chore: audit and slim the ux-test skill

未关闭
  • 哪里不对？
#431 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
4/5
预计耗时
3-5 天
新手友好度
35/100
Issue 类型
重构
描述清晰度
描述清楚
活跃度
活跃
技术栈
csharp, docker
领域
testing, tooling

调研方向

Start by reading .claude/skills/ux-test section by section alongside the simplicity test from issue #400, then record keep or cut decisions here before editing. Check how the allowed-tools entries handle the two docker compose variants and how the gh recipes manage result comments. Done means the file reflects the audit, a real /ux-test PR#N run completes the stated stack and cleanup steps, and a second run updates the existing results comment.

由索引模型根据 Issue 内容生成。

描述

tech-debt

Found while auditing the pipeline for #400 (piece 6). The ux-test skill is the largest single file in .claude/skills/ and has not had the simplicity pass the rest of the workflow is getting. This issue audits it, one section at a time, and then trims it.

What I found

  • It is 307 lines, and its allowed-tools header is about 48 lines of its own. Sixteen of those are the same eight docker compose commands written twice, once for the plain test file (variant B) and once with the local TLS-proxy override (variant L).
  • Steps 1 and 9 carry gh recipes for reading the PR or issue and for finding, updating and creating the marker comment.
  • Step 4 polls with up to 40 separate Bash calls and a Playwright wait between them, because the skill forbids a shell sleep or loop. A single retrying curl may do the same job.
  • Step 5 holds navigation hints (routes, filter behaviour, which task sections start collapsed). These describe UI that changes, so they go stale, and Claude can read them from the code.
  • The report template (Step 8) and the guardrails section restate rules already in the steps.
  • Two lines (Always --body-file and PascalCase enum values) are removed by #400, because pieces 1 and 2 cut those rules everywhere. This issue covers everything else.

Scope

  1. Walk the file one section at a time and apply the standing test from #400: a line stays only if Claude would get it wrong without it, Claude cannot find it in the repo, and a hook, setting or code comment cannot enforce it instead. Record the keep and cut decisions for each section on this issue before editing.
  2. Decide how the two compose variants are handled so the allowed-tools list no longer duplicates every command, or document why it must.
  3. Decide what happens to the gh recipes (the skill is standalone and posts its own results, so the recipes may stay in a shorter form).
  4. Trim the file to match the decisions.

Acceptance criteria

  1. Each section of ux-test has a recorded keep or cut decision on this issue before the file is edited.
  2. allowed-tools has no duplicated command pairs, or a comment in the skill says why a pair remains.
  3. A real run, /ux-test PR#N against an open PR with a UI change, still builds the taskflow-test stack, runs the plan, posts the results comment, tears the stack down and leaves .playwright-mcp/ clean.
  4. A second run on the same PR updates the existing <!-- ux-test-results --> comment instead of adding another.
  5. The live stack and its volumes are untouched (every docker compose command still carries -p taskflow-test).

Out of scope

  • Running ux-test from the feature-factory pipeline. #400 decided it stays manual: after the PR is created, the pipeline prints /ux-test PR#N when the UI changed. Revisit only after this audit.
  • Changing the test stack's compose files.
  • Moving ux-test into a plugin or rewriting it as an agent.
主要语言
C#
星标
1
派生
0
平均合并
2 小时 2 分钟
30 天内合并 PR
98

环境准备

  • 提供 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。

nevridge/TaskFlow 的其他 Issue

查看 nevridge/TaskFlow 的全部 Issue

相似的 Issue

更多 C# Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。