← 返回 Skills 市场
Cross Verify
作者
Irving ZHAO
· GitHub ↗
· v1.0.0
· MIT-0
111
总下载
1
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install cross-verify
功能描述
交叉验证信息准确性、识别偏见和逻辑漏洞。当用户说"这段话有没有问题"、"这个数据可靠吗"、"帮我看看这段有没有偏见"、"这个信息准确吗"、"核查一下这段",或直接丢入一段文字/数据要求检查时触发。Also triggers when the user pastes any text containing stat...
使用说明 (SKILL.md)
Cross-Verify Skill
给定一段文字或数据,对其进行事实核查、偏见识别和逻辑分析,输出结构化的验证报告。
验证维度
1. 事实核查(Fact Check)
- 识别所有可验证的具体声明(数据、日期、百分比、名称、机构)
- 对每个声明进行核查:搜索可靠来源(官方数据、权威媒体、学术文献)
- 标注:✅ 已验证 | ⚠️ 无法确认 | ❌ 与事实不符
- 如果数据存疑,给出参考依据
2. 偏见识别(Bias Detection)
检查信息来源和内容本身的潜在偏见:
- 来源偏见:信息来自哪里?是否有利益关联?是否是一手来源?
- 选择偏差:是否只选了有利证据?是否忽略了对立证据?
- 确认偏误:是否在强化某种既得观点?
- 语言倾向:用词是否带有情感偏向("惊人"、"竟然"、"不过是"等)
- 标注:
来源偏见|选择性呈现|情感化语言|利益相关方
3. 逻辑漏洞(Logic Gaps)
- 因果混淆(相关≠因果)
- 样本偏差(以偏概全)
- 稻草人(反驳弱化后的观点)
- 诉诸权威/情感
- 标注:
因果混淆|样本偏差|逻辑断层
4. 可信度评分(Credibility Score)
对整体内容给出一个 1-10 的可信度评分:
| 评分 | 含义 |
|---|---|
| 9-10 | 多方信源确认,无明显偏见,逻辑严密 |
| 7-8 | 基本可靠,有小问题或信息不完整 |
| 5-6 | 存在明显不确定性,需进一步核实 |
| 3-4 | 存在严重偏见或逻辑漏洞,可信度低 |
| 1-2 | 基本不可信,多处存疑或明显错误 |
评分要给出简要理由。
输出格式
━━━━━━━━━━━━━━━━━━
🔍 交叉验证报告
📊 可信度评分:X / 10 [评分含义]
━━━━━━━━━━━━━━━━━━
✅ 已验证声明
- ...
⚠️ 无法确认的声明
- [声明] → 缺乏可靠来源/存疑
参考:...
❌ 与事实不符
- [声明] → 实际为:...
━━━━━━━━━━━━━━━━━━
🧠 偏见标记
- [类型]:[具体内容]
- ...
🔗 逻辑漏洞
- [类型]:[具体内容]
- ...
📌 信息来源评估
- 来源:[来源名称]
- 利益关联:[有/无/不确定]
- 一手/二手:[判断]
━━━━━━━━━━━━━━━━━━
💡 总结建议
[1-2句话给出综合判断和补充建议]
━━━━━━━━━━━━━━━━━━
工作流程
- 提取声明 — 识别文本中所有具体声明(数据、事实、断言)
- 并行核查 — 使用 web_search 搜索每个关键声明的可信度
- 分析偏见 — 从来源、用词、证据选择角度扫描
- 扫描逻辑 — 查找常见逻辑谬误
- 综合评分 — 给出评分和总结
输出原则
- 不回避:即使结论是"这段基本不可信",也要直接说
- 有依据:每个标注都要有具体证据
- 区分层次:轻微问题、严重问题分开展示
- 有建设性:在总结里给出"如果要做实,需要补充什么"
适用边界
触发:
- 用户主动粘贴文字/数据要求核查
- 用户说"有没有问题"、"可靠吗"、"有没有偏见"
不触发:
- 用户只是在问信息(不要求验证)
- 用户在写小说、诗歌等创作内容
- 纯娱乐/闲聊
安全使用建议
This skill is coherent with its stated purpose and doesn't request credentials or install code. Before using: (1) confirm whether your agent has an actual web/browsing tool enabled — the skill expects to run web searches to verify claims; otherwise evidence may be absent or fabricated; (2) avoid pasting sensitive or private data because the skill will use external search/services to verify claims; (3) if you require provenance guarantees, ask which exact sources were consulted and verify them independently. If you need offline verification only, this skill may not be suitable unless you ensure the agent's browsing is disabled and accept limited checks.
功能分析
Type: OpenClaw Skill
Name: cross-verify
Version: 1.0.0
The 'cross-verify' skill is a standard utility designed for fact-checking, bias detection, and logical analysis of text. The SKILL.md file contains well-structured instructions for the AI agent to perform verification tasks using web searches and provides a clear reporting format without any evidence of malicious intent, data exfiltration, or prompt injection attacks.
能力评估
Purpose & Capability
Name/description (fact check, bias detection, logic review) matches the SKILL.md tasks (extract claims, search sources, label bias/logic, produce a structured report). The skill does not request unrelated credentials or system access.
Instruction Scope
Instructions stay within the stated purpose (extract claims, run web searches, evaluate sources, mark bias and logic gaps). One minor mismatch: the workflow explicitly says to use "web_search" but the skill metadata does not declare any required browsing/tool capability — this is an operational gap (not a security violation) that may cause the agent to either perform no external checks or hallucinate sources if browsing isn't available. The SKILL.md does not instruct reading any user files or environment variables beyond the pasted text.
Install Mechanism
No install spec and no code files (instruction-only). This is lowest-risk from an install perspective: nothing will be written to disk or fetched during install.
Credentials
The skill requests no environment variables, credentials, or config paths. That is proportionate for a fact‑checking assistant. There are no hidden requests for tokens or keys in the instructions.
Persistence & Privilege
always:false and no special persistence or system modifications. The skill does not request permanent presence or elevated privileges.
如何使用
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install cross-verify - 安装完成后,直接呼叫该 Skill 的名称或使用
/cross-verify触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
Initial release of the "cross-verify" skill.
- Provides structured reports for fact-checking, bias detection, and logic analysis of text or data.
- Assigns a credibility score (1–10) with clear grading criteria and a brief rationale.
- Supports detailed marking of verified, unverified, and factually incorrect claims.
- Detects potential biases (source, selection, confirmation, emotional language) and common logic gaps.
- Gives concise summary advice for users seeking verification or cross-checking.
元数据
常见问题
Cross Verify 是什么?
交叉验证信息准确性、识别偏见和逻辑漏洞。当用户说"这段话有没有问题"、"这个数据可靠吗"、"帮我看看这段有没有偏见"、"这个信息准确吗"、"核查一下这段",或直接丢入一段文字/数据要求检查时触发。Also triggers when the user pastes any text containing stat... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 111 次。
如何安装 Cross Verify?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install cross-verify」即可一键安装,无需额外配置。
Cross Verify 是免费的吗?
是的,Cross Verify 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
Cross Verify 支持哪些平台?
Cross Verify 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 Cross Verify?
由 Irving ZHAO(@irvinezhao)开发并维护,当前版本 v1.0.0。
推荐 Skills