← Back to Skills Marketplace
yujietech

Just internal test skill

by yujietech · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
91
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install ops-comment
Description
小红书自动化养号互动 Skill(OpenClaw 多模型版)。 当用户提到小红书养号、自动互动、点赞、收藏、关注、评论引流、 账号活跃度提升、互动任务、定时养号、多账号管理时必须使用。 核心:在用户已登录会话内模拟真人浏览与互动行为,按配置的速率、 抖动、过滤器与每日上限执行点赞/收藏/关注/评论四类动作。 纯...
README (SKILL.md)

小红书自动化养号互动 Skill — xhs-nurture v1.0

在用户已登录的小红书 Web 会话内,模拟真人浏览与互动行为, 自动执行点赞、收藏、关注、评论四类动作,完成养号引流。


模型能力检测(首次加载时执行)

在执行任何任务前,先探测当前可用工具集,确认处于 advanced 层级。

def detect_tier() -> str:
    required_tools = ["navigate", "read_page", "find", "javascript_tool", "computer"]
    has_all = all(tool_available(t) for t in required_tools)
    if has_all:
        return "advanced"   # 完整浏览器自动化(本 Skill 要求)
    return "unsupported"    # 本 Skill 不支持降级运行

如果层级为 unsupported,告知用户:

"本 Skill 需要 OpenClaw 浏览器工具链(navigate, read_page, find, computer 等)。请确认当前环境为 OpenClaw/Cowork 并已连接浏览器。"


任务入口路由

用户触发本 Skill 时,根据意图路由到对应流程:

用户意图 路由目标 参考文档
"开始养号" / "执行互动" → §互动主流程 references/interaction-engine.md
"配置/修改策略" → §配置管理 config/nurture-config.yaml
"查看数据/报告" → §数据看板 references/dashboard.md
"切换账号" → §多账号切换 references/multi-account.md
"设置定时任务" → §定时调度 references/scheduler.md
"停止/暂停" → §会话控制 本文件 §会话控制

互动主流程

Step 1: 登录态验证

1. navigate 到 https://www.xiaohongshu.com
2. read_page 获取页面结构
3. find("用户头像") 或 find("登录按钮")
4. 判断登录状态:
   - 找到头像 → 登录有效,继续
   - 找到登录按钮 → 未登录,通知用户手动登录后重试
   - 页面异常 → 等待 5s 重试,3次失败则终止

Step 2: 加载配置

1. 读取 config/nurture-config.yaml(全局配置)
2. 读取 config/profiles/{当前账号}.yaml(账号配置)
3. 读取 data/nurture-log/ 中今日已有日志,恢复计数器
4. 计算剩余配额:
   remaining = daily_limit - already_done_today
   如果 remaining \x3C= 0 → 通知用户今日配额已满,终止

Step 3: 预热阶段

目的:模拟真人打开 App 先浏览一会儿的行为,避免一上来就开始互动。

1. 停留在首页 3-8 秒(随机)
2. 缓慢滚动浏览 2-4 条笔记(只看不动作)
3. 随机点开 1 条笔记阅读 5-15 秒
4. 返回首页
5. 预热总时长:2-5 分钟

滚动模拟(所有滚动统一逻辑):

scroll_params:
  direction: "down"
  amount: random(2, 5)            # 滚动幅度随机
  pause_after: gaussian(3.0, 1.5) # 滚动后停顿(秒)
  reverse_probability: 0.1        # 10% 概率回滚

Step 4: 选择任务路径

根据配置中的 mode 参数选择路径:

路径 A: 发现页模式 (discover_feed)

1. 确认在首页推荐流
2. 滚动加载新内容
3. 对每条可见笔记:
   a. read_page 获取笔记卡片信息(标题、点赞数、作者)
   b. 过滤器判断是否符合目标(见 references/filters.md)
   c. 符合 → 点击进入详情页 → 执行互动动作
   d. 不符合 → 跳过,继续滚动
4. 每处理 3-5 条后,休息 15-30 秒

路径 B: 搜索页模式 (search)

1. navigate 到搜索页或点击搜索框
2. 从 config.targets.keywords 中随机选取关键词
3. 输入关键词搜索(模拟逐字输入)
4. 在搜索结果中按过滤器筛选
5. 对符合条件的笔记执行互动
6. 每个关键词处理 5-10 条后换下一个关键词

路径 C: 用户主页模式 (user_profile)

1. 从 config.targets.competitor_accounts 中选取目标用户
2. navigate 到该用户主页
3. 浏览其粉丝列表或最近互动用户
4. 对符合过滤条件的用户:
   a. 进入其主页
   b. 浏览 1-3 条笔记
   c. 执行点赞/收藏/关注
5. 每处理 3 个用户后休息

