跳到主内容
智客 ZICQ

技能库 智客分类:Agent 工作流 review-as-json

Review As Json

运行“/审查”并将结果作为单一结构化的JSON对象发布。 CI使用; " /审查 " 直接用于人脸审查.

233 安装量

官方网址:skills.sh

技能介绍

先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。

做什么

运行“/审查”并将结果作为单一结构化的JSON对象发布。 CI使用; " /审查 " 直接用于人脸审查.

何时用

官方 description 未单独写出 Use when。按规范,代理会在用户任务与这段 description 的关键词匹配时激活本技能。

代理如何加载

按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:Review as JSON。

文件分析

文件分析:这是一份仅含 SKILL.md 的指令型技能,代理激活后整份正文进入上下文。

官方 description(原文)

Runs `/review` and emits the result as a single structured JSON object. Used by CI; use `/review` directly for human-facing reviews.

Review as JSON

· 许可:MIT

来源分类:skills.sh agent-skill

SKILL.md 与 Agent 调用

官方规范 ↗
name
review-as-json
description
Runs `/review` and emits the result as a single structured JSON object. Used by CI; use `/review` directly for human-facing reviews.
许可
MIT
  1. 发现技能客户端向 Agent 提供名称与描述目录。
  2. 匹配与调用用户指定或任务匹配后,载入 SKILL.md 指令。
  3. 按需加载按步骤读取参考文档、使用脚本与素材。

具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗

安装这个技能

Skills CLI ↗

先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。

交给 Agent 安装

复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。

把 Agent Skill「review-as-json」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-9dfb36fc2dcbba5e-%E4%BD%9C%E4%B8%BAJson%E5%AE%A1%E6%9F%A5.html
请存为 .cursor/skills/review-as-json/SKILL.md 或 .claude/skills/review-as-json/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。

GitHub 完整包 ↗

终端安装 · Skills CLI

需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。

npx skills add 'https://github.com/elliottlawson/open-review' --list

npx skills add 'https://github.com/elliottlawson/open-review' --skill 'review-as-json'

CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。

阅读排版

name: review-as-json description: Runs /review and emits the result as a single structured JSON object. Used by CI; use /review directly for human-facing reviews. license: MIT metadata: version: "2"

Review as JSON

Run /review on this change and follow it exactly.

Emit the result as a single JSON object instead of a prose report:

{
  "verdict": "approve | changes_needed | hold",
  "summary": "One or two sentences on the change overall.",
  "findings": [
    {
      "severity": "critical | warning | info",
      "type": "issue | suggestion | question",
      "category": "mission | architecture | implementation | craft | security | performance",
      "title": "Short imperative, e.g. 'Sanitize user input before query'",
      "description": "What it is, why it matters, and the consequence.",
      "file": "path/to/file.ts",
      "line": 42,
      "suggestedFix": "Optional concrete fix."
    }
  ],
  "passes": {
    "mission": "met | missing | different | unclear",
    "architecture": "ok | concern | blocking",
    "implementation": "ok | concern | blocking",
    "craft": "ok | concern | blocking",
    "security": "ok | concern | blocking",
    "performance": "ok | concern | blocking"
  },
  "sectionSummaries": {
    "mustFix": "1-2 sentences on the critical findings, if any.",
    "shouldFix": "1-2 sentences on the warnings, if any.",
    "questions": "1-2 sentences on the questions, if any.",
    "suggestions": "1-2 sentences on the suggestions, if any."
  }
}

Rules for the output:

  • Emit only the JSON object. No preamble, no commentary, no markdown fences around it.
  • Severity. critical blocks merge (bugs, security holes, data loss). warning should be addressed but isn't a blocker. info is a note.
  • Summary. Never restate the verdict ("This PR is approved", "Changes requested") — the presentation layer renders it. The summary answers why.
  • sectionSummaries. Explain the underlying theme ("Missing input validation across the new endpoints"), never just count findings.
  • Type. question is something you need answered rather than a defect. suggestion is an optional improvement. Everything else is an issue.
  • Category. Every finding's category is the pass that found it — this is how the passes show up in CI output.
  • Verify before you flag still applies. Every finding needs a concrete file, line, and consequence. If you can't verify it, make it a question.
  • If there are no findings, emit an empty findings array and an approve verdict.

相关技能

Agent 工作流

Skill Creator

创造有效技能指南。 当用户想创造出新的技能(或更新现有的技能),以专业知识,工作流程,或工具集成来扩展克洛德的能力时,应该使用这种技能.

Agent 工作流

Clawdhub

使用ClawdHub CLI搜索,安装,更新并发布从taladhub.com的代理技能. 需要获取苍蝇上的新技能时使用,将安装的技能同步到最新版本或特定版本,或者发布 npm-instainddhub CLI 的新/更新的技能文件夹.

Agent 工作流

Agent Team Orchestration

管弦乐团多代理团队,任务设定周期,交接协议,审查工作流程. 使用时间: (1)建立2+特派员队伍,具有不同专业,(2)确定任务路线和生命周期(收录框_ spec_建设_审查_完成),(3)在特派员之间制定交接协议,(4)建立审查和质量关口,(5)管理特派员之间的交流和文物共享.

Agent 工作流

Superpowers

Spec-first,TDD,子代理驱动的软件开发工作流程. 当:(1)构建任何新功能或应用——触发脑暴_计划_子代理执行回路,(2)调试出一个bug或测试失败——触发系统性的根起过程,(3)用户说"让我们构建","帮助我计划","我想添加X",或"这个被打破",(4)完成一个功…