X Engagement

v4.1.3

X/Twitter 运营辅助。完整 onboarding → Persona 学习 → Browser Relay 浏览器控制 → 记忆系统 → 手动提醒模板 → For You 关注建议 → Following 互动建议

1· 467·3 current·3 all-time
Security Scan
Capability signals
Crypto
These labels describe what authority the skill may exercise. They are separate from suspicious or malicious moderation verdicts.
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
The name/description (X engagement / account operations) match what the skill actually does: onboarding, persona learning, generating candidate comments, and using a local browser runtime to execute confirmed write actions. It does not request unrelated credentials or hidden privileged access.
Instruction Scope
SKILL.md and docs instruct the agent to scrape tweets, generate candidate comments, check local history, and only perform likes/follows/comments after explicit user confirmation. That scope aligns with the stated purpose. Two notes: (1) the instructions rely on running npx browser-relay-cli commands (CDP/raw evaluation) to read DOM and interact with the logged‑in browser; (2) the skill writes and reads local files under ~/memory/daily/hotspots and the repo tree—this is expected but worth being aware of.
Install Mechanism
There is no formal install spec (instruction‑only), which lowers risk. However the runtime docs use npx browser-relay-cli and point to a third‑party GitHub repo; npx may fetch/run code from the npm registry on demand. The skill itself does not automatically download or run remote code, but following the docs will invoke third‑party tooling that the user should vet.
Credentials
The skill requests no environment variables or secrets. It operates on local browser sessions via Browser Relay and uses the user's HOME path for memory (memory/daily/hotspots). Those are proportionate to a social‑media automation assistant. No unrelated credentials or config paths are requested.
Persistence & Privilege
always:false and there is no attempt to modify system crontab by default; scripts explicitly avoid installing persistent schedulers. The skill writes to its own memory subtree and playbooks in the repo—expected for a learning/recording assistant. Note: the agent platform default allows autonomous invocation; this skill's docs emphasize manual confirmation for write actions, but if the agent were misconfigured to bypass confirmations the tool could act on the user's logged‑in browser (this is an operational risk, not an incoherence).
Assessment
This package appears coherent for its stated purpose, but before running anything: 1) Vet the Browser Relay tool (https://github.com/jasonCodeSpace/browser-relay) because the docs rely on npx/browser-relay‑cli which will download/execute third‑party code; only proceed if you trust that project. 2) Inspect the repo and scripts locally (especially any calls to npx or raw CDP commands) and confirm you are comfortable with a tool that can control your logged‑in browser. 3) Backup any existing memory/data you care about—cleanup scripts will delete files under ${HOME}/memory/daily/hotspots when run with --apply. 4) Keep the confirmation steps enabled and avoid automating the confirmation step; the skill is designed to require user confirmation before likes/follows/comments. 5) If you want to run this in a production account, test first on a throwaway account and review Browser Relay's source and npm package provenance.

Like a lobster shell, security has layers — review code before you run it.

latestvk97fqa4j8et9bbgn41gftt480584txzk
467downloads
1stars
5versions
Updated 1w ago
v4.1.3
MIT-0

X 运营辅助 Skill v4.2

快速开始

触发条件:

  • "刷推 [时间]"
  • "运营推特 [时间]"
  • "去X上互动 [时间]"

首次运行: 自动进入 Onboarding 流程(详见 docs/onboarding.md

后续运行: 读取配置 → 生成互动建议 → 用户确认后执行


文档结构

x-engagement/
├── SKILL.md                    # 主入口(本文件)
├── docs/
│   ├── onboarding.md           # Onboarding 流程
│   ├── browser-operations.md   # 浏览器操作模块(基于 Browser Relay)
│   ├── comment-rules.md        # 评论规则(重要!防止错误)
│   ├── human-behavior.md       # 自然节奏与确认规范
│   ├── memory-system.md        # 记忆系统设计
│   ├── cron-jobs.md            # 手动提醒与维护
│   ├── comment-generation.md   # 评论生成逻辑
│   └── natural-language-parser.md # 自然语言时间解析
├── playbooks/
│   ├── comment-strategies.md   # 评论策略(有效/无效)
│   └── changelog.md            # 策略变更记录
├── data/
│   └── engagement/
│       └── YYYY-MM-DD.json     # 每日评论数据
├── templates/
│   ├── persona.md              # Persona 模板
│   ├── config.json             # 配置模板
│   └── daily-log.md            # 每日日志模板
└── scripts/
    ├── setup-cron.sh           # 生成手动提醒模板
    ├── check-cron.sh           # 检查本地运行状态
    └── daily-review.sh         # 每日复盘脚本

核心功能

1. Onboarding(首次运行)

5个阶段:

  1. 浏览器连接 + 登录检查
  2. 选择 Persona(自己或其他账号)
  3. 学习 Persona(抓取100条 → 生成描述)
  4. 刷推习惯配置
  5. 保存配置

详见: docs/onboarding.md


2. 自然节奏与确认

核心原则: 保持自然节奏,但不追求伪装或规避检测。

包含:

