Tabbit开发工具
tabbit-devtools
by carri1sun
Use Tabbit with agent-browser by reading Tabbit's live DevToolsActivePort file, deriving the browser wsEndpoint, and routing browser actions through agent-browser --cdp.
安装
claude skill add --url https://github.com/openclaw/skills文档
Tabbit Devtools
Prefer this skill whenever the request is explicitly about Tabbit or includes phrases like 用我的 tabbit 浏览器, 在 Tabbit 里, Tabbit 当前页, or Tabbit 当前标签.
Treat Tabbit as a Chromium-based browser.
This skill is about how to connect agent-browser to Tabbit.
After the connection is established, handle browser automation and inspection through the normal agent-browser workflow.
Do not implement a parallel browser automation layer, bridge daemon, or custom CDP client inside this skill.
agent-browser Quick Reference
Treat agent-browser as the browser-operation layer after Tabbit endpoint discovery.
The most relevant commands for this skill are:
open <url>snapshot -iclick @e3fill @e5 <text>press Enter
Do not restate a full agent-browser manual here. Use these commands as the default vocabulary for Tabbit tasks, and prefer the official README for any broader command surface.
Quick Path
- Read
~/Library/Application Support/Tabbit/DevToolsActivePortfirst. - If that file does not exist, read
~/Library/Application Support/Tabbit Browser/DevToolsActivePort. - Use both lines in that file:
- line 1: TCP port
- line 2: browser path such as
/devtools/browser/<id>
- Build the full browser endpoint as
ws://127.0.0.1:<port><path>. - Prefer that
wsEndpointoverhttp://127.0.0.1:<port>. Tabbit may expose the browser WebSocket while/json/versionand/json/liststill return404. - Prefer scripts/run_agent_browser_on_tabbit.py for actual browser actions. It injects the live
wsEndpointintoagent-browser --cdp .... - Use scripts/discover_tabbit_cdp.py when you only need structured connection facts.
- Once connected, use the full normal
agent-browserworkflow for page operations.
Workflow
- For Tabbit requests, start by reading
DevToolsActivePortdirectly or by running scripts/discover_tabbit_cdp.py. - Return the connection facts the agent actually needs:
activePortFile,port,browserPath,browserUrl, andwsEndpoint. - Unless the user explicitly asks only for endpoint details, prefer scripts/run_agent_browser_on_tabbit.py immediately so the command becomes
agent-browser --cdp <wsEndpoint> .... - After that handoff, follow the normal
agent-browserworkflow for open, snapshot, click, fill, and other browser commands. - If
agent-browseris unavailable, say so plainly and surface the connection facts instead of inventing a custom CDP bridge.
Guidance
- This skill solves the connection problem, not the general browser-operation problem.
- Return structured connection data first, then any short explanatory note.
- Prefer the lightest possible discovery path:
DevToolsActivePortand the derived browser WebSocket endpoint. - Search the macOS
Tabbitsupport directory first, thenTabbit Browser. - Prefer the full
wsEndpointover a raw port because Tabbit may not expose HTTP discovery routes. - Once a Tabbit task has started through
run_agent_browser_on_tabbit.py, keep using that same wrapper path for the rest of the task unless the user explicitly asks otherwise. - Once connected, use standard
agent-browserpatterns for everything else.
Constraints
- Do not assume a dedicated
tabbit-devtoolsMCP server exists. - Do not assume the generic
chrome-devtoolssession can be retargeted to Tabbit. - Do not turn this skill into a replacement for
agent-browser. - Do not create a custom daemon, long-lived CDP proxy, or one-off WebSocket client for post-connection browser actions.
- Do not promise that
chrome-devtoolsMCP will automatically take over Tabbit. - If
agent-browsercannot be launched in the current environment, stop at connection guidance and explain the limitation. - After connection, the browser workflow belongs to
agent-browser, not to this skill.
Resources
- Setup and direct-connection notes: references/setup.md
- Endpoint discovery rules and environment variables: references/discovery.md
- Endpoint discovery helper: scripts/discover_tabbit_cdp.py
- agent-browser wrapper: scripts/run_agent_browser_on_tabbit.py
- agent-browser upstream docs: https://github.com/vercel-labs/agent-browser
相关 Skills
PPT处理
by anthropics
处理 .pptx 全流程:创建演示文稿、提取和解析幻灯片内容、批量修改现有文件,支持模板套用、合并拆分、备注评论与版式调整。
✎ 涉及PPTX的创建、解析、修改到合并拆分都能一站搞定,连备注、模板和评论也能处理,做演示文稿特别省心。
技能工坊
by anthropics
覆盖 Skill 从创建到迭代优化全流程:起草能力、补测试提示、跑评测与基准方差分析,并持续改写内容和描述,提升效果与触发准确率。
✎ 技能工坊把技能从创建、迭代到评测串成闭环,方差分析加描述优化,特别适合把触发准确率打磨得更稳。
Word文档
by anthropics
覆盖Word/.docx文档的创建、读取、编辑与重排,适合生成报告、备忘录、信函和模板,也能处理目录、页眉页脚、页码、图片替换、查找替换、修订批注及内容提取整理。
✎ 搞定 .docx 的创建、改写与精排版,目录、批量替换、批注修订和图片更新都能自动化,做正式文档尤其省心。
相关 MCP 服务
文件系统
编辑精选by Anthropic
Filesystem 是 MCP 官方参考服务器,让 LLM 安全读写本地文件系统。
✎ 这个服务器解决了让 Claude 直接操作本地文件的痛点,比如自动整理文档或生成代码文件。适合需要自动化文件处理的开发者,但注意它只是参考实现,生产环境需自行加固安全。
by wonderwhy-er
Desktop Commander 是让 AI 直接执行终端命令、管理文件和进程的 MCP 服务器。
✎ 这工具解决了 AI 无法直接操作本地环境的痛点,适合需要自动化脚本调试或文件批量处理的开发者。它能让你用自然语言指挥终端,但权限控制需谨慎,毕竟让 AI 执行 rm -rf 可不是闹着玩的。
EdgarTools
编辑精选by dgunning
EdgarTools 是无需 API 密钥即可解析 SEC EDGAR 财报的开源 Python 库。
✎ 这个工具解决了金融数据获取的痛点——直接让 AI 读取结构化财报,比如让 Claude 分析苹果的 10-K 文件。适合量化分析师或金融开发者快速构建数据管道。但注意,它依赖 SEC 网站稳定性,高峰期可能延迟。