← 返回 Skills 市场
sansan-mei

社交媒体研究助手Skill

作者 梅花三十三 · GitHub ↗ · v1.0.2 · MIT-0
cross-platform ✓ 安全检测通过
272
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install media-research-crawl-skill
功能描述
通过本机 media-agent-crawler HTTP 服务搜集 B站/抖音/YouTube/知乎内容(不依赖 MCP 客户端安装)。当用户要搜集这些平台内容、并已在本机启动应用(默认 http://127.0.0.1:39002)时使用。
使用说明 (SKILL.md)

media-crawler-local

直接调用本机 HTTP 服务,不走 OpenClaw/Cursor 的 MCP 客户端配置。

前置确认

先从用户消息或上下文中提取以下信息,缺少时再询问:

  • 操作类型:搜集内容 / 查询归档 / 读取任务数据
  • 目标链接或关键词
  • 平台(可从链接自动推断)

工具清单

B 站系列

工具名 必填参数 说明
crawl_bilibili url 视频 URL 或 BV 号
crawl_bilibili_search keyword 按关键词触发搜索结果搜集
crawl_bilibili_uploader mid UP 主纯数字 ID,触发视频列表搜集
crawl_bilibili_popular 热门视频搜集
crawl_bilibili_weekly 无(可选 number 每周必看,不传 number 则自动取最新一期
crawl_bilibili_history 无(可选 max/view_at/business/ps/type/page_count 历史记录聚合搜集,不传 page_count 时跟随 dailyRecommendPageCount

所有 B 站工具均支持可选 cookies 参数(字符串,从浏览器插件获取)。

其他平台

工具名 必填参数 说明
crawl_douyin url 抖音视频 URL
crawl_youtube url YouTube 视频 URL 或视频 ID
crawl_zhihu url 知乎问题或回答 URL

归档与数据读取

工具名 必填参数 可选参数 说明
list_archives platform / keyword / limit / sort_by / created_after 列出归档任务,默认返回最多 50 条,按时间倒序
get_task_data task_id type 读取任务目录下的数据文件

list_archives 参数说明:

  • sort_bydate(默认,创建时间倒序)或 status(running → failed → unknown → finished)
  • created_after:ISO 日期,如 2026-03-182026-03-18T10:00:00Z

get_task_datatype 支持以下值(含别名):

type 值 读取的数据
comments / comment 评论数据
danmaku 弹幕数据
subtitles / subtitle / caption / captions 字幕数据
detail / info 视频/帖子详情
all / full 全量聚合数据
summary / ai_summary AI 摘要
不传 返回目录下所有可识别文件

HTTP 端点

服务地址默认 http://127.0.0.1:39002,可通过环境变量 BIL_CRAWL_URL 覆盖。

搜集端点(REST)

POST /start-crawl/{platform}/{encodedUrl}
Content-Type: application/json

{ "source": "ai" }

encodedUrl 需要 encodeURIComponent 编码;platform 取值:bilibili / douyin / youtube / zhihu

MCP 端点(JSON-RPC 2.0)

POST /mcp
Accept: application/json, text/event-stream
Content-Type: application/json

请求体格式:

{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "\x3Ctool>", "arguments": { } } }

脚本用法

所有脚本位于 skills/media-crawler-local/scripts/,工作目录为 openclaw workspace 根。

1. 快速搜集(REST,crawl.sh

bash skills/media-crawler-local/scripts/crawl.sh \x3Cplatform> \x3Curl> [base_url]

示例:

bash skills/media-crawler-local/scripts/crawl.sh bilibili "https://www.bilibili.com/video/BV1xx411c7mD"

2. 通过 MCP 搜集(crawl_mcp.sh,仅支持带 url 的工具)

bash skills/media-crawler-local/scripts/crawl_mcp.sh \x3Ctool_name> \x3Ctarget_url> [base_url]

示例:

bash skills/media-crawler-local/scripts/crawl_mcp.sh crawl_bilibili "https://www.bilibili.com/video/BV1xx411c7mD"

支持工具:crawl_bilibili / crawl_douyin / crawl_youtube / crawl_zhihu

其余工具(bilibili_search / bilibili_uploader / bilibili_popular / bilibili_weekly / bilibili_history / list_archives / get_task_data)请用 mcp_tool.sh

3. 归档查询(list_archives_mcp.sh

bash skills/media-crawler-local/scripts/list_archives_mcp.sh [platform] [keyword] [limit] [base_url]

示例:

bash skills/media-crawler-local/scripts/list_archives_mcp.sh bilibili "蛋神" 20

4. 通用工具调用(mcp_tool.sh

bash skills/media-crawler-local/scripts/mcp_tool.sh \x3Ctool_name> [args_json] [base_url]

示例:

# B 站搜索
bash skills/media-crawler-local/scripts/mcp_tool.sh crawl_bilibili_search '{"keyword":"蛋神"}'

# UP 主视频列表
bash skills/media-crawler-local/scripts/mcp_tool.sh crawl_bilibili_uploader '{"mid":"123456"}'

# 热门视频
bash skills/media-crawler-local/scripts/mcp_tool.sh crawl_bilibili_popular '{}'

# 每周必看(最新一期)
bash skills/media-crawler-local/scripts/mcp_tool.sh crawl_bilibili_weekly '{}'

# 每周必看(指定期数)
bash skills/media-crawler-local/scripts/mcp_tool.sh crawl_bilibili_weekly '{"number":364}'

# 历史记录(默认参数)
bash skills/media-crawler-local/scripts/mcp_tool.sh crawl_bilibili_history '{}'

# 历史记录(指定首屏 cursor)
bash skills/media-crawler-local/scripts/mcp_tool.sh crawl_bilibili_history '{"max":0,"view_at":0,"business":"","ps":20,"type":"all"}'