  • 阅读与判断节奏建议
  • 点赞/关注/评论二次确认
  • 频率限制
  • 人工审核后再执行写操作

详见: docs/human-behavior.md


3. 记忆系统

三层记忆:

memory/daily/hotspots/
├── .onboarding_complete     # Onboarding 标记
├── .config.json             # 用户配置
├── personas/
│   └── [handle].md          # Persona 描述
├── events/                  # 重大事件(永久)
├── tables/                  # 每日热点(7天)
└── history/
    ├── comments/            # 评论历史(避免自相矛盾)
    └── daily/               # 每日日志

关键功能:

  • 记录每次评论内容
  • 记录用户说过的话(如"昨天出去吃饭了")
  • 评论前检查历史,避免矛盾

详见: docs/memory-system.md


4. 手动提醒与维护

默认策略:

  • 不自动安装 cron
  • 不自动修改 crontab
  • 仅生成手动提醒模板

可手动执行的维护项:

  • 每日热点总结
  • 记忆清理预览
  • 记忆清理执行(需显式 --apply

生成手动模板:

./scripts/setup-cron.sh

详见: docs/cron-jobs.md


5. 刷推流程

⚠️ 重要规则(必须遵守):

  1. 只在 Following 的 Recent 页面评论(不是 Popular)
  2. 评论前检查历史(避免重复评论同一博主)
  3. 记录所有评论(保存到历史文件)
  4. 评论发送前必须得到用户确认

详见: docs/comment-rules.md(必读!)

For You 页面:

  1. 浏览(自然阅读节奏)
  2. 关注建议(根据配置条件)

Following 页面:

  1. 确保是 Recent(不是 Popular)
  2. 点赞建议(有价值的推文)
  3. 评论建议(2小时内,使用 persona 风格)
  4. 用户确认后执行
  5. 记录评论到历史(避免重复)

详见: docs/comment-generation.md


6. 浏览器操作(基于 Browser Relay)

使用 browser-relay-cli

  • 连接你们自己的 Browser Relay 运行时
  • 复用本地已登录 Chrome/Chromium
  • 支持 DOM 优先、截图兜底的受控操作
  • 仓库:https://github.com/jasonCodeSpace/browser-relay

核心操作:

  • 读取推文
  • 生成评论建议
  • 点赞/关注/评论确认后执行
  • 滚动与截图验证

安全边界:

  • 不默认安装持久任务
  • 不默认自动发送评论
  • 不把 Browser Relay 当 stealth bot

详见: docs/browser-operations.md


使用示例

首次使用

用户: 刷推
Bot: 开始 Onboarding...
     1. 检查浏览器...
     2. 请选择 persona...
     3. 学习中...
     4. 配置刷推习惯...
     5. 完成!开始刷推...

后续使用

用户: 刷推半小时
Bot: 读取配置...
     For You: 给出 6 个建议关注账号
     Following: 给出 6 条候选互动
     你确认后我再执行...

关键特性

特性说明
完整 Onboarding5阶段引导,学习 persona
自然节奏与确认阅读节奏、二次确认、频率建议
记忆系统评论历史、用户信息、热点表格
手动维护无自动 cron,仅手动模板
避免矛盾评论前检查历史记录
Browser Relay 集成使用本地 browser-relay-cli

自我进化系统

核心理念

没有记忆的AI,只是一个聪明的工具。 有记忆且能进化的AI,才是会成长的伙伴。

进化闭环

采集数据 → 分析对比 → 得出结论 → 更新规则 → 下次执行

三大机制

1. Playbook 系统

  • playbooks/comment-strategies.md - 记录有效/无效策略
  • playbooks/changelog.md - 记录策略变更
  • Agent 可以更新自己的规则

2. 数据采集

  • data/engagement/YYYY-MM-DD.json - 每日评论数据
  • 记录:时间、作者、内容、结果
  • 用于后续分析和优化

3. 每日复盘(22:00)

  • 统计今日数据
  • 分析有效策略
  • 更新 Playbook
  • 生成明日建议
  • 推送报告给用户

文件结构

x-engagement/
├── playbooks/
│   ├── comment-strategies.md  # 评论策略(有效/无效)
│   └── changelog.md           # 策略变更记录
├── data/
│   └── engagement/
│       └── YYYY-MM-DD.json    # 每日评论数据
└── scripts/
    └── daily-review.sh        # 每日复盘脚本

使用示例

Agent 学习过程

  1. 发现「妙啊」评论效果好
  2. 在 Playbook 中记录:「妙啊」适用于技术分享,数据支撑:2026-03-02
  3. 下次刷推时读取这条规则
  4. 考虑在类似推文上使用相同策略

进化效果

  • Agent 越用越聪明
  • 自动学习什么评论有效
  • 持续优化策略
  • 避免重复错误

必读文档

按顺序阅读:

  1. docs/onboarding.md - 了解首次运行流程
  2. docs/human-behavior.md - 了解人类行为模拟
  3. docs/memory-system.md - 了解记忆系统
  4. docs/comment-generation.md - 了解评论生成

版本: 4.0.0 更新: 2026-03-02 改进: 结构化设计 + 记忆系统 + 定时任务 + 人类行为规范

Comments

Loading comments...