路径 D: 评论区模式 (comment_section)

1. 找到目标赛道的热门笔记(搜索 + 按热度排序)
2. 进入笔记详情页
3. 滚动浏览评论区
4. 生成 AI 评论(见 references/comment-generation.md)
5. 点击评论框 → 逐字输入 → 发送
6. 每条评论后等待 ≥ 90s

Step 5: 执行互动动作

通用执行逻辑(所有动作共享):

def execute_action(action_type, target):
    # 1. 限额检查
    if counters[action_type] >= daily_limits[action_type]:
        return "limit_reached"
    
    # 2. 间隔检查
    elapsed = now() - last_action_time
    min_interval = get_min_interval(action_type)
    if elapsed \x3C min_interval:
        wait(min_interval - elapsed + jitter())
    
    # 3. 动作前停留(模拟阅读)
    dwell_time = random_gaussian(mean=8, std=3, min=3, max=20)
    wait(dwell_time)
    
    # 4. 执行动作(见各动作详细逻辑)
    result = perform_action(action_type, target)
    
    # 5. 动作后处理
    if result.success:
        counters[action_type] += 1
        log_action(action_type, target, success=True)
    else:
        handle_failure(result.error)
    
    # 6. 连续动作检查
    if counters["total"] % cooldown_threshold == 0:
        cooldown(random(120, 300))
    
    return result

各动作详细逻辑

点赞 (like)

1. find("点赞按钮") 或 find("爱心图标")
2. 检查是否已点赞(按钮状态/颜色)
3. 已点赞 → 跳过
4. 未点赞 → 鼠标移动到按钮(贝塞尔曲线)→ 随机偏移 ±3px → 点击
5. 等待 500ms 确认状态变化
6. 验证点赞成功(按钮变色/数字+1)

收藏 (collect)

1. find("收藏按钮") 或 find("星标图标")
2. 检查是否已收藏
3. 逻辑同点赞,但频率更低(点赞:收藏 ≈ 4:1)
4. 收藏前额外停留 2-5s(模拟"觉得值得收藏"的决策时间)

关注 (follow)

1. 进入目标用户主页(如果不在的话)
2. find("关注按钮")
3. 检查是否已关注
4. 未关注 → 先浏览用户主页 5-10s → 浏览 1-2 条笔记
5. 返回主页 → 点击关注
6. 关注后在其主页再停留 3-5s
7. 关注间隔 ≥ 60s

评论 (comment)

1. 调用 AI 生成评论(见 references/comment-generation.md)
2. find("评论输入框") → 点击激活
3. 逐字输入评论:
   - 字符间隔:gaussian(120ms, 40ms)
   - 偶尔停顿 500-2000ms(模拟思考)
   - 5% 概率打错字然后删除重打
4. 输入完成后停顿 1-3s(模拟检查)
5. find("发送按钮") → 点击
6. 等待 1s 确认发送成功
7. 评论后等待 ≥ 90s 再执行下一动作

Step 6: 中场休息

每 15-20 分钟触发一次中场休息:
1. 停止互动动作
2. 纯浏览行为 3-8 分钟:
   - 随机滚动首页
   - 点开 1-2 条笔记看看(不互动)
   - 可能切到"我的"页面看看
3. 恢复互动

Step 7: 收尾阶段

触发条件:会话时长到达上限 OR 日配额用完 OR 用户主动停止
1. 逐步减速(最后 5 个动作间隔加倍)
2. 纯浏览 1-3 分钟
3. 保存状态到 data/nurture-log/
4. 生成会话摘要报告
5. 输出给用户:
   "本次互动完成:点赞 {n}、收藏 {n}、关注 {n}、评论 {n}
    耗时 {t} 分钟,无异常。今日剩余配额:点赞 {r}..."

会话控制

指令 行为
"暂停" 保存当前状态,停止动作,保持页面不关闭
"继续" 从暂停点恢复,重新验证登录态后继续
"停止" 执行收尾流程后完全终止
"状态" 报告当前计数、运行时长、剩余配额

异常处理

def handle_exception(error_type):
    match error_type:
        case "button_no_response":
            wait(30)
            retry(max=2)
            
        case "toast_warning":
            pause(minutes=5)
            log("warning", "平台 Toast 警告")
            
        case "captcha_detected":
            stop_immediately()
            notify_user("检测到验证码,已暂停所有操作。请手动完成验证后告知我继续。")
            
        case "logged_out":
            stop_immediately()
            save_state()
            notify_user("登录态已失效,请重新登录后告知我继续。")
            
        case "violation_notice":
            stop_immediately()
            save_state()
            notify_user("检测到违规提示,已终止操作。建议 24 小时内不再执行自动互动。")
            
        case "page_error":
            retry(max=3, backoff="exponential")
            if still_failing:
                stop_and_notify()

