检查

thekinv212026-10-26查看 1254复制 1595

在合并前确认变更。`/check verify` 驱动真实应用来验证行为是否符合规格(满足每一项验收标准,构建每一个界面)。`/check review` 在一个全新的模型上运行资深代码审查,该模型未编写过此代码。在 /develop 之后执行 verify,在 PR 之前执行 review。写入 docs/reviews/,绝不修改代码。

Download (.md)

📘如何使用这个技能

  1. 1
    复制技能内容·点击上方「Copy SKILL.md」复制全部内容,或 Download 下载到本地
  2. 2
    打开你的 AI 客户端·支持 Claude Code、Trae、Cursor、ChatGPT、DeepSeek、Kimi 等任意 AI 助手
  3. 3
    粘贴为系统提示/知识文件·Claude Code:放入 .claude/skills/ 目录;普通对话:作为首条消息发送并说「请严格按这份技能文档执行」
  4. 4
    直接提出你的需求·AI 会按技能定义的工作流、格式和约束来处理,无需再解释背景

💡 提示:技能生效后,每次对话只需一句需求,AI 会按 SKILL.md 的流程执行。

SKILL.md4250 chars
---
name: check
allowed-tools: Bash, Read, Grep, Glob, Write, Agent
argument-hint: [verify | review]
description: "Confirm a change before merge. `/check verify` drives the real app to prove behavior against the spec (every acceptance criterion met, every surface built). `/check review` runs a senior code review on a fresh model, one that did not write the code. Verify after /develop, review before a PR. Writes to docs/reviews/, never edits code."
---

## Output style (plain words, no dashes, no hyphens)

<!-- OUTPUT-STYLE:START -->
Write everything this skill produces, files and messages alike, in plain simple language. Talk to the reader as `you`, warm and direct like a colleague, and present every step as a recommendation they may run or skip, never an order. Keep technical terms that carry real meaning; explain each in plain words. Never use a dash or a hyphen as punctuation: no em dash, no en dash, and no hyphenated compounds. Write `read only`, not `read-only`. Say it in simple words, or reword the sentence. Code, file paths, command flags, and values other skills match on keep their hyphens. Use short sentences, commas, or parentheses. Clear beats clever.
<!-- OUTPUT-STYLE:END -->

## What this skill does

`/check` is the gate before merge. Two modes, separate jobs, usually both, verify first:

- **`verify`** (runtime proof): run the real app and watch the change behave. Proves it works and conforms to the spec (every acceptance criterion met, every specced surface built), which green tests never reveal. Read only on code, no durable files, main thread. Typically after `/develop`.
- **`review`** (fresh model code review): a senior read of the diff on a **different model than wrote the code** (a model reviewing its own output shares its blind spots). Writes findings ranked by severity to `docs/reviews/`. Read only. Typically before a PR.

Neither mode edits code. `verify` points failures at `/debug` or `/develop`; `review` reports findings to fix.

## Pick the mode (route before doing anything else)

First step, before reading any mode file or touching the repo. Look at what followed `/check`:

- **Starts with `verify` (or `run`)** → read `modes/verify.md`, follow it fully. Pass remaining arguments (feature name, scope) through.
- **Starts with `review`** → read `modes/review.md`, follow it fully. Pass the steering through unchanged (e.g. `/check review with opus`, `/check review uncommitted`).
- **No mode word, or ambiguous** (bare `/check`, or a feature name with no mode like `/check auth`) → do NOT guess, do NOT default. Show the two options as a plain text panel and **stop and wait** for the engineer's choice.

**How to present the choice (plain text, works on every agent, no interactive modal):**

Print exactly this, then stop and wait. Do not assume `verify` until they answer. Route on their typed word (`verify` / `review` / `both`).

```
Which check do you want to run? Type one:
  • verify  run the real app and prove the change works against its spec (usually right after /develop)
  • review  a fresh model senior read of the diff, ranked findings (usually right before a PR)
  • both    verify first, then review
```

No interactive picker or modal: a typed choice shown inline behaves the same in every AI tool. (The `argument-hint` frontmatter also surfaces `verify | review` in Claude Code autocomplete; other tools ignore it, which is why this inline panel is the portable path.)

If a feature name came with no mode (`/check auth`), carry it as the target once they pick; still ask the mode.

Do not mix modes in one run. On **both**, do `verify` first, then offer `review` next.

## Portability (any OS, any agent)

Any Agent Skills client on macOS, Linux, or Windows. `git` is the only required CLI. Other shell snippets are POSIX reference, not literal scripts: use your agent's own cross platform file, process, and browser tools. Each mode file adds its own portability notes. No subagent support falls back to inline, noted per mode.

Bundled files: `modes/verify.md`, `modes/review.md`, plus `review-agent-prompt.md` and `review-guide.md` for review. Read only the mode file you routed to; resolve the review bundled files to absolute paths when spawning the reviewer.
来源:skillsdirectory

评论0

登录后参与评论

还没有评论,来抢沙发吧