Xhunt Hot Tweets
v0.1.0从 XHunt 抓取 X/Twitter 热门推文榜单并输出中文摘要与互动数据。用户提到“最火推文/热门帖子/Top20/4小时最火/只要AI/AI最火/给我链接+摘要”等请求时使用。支持 group=global|cn、hours=1|4|24、tag=ai|crypto|全部,并可切换过滤模式(全部/仅AI产...
MIT-0
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description match the behavior in SKILL.md: it scrapes https://trends.xhunt.ai and returns links, one-line Chinese summaries, and engagement stats. No unrelated env vars, binaries, or installers are requested.
Instruction Scope
Instructions are focused on fetching the target URL (browser snapshot preferred, fallback to web_fetch), extracting ranking fields, applying simple filtering modes, and producing a strict output format. The skill does not instruct the agent to read arbitrary files, access other services, or exfiltrate secrets.
Install Mechanism
No install spec and no code files — instruction-only skill. Lowest install risk (nothing written to disk by the skill itself).
Credentials
No environment variables, credentials, or config paths required. Network access to trends.xhunt.ai is reasonable and proportional to the stated function.
Persistence & Privilege
Defaults: not always-included and model invocation allowed (normal). The skill does not request persistent system-wide changes or elevated privileges.
Assessment
This skill appears coherent and lightweight: it only needs network access to https://trends.xhunt.ai and prefers a browser-snapshot tool for extraction. Before installing, confirm that your agent runtime has the necessary web-scraping/browser snapshot capability and that you trust the upstream site (results come from that domain). Be aware that mode=all will include political/controversial posts if present; the skill will return whatever is on XHunt and will not invent data but may return NA for missing fields. Because it is instruction-only and requests no secrets or installs, direct credential exfiltration is unlikely — still review runtime permissions for network access and autonomous invocation if you want to limit automatic scraping frequency.Like a lobster shell, security has layers — review code before you run it.
latest
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
SKILL.md
XHunt 热门推文提取(v2)
目标:稳定输出“可直接转发/复盘”的热门推文清单(链接 + 一句话摘要 + 核心互动数据)。
1) 参数解析(先做)
默认参数:
group=globalhours=24tag=ailimit=20mode=all(不过滤政治)
按用户话术覆盖:
- “华语区”→
group=cn - “4小时/1小时/24小时”→
hours - “只要AI”→
tag=ai - “全部最火”→ 不传
tag - “Top10/Top20/Top30”→
limit - “只看AI产品/模型更新”→
mode=ai-product-only - “可以有政治争议”→
mode=all
URL 模板:
- 基础:
https://trends.xhunt.ai/zh/tweets?group=<group>&hours=<hours> - 有标签:在末尾追加
&tag=<tag>
2) 抓取策略(固定顺序)
- 优先
browser打开目标 URL +snapshot(refs=aria)。 - 从榜单卡片提取:排名、作者、链接、标题、views、likes、retweets、热度。
- 若 browser 失败,降级
web_fetch,并在输出首行标注:说明:本次使用降级抓取,字段可能不完整。
- 若两种抓取都失败,直接返回错误说明,不得编造数据:
说明:抓取失败(网络或页面结构变化),请稍后重试。
3) 过滤规则
mode=all
- 不做内容过滤(含政治争议)。
mode=ai-product-only
保留:
- 模型发布/更新(如 Claude、GPT、Qwen、Gemini 等)
- AI 产品与工具更新(如 Cursor、Perplexity、Notion AI、Devin 等)
- AI agent/workflow/开发工具实操内容
- AI 基础设施与算力(与 AI 直接相关)
剔除:
- 纯政治选举/党争(无 AI 主体)
- 纯加密炒作(无 AI 主体)
- 娱乐八卦与空洞内容
若过滤后不足 limit:
- 继续向后补齐;仍不足则如实返回实际条数,不编造。
4) 输出格式(严格)
首行口径:
数据源:XHunt 推文榜(<group> / <hours>h / <tag或全部>)
列表:每条固定 3 行
N) <tweet_url>摘要:<一句中文摘要>数据:<views>|<likes>|<retweets>|热度<score>
末尾补“观察”3条:
- 当前最强话题簇
- 高互动内容的共同结构
- 适合用户二次创作的切入角度
5) 质量门槛
- 不输出无链接条目。
- 不输出“内容为空”类摘要;遇到则跳过补位。
- 摘要必须是“信息点 + 结果/影响”,避免空话。
- 数据字段缺失时写“NA”,不要臆造。
- 若过滤后不足
limit,返回真实条数,并在末尾加一句:说明:符合条件条目不足 <limit>,已返回全部可用结果。
6) 快速模板(执行时可直接套)
用户:四小时最火帖子,只要AI,给我Top20,每个一条摘要
执行参数:
group=globalhours=4tag=ailimit=20mode=ai-product-only
Files
5 totalSelect a file
Select a file to preview.
Comments
Loading comments…
