Back to Details

check

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

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

输出风格（平实语言，无破折号，无连字符）

<!-- 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 -->
<!-- OUTPUT-STYLE:START -->
本技能生成的所有内容（包括文件和消息）均需使用平实易懂的语言撰写。以“你”称呼读者，语气亲切直接，如同与同事交流，将每个步骤表述为读者可选择执行或跳过的建议，而非命令。保留具有实际意义的技术术语，并以平实语言逐一解释。禁止使用破折号或连字符作为标点：包括em dash、en dash以及带连字符的复合词。请写“read only”，而非“read-only”。用简单的词汇表达，或改写句子。代码、文件路径、命令标志以及其他技能需匹配的值可保留其连字符。使用短句、逗号或括号。清晰胜于巧妙。
<!-- OUTPUT-STYLE:END -->

What this skill does

本技能的功能

/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.
/check
是合并前的把关环节。它包含两种独立的模式，通常会结合使用，建议先执行验证：
  • verify
    （运行时验证）    ：启动真实应用，观察变更的运行表现。验证变更是否正常工作且符合规范（满足所有验收标准，完成所有指定功能模块），而这是普通测试无法完全体现的。仅读取代码，不生成持久化文件，在主线程运行。通常在 
    /develop
    之后执行。
  • review
    （全新模型代码审查）    ：由未编写该代码的其他模型对代码差异进行资深审查（模型审查自己生成的代码会存在盲区）。将按严重程度排序的审查结果写入 
    docs/reviews/
    目录。仅读取代码。通常在提交PR之前执行。
两种模式均不会修改代码。
verify
若发现问题，会引导至 
/debug
或 
/develop
；
review
则会报告需要修复的问题。

Pick the mode (route before doing anything else)

选择模式（先确定模式再执行其他操作）

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.
第一步，在读取任何模式文件或操作代码仓库之前，先查看 
/check
之后的参数：
  • 以 
    verify
    （或 
    run
    ）开头     → 阅读 
    modes/verify.md
    ，严格按照其中步骤执行。传递剩余参数（功能名称、范围）。
  • 以 
    review
    开头     → 阅读 
    modes/review.md
    ，严格按照其中步骤执行。保持引导参数不变（例如 
    /check review with opus
    、
    /check review uncommitted
    ）。
  • 无模式词或参数模糊（仅输入 
    /check
    ，或仅输入功能名称如 
    /check auth
    但未指定模式）→ 请勿猜测，请勿默认选择。以纯文本面板展示两个选项，停止操作并等待工程师选择。
如何展示选项（纯文本，适用于所有Agent，无需交互式弹窗）：
严格打印以下内容，然后停止并等待。在得到答复前不要默认选择 
verify
。根据工程师输入的词汇（
verify
/ 
review
/ 
both
）进行路由。
你想要运行哪种检查？请输入其中一项：
  • verify  启动真实应用，验证变更是否符合规范（通常在 /develop 之后执行）
  • review  由全新模型对代码差异进行资深审查，结果按严重程度排序（通常在提交PR之前执行）
  • both    先执行verify，再执行review
请勿使用交互式选择器或弹窗：输入式选择在所有AI工具中的表现一致。（
argument-hint
前置元数据也会在Claude Code的自动补全中显示 
verify | review
；其他工具会忽略该元数据，因此这种内嵌面板是通用方案。）
若输入时仅提供了功能名称（如 
/check auth
），在用户选择模式后，将该名称作为目标；仍需先询问模式。
一次运行请勿混合模式。若选择 both，先执行 
verify
，再执行 
review
。

Portability (any OS, any agent)

可移植性（支持任意操作系统、任意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.
适用于macOS、Linux或Windows系统上的任意Agent Skills客户端。仅需安装 
git
作为必备CLI工具。其他Shell代码片段仅作为POSIX参考，而非字面脚本：请使用Agent自身的跨平台文件、进程和浏览器工具。每个模式文件都有各自的可移植性说明。若不支持子Agent，则退化为内嵌执行，各模式会对此进行说明。
捆绑文件包括：
modes/verify.md
、
modes/review.md
，以及用于审查的 
review-agent-prompt.md
和 
review-guide.md
。仅读取路由到的模式文件；启动审查器时，需将审查相关的捆绑文件解析为绝对路径。