← 返回 Skills 市场
453
总下载
0
收藏
2
当前安装
1
版本数
在 OpenClaw 中安装
/install auto-researcher
功能描述
自主研究助手 - 深度调研、交叉验证、生成引用报告
使用说明 (SKILL.md)
Auto Researcher - 自主研究助手
🎯 核心功能
自主深度研究任何主题:
- 多源交叉验证
- CRAAP 可信度评估
- 生成引用报告
- 知识图谱构建
- 持续监控更新
灵感来源: OpenFang Researcher Hand
📚 研究方法
5 阶段研究流程
1. 定义 (Define)
- 澄清问题
- 识别已知/未知
- 设定范围
2. 搜索 (Search)
- 多策略搜索
- 多样化来源
- 查询优化
3. 评估 (Evaluate)
- CRAAP 框架
- 提取数据
- 记录限制
4. 综合 (Synthesize)
- 整合发现
- 解决矛盾
- 识别不确定性
5. 验证 (Verify)
- 交叉检查
- 标注置信度
🔍 CRAAP 评估框架
Currency (时效性)
- 何时发布/更新?
- 信息是否仍然当前?
- 链接是否有效?
- 技术主题>2 年可能过时
Relevance (相关性)
- 是否直接回答问题?
- 目标受众是谁?
- 详细程度是否合适?
- 是否愿意引用?
Authority (权威性)
- 作者资质?
- 出版机构?
- 联系信息?
- 域名类型?(.gov/.edu/组织)
Accuracy (准确性)
- 是否有证据支持?
- 是否经过审核?
- 能否从其他来源验证?
- 是否有事实错误?
Purpose (目的)
- 为什么存在?
- 信息/商业/说服/娱乐?
- 偏见是否明显?
- 作者是否受益?
评分标准
A (权威): 通过全部 5 项
B (可靠): 通过 4/5 项
C (有用): 通过 3/5 项,需谨慎
D (弱): 通过≤2/5 项
F (不可靠): 失败,不要引用
🔎 搜索优化技巧
查询构建
| 技巧 | 语法 | 示例 |
|---|---|---|
| 精确短语 | "..." |
"AI Agent 操作系统" |
| 站内搜索 | site:... |
site:zhihu.com OpenClaw |
| 排除 | - |
AI -artificial |
| 文件类型 | filetype:... |
filetype:pdf 报告 |
| 时间范围 | after:... |
after:2025-01-01 |
| OR 操作符 | OR |
(OpenClaw OR OpenFang) |
| 通配符 | * |
"如何用*赚钱" |
多策略搜索模式
主题:OpenClaw 赚钱方法
搜索查询组合:
1. "OpenClaw 赚钱" site:zhihu.com
2. "OpenClaw 变现" after:2025-01-01
3. "OpenClaw Skill" 开发 外包
4. "ClawHub" 技能市场 收入
5. OpenClaw vs OpenFang 对比
6. "AI Agent" 副业 2025
7. site:github.com openclaw skills
8. filetype:pdf openclaw 文档
📊 报告生成
输出格式
# 研究报告:[主题]
## 执行摘要
[200 字核心发现]
## 研究方法
- 搜索查询:[列出使用的查询]
- 来源数量:[N 个]
- 研究时间:[日期范围]
## 核心发现
### 发现 1: [标题]
**内容**: [详细描述]
**来源**: [引用,CRAAP 评分]
**置信度**: [高/中/低]
### 发现 2: [标题]
...
## 相互矛盾的信息
[列出不同来源的矛盾点,分析原因]
## 知识缺口
[识别尚未解答的问题]
## 参考文献
[APA 格式引用列表]
## 附录
- 完整搜索结果
- 原始数据
- 方法论细节
🧠 知识图谱构建
实体类型
- 人物 (Person)
- 组织 (Organization)
- 概念 (Concept)
- 产品 (Product)
- 事件 (Event)
- 地点 (Location)
关系类型
- 属于 (belongs_to)
- 创建 (created_by)
- 使用 (uses)
- 竞争 (competes_with)
- 影响 (influences)
- 引用 (cites)
存储格式
{
"entities": [
{
"id": "openclaw",
"type": "Product",
"name": "OpenClaw",
"attributes": {
"description": "开源 AI 助手",
"language": "TypeScript",
"creator": "Peter Steinberger"
}
}
],
"relations": [
{
"from": "openclaw",
"to": "peter_steinberger",
"type": "created_by"
}
]
}
📋 使用示例
激活研究
帮我研究一下"知乎盐选投稿指南",要详细的
查看进度
研究进行得怎么样了?
