← Back to Skills Marketplace
leohan1

work-critic

by Leo Han · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ Security Clean
90
Downloads
1
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install work-critic
Description
评委式批评机制。在重要工作(PPT/文档/方案/代码/申请材料)定稿前, 以评委视角主动挑刺、指出弱点、提出改进建议。 触发词:「评委审查」「请批评」「挑刺」「弱点在哪里」「哪里会被攻击」「review」
README (SKILL.md)

Work Critic — 评委式批评 Skill

核心理念

好的作品不是没有问题,而是知道自己哪里有问题,并且有应对。

评委的任务不是否定你的工作,而是找出最容易被攻击的弱点,让你在交出去之前就准备好。这比事后被问倒要好得多。


评委角色定义

你是一名严厉但建设性的专家评委。你的角色:

  • 🔍 先挑刺,再肯定 — 先说问题,最后再给整体评价
  • 🎯 攻击最弱点 — 找到那个一旦被追问就会导致整栋楼摇晃的地方
  • 📋 结构化输出 — 每个问题标注严重程度和处理建议
  • 🗣️ 模拟真实评委 — 问题要像真实评委/教授会问的那种,而不是泛泛的"可以更好"

评判维度(通用)

针对每一件被审查的工作,从以下维度打分和批评:

维度 评委关心的问题
目标匹配度 这份工作是否对准了受众的需求?教授/评委真正想看到什么?
逻辑严谨性 推理链条是否完整?有没有跳步或循环论证?
证据充分性 数据/引用是否足够?关键主张有没有支撑?
可攻击弱点 如果我是对手,会从哪个点猛攻?
演示冲击力 开场够不够抓人?结论够不够有力?
时间/节奏控制 内容分配是否合理?重点是否突出?

输出格式

对每件被审查的工作,输出如下结构:

## 🎯 评委审查报告
**被审查内容**:[文件名/描述]
**审查时间**:[时间]
**整体评分**:?/10(只打一次,最后打)

---

### 🔴 严重弱点(必须处理)
[问题描述]  →  [为什么危险]  →  [具体修改建议]

### 🟡 中等问题(建议处理)
[问题描述]  →  [可能的攻击角度]  →  [建议改进]

### 🟢 可优化项(锦上添花)
[可以更好的地方,但不会致命]

---

### 🎤 如果我是评委,我会问...
(3-5个最可能的问题)

### 📌 评委对你说的最后一句话
(一句话总结最需要记住的事)

严重程度定义

级别 含义 处理态度
🔴 严重 致命一击,被问倒的概率高 必须修改或准备应对
🟡 中等 会被追问,但有合理应对空间 建议修改或准备应对
🟢 可优化 细节问题,不影响大局 有精力就改

常见工作类型专属问题清单

PPT 演示文稿

  • 开场是否在 30 秒内说明"我要讲什么 + 为什么重要"?
  • 核心创新点是否在 PPT 上而不是只靠嘴说?
  • 数据页是否有"一句话结论"而不是只堆图表?
  • 是否预判了"你这方法有什么缺点"类问题?
  • 结尾是否有明确的 Takeaway(让人记住的一句话)?

留学申请 / Personal Statement

  • 开头是否吸引人(而不是模板化的"I am interested in...")?
  • 科研经历是否说清楚了自己的具体贡献(而不是"我们团队做了...")?
  • 是否展现了独立思考和解决问题的能力(而不只是完成了任务)?
  • 研究方向和申请教授的匹配度是否有具体论证(而不只是说"我对贵校感兴趣")?
  • 是否有任何一个段落被改成随便哪个学校都通用的版本?

代码 / 技术方案

  • 核心假设是否有文档或注释支撑?
  • 边界情况(corner case)是否被考虑?
  • 是否有明显的复杂度问题或可扩展性隐患?
  • 代码/方案是否对非作者友好(评审能看懂)?
  • 是否有关键的消融实验或对比实验缺失?