鼠标移动模拟

所有鼠标移动使用贝塞尔曲线(见 references/anti-detection.md):

move_to_element(target):
  1. 获取当前鼠标位置 (x0, y0)
  2. 获取目标元素中心 (x1, y1) + random_offset(±3px)
  3. 生成 2 个随机控制点(贝塞尔曲线)
  4. 沿曲线分 15-25 步移动
  5. 每步间隔 10-30ms(模拟手速)
  6. 最后 3 步减速(模拟精确对准)
  7. 添加 ±1px 微颤抖

配置管理

当用户要求修改配置时:

1. 读取当前 config/nurture-config.yaml
2. 根据用户指令修改对应字段
3. 验证配置合法性(限额不超安全阈值)
4. 写回配置文件
5. 确认修改内容给用户

安全阈值上限(不可超过):

  • 每日点赞 ≤ 200
  • 每日收藏 ≤ 50
  • 每日关注 ≤ 30
  • 每日评论 ≤ 20
  • 单次会话 ≤ 60 分钟

数据输出

每次会话结束后自动:

  1. 追加操作日志到 data/nurture-log/{date}.jsonl
  2. 如果是当天最后一个会话,生成日报 HTML
  3. 周日会话结束后额外生成周报

用户可随时请求查看数据:

  • "看今天的数据" → 输出今日统计摘要
  • "生成报告" → 生成并打开 HTML 看板
  • "看趋势" → 输出近 7 天对比数据
Usage Guidance
Install only if you knowingly want an agent to control a logged-in Xiaohongshu account for automated engagement. The main risk is not hidden malware in code, but the documented behavior: automated public actions, anti-detection techniques, scheduled multi-session use, and persistent local logs.
Capability Analysis
Type: OpenClaw Skill Name: ops-comment Version: 1.0.0 The xhs-nurture bundle is a comprehensive automation tool for Xiaohongshu (XHS) designed to simulate human interaction (likes, follows, and comments) within a user's browser session. It employs sophisticated anti-detection measures, such as Bezier-curve mouse movements, Gaussian-distributed typing rhythms, and randomized scrolling behaviors (detailed in references/anti-detection.md). The skill includes robust rate-limiting, safety thresholds, and content filters (e.g., in config/nurture-config.yaml and references/rate-control.md) to prevent account flagging. All high-risk capabilities, such as mouse/keyboard control and DOM interaction, are strictly aligned with the stated purpose of social media account management, and there is no evidence of data exfiltration, credential theft, or malicious prompt injection.
Capability Assessment
Purpose & Capability
The stated purpose is coherent, but it is high-impact: it automates public social-media interactions for account nurturing and traffic attraction using the user's logged-in account.
Instruction Scope
The workflow instructs the agent to navigate, click, type, follow, and send comments without clear per-action user confirmation, including generated public comments.
Install Mechanism
There is no install spec or executable code to scan, but the included dashboard template loads Chart.js from a remote CDN at runtime.
Credentials
The skill requires browser automation over an already-authenticated Xiaohongshu session, which gives the agent broad account authority despite no declared credential contract.
Persistence & Privilege
Artifacts describe logs, state files, multi-account profiles, cron-style scheduled runs, and failure recovery, creating ongoing automated account activity beyond a single user-directed action.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install ops-comment
  3. After installation, invoke the skill by name or use /ops-comment
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Initial release of the 小红书自动化养号互动 Skill (xhs-nurture) for the OpenClaw toolchain. - Simulates human-like browsing and automated interaction (like, favorite, follow, comment) on logged-in Xiaohongshu web sessions. - Fully browser-based DOM operations; does not use headless browsers or API reverse engineering. - Includes configurable interaction rates, filters, daily limits, and multi-account management. - Robust session management, anomaly handling, and exception reporting. - Generates detailed logs, daily/weekly reports, and supports flexible user commands for statistics and configuration.
Metadata
Slug ops-comment
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is Just internal test skill?

小红书自动化养号互动 Skill(OpenClaw 多模型版)。 当用户提到小红书养号、自动互动、点赞、收藏、关注、评论引流、 账号活跃度提升、互动任务、定时养号、多账号管理时必须使用。 核心:在用户已登录会话内模拟真人浏览与互动行为,按配置的速率、 抖动、过滤器与每日上限执行点赞/收藏/关注/评论四类动作。 纯... It is an AI Agent Skill for Claude Code / OpenClaw, with 91 downloads so far.

How do I install Just internal test skill?

Run "/install ops-comment" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Just internal test skill free?

Yes, Just internal test skill is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Just internal test skill support?

Just internal test skill is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Just internal test skill?

It is built and maintained by yujietech (@yujietech); the current version is v1.0.0.

💬 Comments