# 历史记录(指定采集页数)
bash skills/media-crawler-local/scripts/mcp_tool.sh crawl_bilibili_history '{"page_count":2}'

# 读取任务评论数据
bash skills/media-crawler-local/scripts/mcp_tool.sh get_task_data '{"task_id":"BV1xx411c7mD-123456","type":"comments"}'

执行流程

  1. 健康检查GET /(连不上则提醒用户先启动应用)。
  2. 发起搜集:优先用 REST 端点(crawl.sh),需要额外工具参数时用 MCP(mcp_tool.sh)。
  3. 结果处理
    • 给用户简要摘要(任务 ID、状态、关键字段)
    • 内容很多时仅展示前几条,说明可通过 get_task_data 继续读取或过滤

故障处理

错误 处理方式
连接失败 提醒先启动 Electron 应用(bun run start / dev
401 / 403 提示检查 cookies 是否已在 store 中,或让用户重新从插件导入
429 按返回的 Retry-After 退避,不密集重试
5xx 最多重试 1 次,返回错误摘要与建议
task_id 不存在 先用 list_archives 查询正确的任务 ID
安全使用建议
This skill is a local client for a crawler running on your machine. Before installing or using it: 1) Ensure the crawler service (Electron app) is really running on localhost (or on a host you trust). If you set BIL_CRAWL_URL, do not point it at an untrusted remote server — the skill will send URLs/arguments there. 2) Be cautious when supplying 'cookies' strings (they can contain login tokens). 3) The package metadata omits required binaries: the scripts call curl and node; make sure those are present and review their versions. 4) If you want stronger assurance, inspect or run the actual media-agent-crawler service code (the skill only calls that service). Overall the skill appears coherent for its stated purpose, with the above operational caveats.
功能分析
Type: OpenClaw Skill Name: media-research-crawl-skill Version: 1.0.2 The skill bundle is a legitimate utility designed to interface with a local media crawler service (defaulting to http://127.0.0.1:39002) for platforms like Bilibili, Douyin, YouTube, and Zhihu. The provided bash scripts (e.g., crawl.sh, mcp_tool.sh) act as wrappers for curl commands to interact with the service's REST and JSON-RPC endpoints. The instructions in SKILL.md are well-aligned with the stated purpose, and there is no evidence of data exfiltration, malicious execution, or harmful prompt injection.
能力评估
Purpose & Capability
The name/description match the actual behavior: the skill is a client for a local media-agent-crawler service (default http://127.0.0.1:39002). The included scripts and SKILL.md describe only crawl/list/get operations for the declared platforms and do not request unrelated cloud credentials or access to unrelated subsystems.
Instruction Scope
Runtime instructions and scripts only construct JSON payloads and call REST (/start-crawl/...) or MCP (/mcp) endpoints on the configured base URL. They do not read arbitrary files, shell history, or system credentials. The SKILL.md correctly documents endpoints, parameters (including optional cookies), and expected behavior.
Install Mechanism
There is no install spec (instruction-only skill) and no downloads/extraction. The provided scripts are simple wrappers that use curl and node via inline node -e snippets; they do not install external code.
Credentials
The skill does not require credentials and does not declare required env vars in registry metadata, but SKILL.md and scripts reference an optional BIL_CRAWL_URL environment variable. The scripts also assume command-line tools (curl, node) exist even though 'required binaries' lists none. Also note the optional 'cookies' parameter can contain session cookies — providing those grants the crawler access to account-protected content and should be considered sensitive.
Persistence & Privilege
always:false and no special persistence is requested. The skill does not modify other skills or system settings. The agent may invoke it autonomously (default), which is normal — this simply allows the agent to call the local crawler when relevant.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install media-research-crawl-skill
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /media-research-crawl-skill 触发
  4. 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
v1.0.2
- 移除了对小红书(xhs)平台的支持,相关工具说明与脚本调用均已更新 - 文档中所有相关参数和脚本说明同步删去了 xhs - 其余平台(B站/抖音/YouTube/知乎)功能保持不变
v1.0.1
- 新增 B 站历史记录聚合采集工具 crawl_bilibili_history,支持可选参数(如页数、首屏 cursor)。 - mcp_tool.sh 与相关文档示例补充了 crawl_bilibili_history 的多种参数调用用法。 - 说明 crawl_mcp.sh 仅支持部分工具,其余包含 crawl_bilibili_history 的工具需通过 mcp_tool.sh 调用。
v1.0.0
Initial release of media-crawler-local. - Enables local collection of Bilibili, Douyin, Xiaohongshu, YouTube, Zhihu content via a media-agent-crawler HTTP service running on the user's machine. - Operates directly through local HTTP endpoints without requiring MCP client configuration. - Adds bash scripts for common actions: collecting content, querying task archives, and tool invocation. - Supports a range of tools for content collection, archive listings, and data retrieval with flexible parameters. - Guidance included for error handling and normal operation workflow.
元数据
Slug media-research-crawl-skill
版本 1.0.2
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 3
常见问题

社交媒体研究助手Skill 是什么?

通过本机 media-agent-crawler HTTP 服务搜集 B站/抖音/YouTube/知乎内容(不依赖 MCP 客户端安装)。当用户要搜集这些平台内容、并已在本机启动应用(默认 http://127.0.0.1:39002)时使用。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 272 次。

如何安装 社交媒体研究助手Skill?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install media-research-crawl-skill」即可一键安装,无需额外配置。

社交媒体研究助手Skill 是免费的吗?

是的,社交媒体研究助手Skill 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

社交媒体研究助手Skill 支持哪些平台?

社交媒体研究助手Skill 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 社交媒体研究助手Skill?

由 梅花三十三(@sansan-mei)开发并维护,当前版本 v1.0.2。

💬 留言讨论