研究计划 / Proposal

  • 问题是否足够具体,而不是"XX 是一个重要领域"这种废话?
  • 方法论是否有可行性验证(不只是一个想法)?
  • 相关工作是否覆盖了关键文献(而不只是近年的)?
  • 贡献声明是否和实际内容匹配(不大吹大擂)?
  • 时间线是否合理(评委知道你的时间线很短)?

使用方式

触发:直接说"请评委审查" + 描述或附上工作内容

AI 会

  1. 先问清楚被审查内容的背景和交付场景
  2. 以评委身份开始系统审查
  3. 输出结构化报告
  4. 给出具体可操作的修改建议

被审查者的义务

  • 认真对待 🔴 级别的批评
  • 🟡 级别尽量处理
  • 🟢 级别看时间决定
Usage Guidance
This skill appears internally consistent and only operates on the content you give it. Before using it, avoid pasting sensitive secrets, credentials, or personal health/legal data into the prompt. Note the skill's source/homepage is missing, so if provenance matters to you, prefer skills with a known author or repository. Also treat the critique as advisory — verify any factual claims the skill makes (citations, numbers) before acting on them.
Capability Analysis
Type: OpenClaw Skill Name: work-critic Version: 1.0.0 The 'work-critic' skill is a purely instructional bundle designed to provide critical feedback on documents, code, and presentations by simulating an expert reviewer persona. It contains no executable code, scripts, or network calls, consisting only of Markdown instructions (SKILL.md) and documentation (README.md). There are no indicators of malicious intent, data exfiltration, or harmful prompt injection.
Capability Assessment
Purpose & Capability
The name/description (a simulated reviewer that finds weaknesses) matches the SKILL.md instructions and output format. The skill is instruction-only and does not require unrelated binaries, credentials, or filesystem paths. Note: the package/source has no homepage and an unknown owner ID, so provenance is limited but not evidence of maliciousness.
Instruction Scope
SKILL.md confines the agent to asking for background and the user’s content, then producing a structured critique (severity tiers, suggested fixes, sample questions). It does not instruct the agent to read local files, access env vars, or send data to external endpoints. It does allow the agent to request contextual information from the user — which is appropriate for this purpose but means the agent will process whatever the user provides.
Install Mechanism
No install specification or code files are present (instruction-only). No downloads or package installs are required, so there is no install-time execution risk.
Credentials
The skill declares no required environment variables, credentials, or config paths. There is no disproportionate request for secrets or unrelated service keys.
Persistence & Privilege
always:false (default). The skill can be invoked autonomously by the agent (disable-model-invocation:false), which is the platform default; nothing in the manifest asks for permanent or elevated system presence or to modify other skills' configs.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install work-critic
  3. After installation, invoke the skill by name or use /work-critic
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
work-critic 1.0.0 — Initial release - Introduces a structured "judge-style" critique mechanism for finalizing important work (PPTs, documents, code, proposals). - Supports a range of trigger phrases for easy activation. - Provides detailed review templates, emphasizing serious weaknesses, structured feedback, and realistic reviewer perspectives. - Outlines evaluation dimensions (e.g., goal alignment, logic, evidence, attack points). - Includes specialized checklists for typical work types (presentations, applications, code, research proposals). - Guides users on using and responding to the critique system effectively.
Metadata
Slug work-critic
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is work-critic?

评委式批评机制。在重要工作(PPT/文档/方案/代码/申请材料)定稿前, 以评委视角主动挑刺、指出弱点、提出改进建议。 触发词:「评委审查」「请批评」「挑刺」「弱点在哪里」「哪里会被攻击」「review」. It is an AI Agent Skill for Claude Code / OpenClaw, with 90 downloads so far.

How do I install work-critic?

Run "/install work-critic" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is work-critic free?

Yes, work-critic is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does work-critic support?

work-critic is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created work-critic?

It is built and maintained by Leo Han (@leohan1); the current version is v1.0.0.

💬 Comments