body
维护者通常 1 天内回复
还没有人认领这个 Issue。
- #7130 来自 @kyle-sexton —— 已关闭，未合并
评估
调研方向
Start with the SKILL.md body of /user-interface:write-css under plugins/user-interface and its topic reference files, which hold the eight always-on rules and the css-rule-* and css-* eval cases. The claude plugin eval runs must be done from a terminal outside Claude Code, so check the plan's keep-in-body test before starting. Done means every rule has recorded body, reference and no-plugin pass counts on the issue, and SKILL.md keeps only the rules that pass the 4-of-5 and beat-by-2 threshold.
由索引模型根据 Issue 内容生成。
描述
Scope
/user-interface:write-css ships its eight always-on CSS rules (hover, logical, outline, ease-in, color, transition-all, motion, duration) as provisional one-liners in the SKILL.md body, with their detail in the topic references. The plan's keep-in-body test was never run. That test runs each rule case in three arms: the rule in the body, the rule moved to its topic reference, and no plugin. It keeps a rule in the body only when the body arm passes at least 4 of 5 runs and beats the reference arm by at least 2.
The eval could not run from the implementing session: claude plugin eval has to run from a terminal outside Claude Code. The no-plugin baseline was run and is recorded on the write-css PR.
Steps
- With write-css installed from main, run the eight
css-rule-*cases with the plugin, atruns: 5, oneclaude plugin eval plugins/user-interface --case <name>per case. - Make a variant copy of
plugins/user-interfaceoutside the repo with each rule line removed from the SKILL.md body (its reference file unchanged), and run the same eight cases against the copy. - Apply the threshold above and move each rule that fails it out of the body into its topic reference.
- Also run the nine other
css-*cases with the plugin, and record the with/without delta on this issue.
Done when
Each rule's body/reference/without pass counts are recorded here, and SKILL.md keeps only the rules that met the threshold.
🤖 Generated with Claude Code
- 主要语言
- Shell
- 星标
- 22
- 派生
- 2
- 平均合并
- 6 小时 58 分钟
- 30 天内合并 PR
- 951
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
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 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
-
priority: medium status: needs-decision work-class: scoped
难度 2/5 1-3 小时 新手友好度 78/100
melodic-software/claude-code-plugins#9339 · 2 条评论 ·
维护者通常 1 天内回复
-
pipeline-canary priority: low status: ready work-class: mechanical
难度 1/5 1 小时以内 新手友好度 88/100
melodic-software/claude-code-plugins#9263 · 1 条评论 ·
维护者通常 1 天内回复
-
good first issue priority: low
难度 1/5 1 小时以内 新手友好度 90/100
melodic-software/claude-code-plugins#9251 · 1 条评论 ·
维护者通常 1 天内回复
-
area: security needs-triage priority: low
难度 2/5 1-3 小时 新手友好度 79/100
melodic-software/claude-code-plugins#9211 · 1 条评论 ·
维护者通常 1 天内回复
-
priority: low
难度 2/5 1-3 小时 新手友好度 68/100
melodic-software/claude-code-plugins#9202 · 1 条评论 ·
维护者通常 1 天内回复
查看 melodic-software/claude-code-plugins 的全部 Issue
相似的 Issue
-
Python Volcengine provider returns failures instead of raising them可能已有人在做 @Ricardo-M-L 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 79/100
agent-infra/sandbox#248 ·
-
chore priority:low size:S
难度 2/5 1-3 小时 新手友好度 74/100
celo-org/celopedia-skills#98 ·
维护者通常 1 天内回复
-
Git for Windows shouldn't be locked on their `.windows.1` versions可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭git windows
难度 2/5 1-3 小时 新手友好度 78/100
jenkins-infra/packer-images#3179 ·
维护者通常 1 天内回复
-
ready-for-pr
难度 2/5 1-3 小时 新手友好度 72/100
kunchenguid/firstmate#7679 · 1 条评论 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
MISP/misp-docker#478 ·
维护者通常 1 天内回复