获取报告
把研究结果整理成报告,要 APA 引用格式
持续监控
持续监控"OpenClaw 新功能",有更新告诉我
🔧 配置选项
# 研究深度
research_depth = "deep" # basic/standard/deep
max_sources = 20 # 最多引用来源数
min_craap_score = "B" # 最低可信度
# 输出设置
output_format = "markdown" # markdown/pdf/html
citation_style = "APA" # APA/MLA/Chicago
# 监控设置
monitor_enabled = false # 是否持续监控
monitor_frequency = "daily" # daily/weekly
📊 仪表盘指标
{
"researcher_reports_generated": 0,
"researcher_sources_evaluated": 0,
"researcher_entities_stored": 0,
"researcher_relations_stored": 0,
"researcher_last_report_date": null,
"researcher_avg_credibility_score": 0
}
🎯 应用场景
场景 1: 市场调研
研究"小红书 AI 工具博主"的变现方式
- 分析 Top 10 博主
- 统计变现模式
- 估算收入范围
- 给出进入建议
场景 2: 竞品分析
研究 OpenClaw 的竞品
- OpenFang vs ZeroClaw vs CrewAI
- 功能对比
- 性能基准
- 市场份额
场景 3: 技术调研
研究"AI Agent 自主赚钱"的可行性
- 现有案例
- 技术栈
- 法律风险
- 实操步骤
场景 4: 持续监控
监控"知乎盐选过稿率"变化
- 每周收集数据
- 追踪趋势
- 异常 alert
- 生成月报
📝 从 OpenFang 借鉴的功能
- ✅ CRAAP 评估框架 (直接采用)
- ✅ 5 阶段研究流程 (优化适配)
- ✅ 知识图谱存储 (简化实现)
- ✅ 多源交叉验证 (完整保留)
- ✅ 引用报告生成 (APA/MLA 支持)
🔧 OpenClaw 适配
| OpenFang 功能 | OpenClaw 实现 |
|---|---|
shell_exec |
exec 工具 |
knowledge_add_entity |
memory/store JSON |
web_search |
searxng skill |
web_fetch |
web_fetch 工具 |
schedule_create |
qqbot-cron |
dashboard |
SESSION-STATE.md |
此 Skill 受 OpenFang Researcher Hand 启发创建
安全使用建议
This skill is essentially a detailed set of instructions for how an agent should perform research. It's not installing anything, but it tells the agent to use tools that can fetch web data, run shell commands, store information, and set up ongoing monitoring. Before installing or enabling it: (1) Confirm which agent tools (exec, web_fetch, searxng, memory/store, scheduler) are available and what privileges they have; (2) If you don't want shell access or continuous network monitoring, disable or remove those tools or run the skill in a restricted context; (3) Test with non-sensitive queries to observe behavior and storage locations (where entities and SESSION-STATE.md would be written); (4) Be cautious about autonomous/recurring monitoring — limit frequency or require manual approval for scheduled tasks; (5) Because source/homepage is unknown, prefer not to grant it high trust or persistent privileges until you can verify the author or origin.
功能分析
Type: OpenClaw Skill
Name: auto-researcher
Version: 1.0.0
The `SKILL.md` file, which serves as instructions for the AI agent, explicitly maps `shell_exec` functionality to the `exec` tool and `schedule_create` to `qqbot-cron`. The `exec` tool typically allows for arbitrary command execution, presenting a significant remote code execution (RCE) vulnerability if the agent is subsequently prompted to execute malicious commands. The instruction to use a scheduling tool (`qqbot-cron`) also introduces a risk for persistence or repeated unauthorized actions. While the skill's stated purpose is benign research, these instructions enable high-risk capabilities for the agent, classifying it as suspicious due to the potential for misuse.
能力评估
Purpose & Capability
The skill's name, description, and SKILL.md align: it is a researcher assistant that outlines search, evaluation, synthesis, and monitoring workflows. However, the doc explicitly maps features to tools (e.g., exec/shell_exec, web_fetch, searxng, memory/store, schedule_create) even though the skill declares no required binaries, env vars, or install steps. That mismatch is not necessarily malicious but is a design inconsistency: the skill expects capabilities that are not enumerated in the metadata.
Instruction Scope
The SKILL.md instructs the agent to perform multi-source web search/fetching, use shell execution (exec), store entities in memory/JSON, and set up continuous monitoring/schedules. Those are broad runtime actions: shell execution can run arbitrary commands, web_fetch can access external sites, and scheduling/monitoring implies recurring activity. The instructions do not constrain or limit these actions (no explicit bans on private files, system paths, or where results are sent), giving the agent broad discretion.
Install Mechanism
This is a prompt-only skill with no install spec and no code files, which minimizes disk writes and direct install risk. No downloads, packages, or binaries are installed by the skill itself.
Credentials
The skill declares no required environment variables or credentials, which keeps its declared surface small. But the instructions assume access to external tools/APIs (web fetch/search, scheduling, memory store) that in practice may require credentials or tokens. The absence of declared env vars is a potential omission rather than proof of safety — enabling it may rely on other skills/tools that have their own secrets.
Persistence & Privilege
always:false (good). However, the skill describes continuous monitoring and schedule creation, which implies persistent/recurring activity if the agent and platform allow it. Autonomous invocation is permitted by default on the platform; combined with monitoring/scheduling this increases potential for ongoing network activity or repeated access to data. The skill does not request to modify other skills or system-wide config.
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 auto-researcher - 安装完成后,直接呼叫该 Skill 的名称或使用
/auto-researcher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
auto-researcher 1.0.0 初始发布:自主研究助手
- 实现 5 阶段系统性研究流程(定义、搜索、评估、综合、验证)
- 集成 CRAAP 可信度评估,自动评分来源
- 支持多源交叉验证并生成 APA/MLA 格式引用报告
- 提供知识图谱构建和实体-关系存储模板
- 包含报告输出模板、配置项与仪表盘指标
- 借鉴 OpenFang 主要研究功能并适配 OpenClaw 生态
元数据
常见问题
Auto Researcher 是什么?
自主研究助手 - 深度调研、交叉验证、生成引用报告. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 453 次。
如何安装 Auto Researcher?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install auto-researcher」即可一键安装,无需额外配置。
Auto Researcher 是免费的吗?
是的,Auto Researcher 完全免费(开源免费),可自由下载、安装和使用。
Auto Researcher 支持哪些平台?
Auto Researcher 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 Auto Researcher?
由 xiaomo(@bandwe)开发并维护,当前版本 v1.0.0。
推荐 Skills