Catch the accessibility failures that ship in almost every AI-built UI. Use after building any interactive component.
中文简介正在准备中。当前显示原文说明。
Archive228/loopkit7552026年7月15日 更新
免费在 GitHub 公开
my-new-skill
One sentence that describes WHEN to invoke this skill. Write it as a trigger phrase, not a summary — include the words a user would say ("fix bug", "add feature", "review diff"). The agent routes on this field.
中文简介正在准备中。当前显示原文说明。
查看安装方法包含的文件（3）
- SKILL.md2.5 KB
- evidence/README.md527 B
- POST.md1.3 KB
SKILL.md（原文）
安装前可以查看交给 AI 代理的指令内容。
My New Skill
<!-- Anthropic-shaped SKILL.md template. Copy this folder to skills/<your-name>/ and edit. Keep the body under ~150 lines — long bodies are a context tax on every turn where this skill is loaded. Recommended structure: 1. One-paragraph statement of the failure mode this skill exists to prevent. 2. A short checklist or numbered procedure the agent should follow. 3. Red flags — the shortcuts an agent takes when it wants to skip this. 4. Evidence — a required before/after pair proving the skill actually changes behaviour. Ships in the sibling evidence/ folder. 5. (optional) `references/` sibling files with deeper material the skill lazy-loads only when needed. Do not include: - Marketing copy. - Duplicates of other skills (link, do not copy). - Project-specific rules — those belong in .claude/rules/. Frontmatter contract (validated by scripts/validate-skills + CI): - `name` — kebab-case, unique across the repo, matches the directory name. - `description` — one non-empty line, phrased as a trigger. - Anything else is optional and not currently enforced. -->Without this skill, the agent tends to <failure mode>. This skill enforces <the alternative>.
Steps
- <step>
- <step>
- <step>
Red flags — STOP
| Thought | Reality |
|---|---|
| "I've done this before, I don't need the checklist" | Skills evolve; re-read the current SKILL.md. |
| "This case is simple" | Simple cases become complex mid-task. |
| "The user didn't ask for this rigor" | The user asked for a working result. Rigor is how you get one. |
Evidence
Every skill ships with a real before/after pair proving it changes agent behaviour on a concrete task. Without this, the skill is a claim, not a tool.
evidence/before.md— a verbatim snippet of what the agent does on a representative task WITHOUT this skill loaded. Include the prompt, the agent's response, and what went wrong (or what got skipped).evidence/after.md— the same prompt with this skill loaded. Show the changed behaviour and why it is better.
If you cannot produce a before/after pair, the skill is not ready to ship. Land it as a draft in a branch, gather the evidence, then merge.
References
references/<topic>.md(only if needed; keep the body lightweight)
评价
还没有评价。欢迎分享你的使用感受。
评论
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
同一仓库的技能
Before compaction Loopkit extracts decisions into claude-decisions.json (machine-readable). Read it alongside claude-progress.txt at session start — prose is for humans, JSON is for the loop.
中文简介正在准备中。当前显示原文说明。
Archive228/loopkit7552026年7月15日 更新
Review a diff against the goal spec assuming the code is BROKEN. The reviewer that lives in the maker's head always agrees with itself — this pulls review into a hostile, separate pass. Invoke after every code change before marking work done.
中文简介正在准备中。当前显示原文说明。
Archive228/loopkit7552026年7月15日 更新
Verify that an endpoint checks ownership, not just authentication. Use on any handler that reads or mutates user data.
中文简介正在准备中。当前显示原文说明。
Archive228/loopkit7552026年7月15日 更新
Find the exact commit that introduced a bug. Use when something worked before and broke, and you don't know which change did it.
中文简介正在准备中。当前显示原文说明。
Archive228/loopkit7552026年7月15日 更新
Before picking new work, smoke-test the last "completed" feature. If it's broken, revert and re-open it before touching anything else. Kills the "looks shipped, isn't shipped" bug across sessions.
中文简介正在准备中。当前显示原文说明。
Archive228/loopkit7552026年7月15日 更新