一键导入
claude-skills-workflow-bundle
claude-skills-workflow-bundle 收录了来自 liush2yuxjtu 的 3 个 skills,并提供仓库级职业覆盖和站内 skill 详情页。
这个仓库中的 skills
MR 提交前 4 段式自检 + 默认就地修复:QA → e2e 默认改源码(不 defer 到 follow-up MR)→ diff 展示(代码全量 / 文档仅摘要)→ 冲突默认按 safe-category + intent 自动解(撞模糊才升用户)→ MR body 引用了未跟踪的视觉证据(mp4 / webm / PNG / 草稿 workflow)并对该证据做"能不能 让 reviewer 一眼接受"的质量复盘。**默认行为是动手修,不是报告遗留**。 Use this skill whenever the user says any of: "pre-mr", "/pre-mr", "pre-MR", "prepare merge request", "准备 MR", "MR 前检查", "开 MR 前", "ready to push", "ship MR", "合并前检查", "MR checklist", "push 前自检", "MR 提交流程", "before I push", "before opening MR", or asks to verify a branch is ready to merge into origin/main with diff, conflict, visual-evidence, and evidence-quality gates. Do NOT trigger for plain "diff" or "merge" requests — those are too broad and don't imply the full 4-gate pre-MR pipeline.
一条龙把"当前 git worktree → 远端分支 → 已开 MR"的最后一公里跑完: worktree 探测 → 委托 /pre-mr 4 段自检 → fetch + rebase(默认拉齐) → zh-CN/视觉证据补齐(默认 ≥ 0.30 安全垫) → push --force-with-lease → **GitLab 协议探活(HTTPS → HTTP → OAuth 三段退化)** → 创建 MR → 输出 web_url + 用户待办 + 把产物留在 .agent/pr-assets/ 便于 recall。 **默认行为是动手把 MR 开出来,不是报告剩余项**。 Use this skill whenever the user says any of: "/submit-mr", "submit mr", "open the MR", "ship it", "ship MR", "push my branch", "push 我这条分支", "提个 MR", "提 MR", "搞定 MR", "把这条分支提了", "准备合并", "合并准备", "把分支推上去", "开 MR", "发 MR", "push and open MR", "create the merge request", or asks to take a clean worktree all the way to an opened GitLab MR in one shot. Special-case for winchannel internal GitLab (gitlab008.its.winchannel.net) where HTTPS API often returns EOF and the skill silently falls back to HTTP + $GITLAB_TOKEN. Do NOT trigger for plain "pre-mr"/"准备 MR" — that's the 4-gate self-check sub-step, delegate to /pre-mr instead. Do NOT trigger for "commit"/"git push" with no MR intent.
Full QA loop for the current worktree: diff → host → test → handoff. Runs `git diff` to enumerate every new/changed feature, launches the per-worktree dev stack via `/worktree-self-host`, exercises all features (old + new) through `/playwright-cli`, generates a test report, and finally pops the app open in Chrome for manual inspection. Use this skill whenever the user says "QA this worktree", "test my changes", "smoke test the worktree", "run full QA", "test everything end-to-end", "playtest my app", or mentions wanting to both auto-test AND hand-test their current worktree changes — even if they don't say "worktree" explicitly. Also trigger when the user wants a diff-to-test pipeline, or says something like "diff → host → test → let me try it".