Token导航 LogoToken导航

Agent Skill · 研究检索

surfagent-gmail(surfagent Gmail 搜索)

surfagent-gmail 用于查找、检索和筛选相关信息,适合在 OpenClaw 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

surfagent-gmail 是研究检索类 Agent Skill,可用宿主为 OpenClaw,通过ClawHub获取,权限为操作浏览器;许可证为 MIT-0,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 26,930。

总安装 3,480周安装 145GitHub Stars 公开资料未说明下载 1,160核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT-0
版本
1.3.0

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:surfagent-gmail(surfagent Gmail 搜索)
来源仓库:https://github.com/surfagentapp/surfagent-gmail
安装命令:
openclaw skills install surfagent-gmail
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install surfagent-gmail

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

Gmail 平台技能支持邮件检查、分类与自动回复。可运行撰写任务并管理标签体系。surfagent-gmail 属于研究检索类 Skill,可作为该场景下的辅助能力补充。适用于客服或日程提醒类自动化场景。需启用 Google API 并获取必要作用域权限。建议设置过滤器避免重复发送通知。

适用宿主包括 OpenClaw,接入前应确认版本、权限和运行环境要求。

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 OpenClaw;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 openclaw skills install surfagent-gmail;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要操作浏览器,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

surfagent-gmail 怎么安装?

在终端执行 openclaw skills install surfagent-gmail。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

surfagent-gmail 支持哪些 AI 工具?

公开资料标注支持 OpenClaw。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

surfagent-gmail 需要什么权限?

权限级别是操作浏览器。该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

surfagent-gmail 是免费的吗?

来源标注的许可证是 MIT-0。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

surfagent-gmail 用的人多吗?

公开数据是累计安装 3,480 次,近一周安装 145 次,下载 1,160 次。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

surfagent-gmail 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 专注 OpenClaw 一种宿主,触发方式和使用习惯不用来回切换。
  • 许可证明确为 MIT-0,商用和分发边界好判断。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 OpenClaw,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 需要许可证清楚、并且有第三方扫描结果的团队项目。

不太适合

  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
surfagent-gmail

surfagent Gmail 搜索 · 本页

3,480—操作浏览器MIT-0通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

name
gmail
description
Gmail platform skill for SurfAgent, covering mailbox checks, mailbox triage, latest-thread opening, compose and reply task runners, tab hygiene, sent-message verification, proof rules, and when to use the Gmail adapter over raw browser control.
version
1.3.0
metadata
openclaw
homepage
https://surfagent.app
emoji
📧

Gmail

Gmail-specific operating skill. Use this with the core browser-operations skill.

This skill teaches agents how to work Gmail without guessing at selectors, lying about send success, or burning tokens on giant browser reads.

1. Use this skill for

  • inbox, spam, sent, drafts, and outbox-style mailbox checks
  • deterministic latest-thread opening from a mailbox view
  • mailbox triage with urgency and action-signal scoring
  • compose flows
  • reply flows inside opened threads
  • draft filling
  • send verification
  • Sent Mail and in-thread reply verification
  • Gmail-specific blockers and recoveries
  • deciding when to use the Gmail adapter instead of raw browser control

2. Tool preference

Use this order:

  1. Gmail deterministic task runners
  2. Gmail adapter state tools
  3. Gmail adapter compose/reply tools
  4. targeted browser evaluation only when needed
  5. raw generic browser actions as fallback

Prefer Gmail-native verbs over rediscovering Gmail from scratch every run.

3. Gmail truths that matter

Gmail is not a normal static form page.

It has:

  • dynamic compose surfaces
  • editor behavior that can reject naive HTML injection
  • mailbox navigation state that can be misleading from generic perception alone
  • send flows where a click is not proof

Important: Gmail enforces Trusted Types in places, so direct innerHTML approaches are unreliable for rich content workflows.

4. Core Gmail loop

Default loop:

  1. confirm current Gmail state or mailbox
  2. choose the narrowest deterministic Gmail task runner that fits
  3. verify the target surface is real
  4. fill or extract only what the task needs
  5. verify draft, mailbox, or thread state
  6. send only when requested
  7. verify the visible result with mailbox/thread proof

5. Verified compose flow

Known-good pattern:

  • confirm inbox or current mailbox state first
  • open compose deliberately
  • confirm dialog is actually open
  • fill to, subject, and body
  • inspect composer state after fill
  • send using the real Gmail Send control
  • confirm Gmail shows a send alert like Message sent
  • open Sent Mail
  • verify the sent message exists and the content renders as expected

6. Proof rules

For Gmail, success requires all three layers:

  1. compose proof

- compose dialog open - target fields present - intended values visible in fields

  1. send proof

- send control activated - Gmail alert or live region confirms Message sent - compose closes or transitions correctly

  1. render proof

- Sent Mail contains the message, or - opened sent thread visibly preserves the result

Do not claim success from only:

  • clicking Send
  • a selector existing
  • lack of an error
  • draft text being present before send

7. Rich formatting rule

If the task cares about formatting, proof must come from rendered output, not only compose state.

That means:

  • verify after send in Sent Mail or opened sent thread
  • use screenshot or render inspection when formatting matters
  • do not assume editor formatting survived the send path

8. Gmail adapter, when to use it

Prefer the Gmail adapter for:

  • opening Gmail
  • mailbox state inspection
  • compose open/state
  • filling drafts
  • sending drafts
  • opening Sent Mail
  • visible thread extraction

Current Gmail adapter verbs:

  • gmail_health_check
  • gmail_open
  • gmail_get_state
  • gmail_open_label
  • gmail_extract_visible_threads
  • gmail_open_visible_thread_row
  • gmail_get_open_message
  • gmail_open_compose
  • gmail_open_reply
  • gmail_get_composer_state
  • gmail_fill_compose_draft
  • gmail_send_compose
  • gmail_open_sent

Deterministic Gmail task runners:

  • gmail_compose_and_send_task
  • gmail_reply_and_send_task
  • gmail_check_mailbox_task
  • gmail_open_latest_thread_task
  • gmail_triage_mailbox_task

9. When raw browser control is still acceptable

Use targeted evaluate or browser control when:

  • you need render-only verification not yet wrapped by the adapter
  • you need a one-off Gmail UI probe
  • the adapter is missing a narrow site action
  • Gmail has drifted into an odd hydration state and you need a focused diagnosis before changing the adapter

Even then:

  • keep calls targeted
  • inspect only the relevant surface
  • verify the visible result

10. Common Gmail blockers

Watch for:

  • compose dialog not actually open
  • inline reply surface behaving differently from full compose
  • wrong mailbox or label state
  • stale draft surface
  • editor accepting text differently than expected
  • send button present but not truly actionable
  • rich formatting lost in final render
  • account/login mismatch
  • multiple Gmail tabs causing verification to read the wrong surface

Current adapter rule: Gmail task runners should enforce tab hygiene and collapse duplicate Gmail tabs down to one working tab when they take control of the surface.

Mailbox triage is heuristic and proof-oriented. It is for ranking visible threads and guiding the next action, not for pretending to be a full Gmail classification engine.

When blocked:

  • restate the blocker plainly
  • say whether it is adapter-fixable, retryable, or human-blocked
  • do not quietly downgrade the proof standard

11. Token-efficiency rules for Gmail

Prefer:

  • task runners over hand-built action chains
  • state checks over full page reads
  • compose-state inspection over broad DOM dumps
  • visible thread extraction over giant inbox scraping
  • small post-action verification steps over repeated full reads

Avoid:

  • full-page HTML grabs for simple compose tasks
  • repeated generic perception when Gmail-specific state tools exist
  • large reads of the inbox when only one draft or sent result matters

12. Minimal Gmail checklist

Before claiming a Gmail task done, confirm:

  • correct account
  • correct mailbox or thread
  • compose, reply, mailbox, opened-thread, or triage surface is real
  • intended values, extracted rows, or triage ranking are present
  • send or action actually happened
  • resulting message or state is visible afterward
  • duplicate Gmail tabs were not left behind by the workflow

13. Relationship to other docs

Use alongside:

  • browser-operations for the universal browser rules
  • Gmail adapter docs for the concrete MCP/tool surface
  • runtime wrappers for Claude Code, Codex, Cursor, Hermes, and others

平台分布

OpenClaw

95.7%
按下载量换算1,110

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills