← 返回 Skills 市场
Dailynewsreport
作者
FrankLe1117
· GitHub ↗
· v1.0.2
· MIT-0
384
总下载
1
收藏
2
当前安装
3
版本数
在 OpenClaw 中安装
/install daily-news-report
功能描述
定时抓取并整合多源新闻,智能去重分类,改写成行业研报级一句话新闻,输出结构化日报并支持Telegram推送。
使用说明 (SKILL.md)
Daily News Brief
Purpose
每日定时抓取、处理和推送新闻资讯的 skill。支持全量模式和重点领域模式,自动去重、分类、排序和改写新闻,输出结构化的日报。
Capabilities
- 定时抓取多个新闻源(财联社、IT之家、36氪等)
- 智能去重:URL去重、标题规范化去重、多源事件优先级去重
- 显式规则分类:政策类、上市公司、非上市公司/产业动态
- 一句话新闻改写:保留关键数据,行业研报级表达
- 智能排序:按分类优先级、主题相关性、数据完整性、时间新鲜度
- 支持全量模式和重点领域监控模式
- Markdown格式输出,支持Telegram推送
Configuration
config/default.json
{
"mode": "all", // "all" 或 "focused"
"focus_topics": ["robot", "real_estate", "ai"],
"sources": {
"primary": ["cls", "ithome", "36kr"],
"backup": ["jiemian", "yicai", "zqrb", "xinhua", "ssnews", "mnjs", "gov"]
},
"schedule": ["08:00", "17:30", "22:30"],
"max_items": 15,
"include_links": true,
"timezone": "Asia/Shanghai",
"include_keywords": [],
"exclude_keywords": ["广告", "推广", "营销"]
}
Source Retention Rules
- 同一事件多来源时优先保留财联社
- 如果没有财联社,则保留信息最完整、关键数据最全的一条
- 最终日报里同一事件只保留一条
Classification Rules
政策类
标题或正文出现这些关键词时优先归为政策:
- 政府、政策、国务院、发改委、工信部、证监会、财政部、商务部
- 标准、规划、指导意见、通知、方案、办法、条例
上市公司
标题或正文里出现股票代码或明确识别为上市公司:
- (603xxx.SH)、(000xxx.SZ)、(688xxx.SH)、(xxxx.HK)
非上市公司 / 产业动态
不属于政策类,也不属于上市公司时归到这里
Processing Pipeline
定时触发 → 配置加载 → 新闻源抓取 → 数据标准化 → 去重 → 过滤 → 分类 → 改写 → 排序 → 截取 → 输出 → 推送
Fixed Rewrite Prompt
const REWRITE_PROMPT = `请将我提供的新闻按照以下规则改写成"一句话新闻":
1. 结构要求:
* 用 1–2 句完成整条新闻。
* 主题要明确,突出核心事件(如发布、布局、出货、融资、合作、产品突破等)。
2. 内容要求:
* 必须保留所有关键数据(如营收、出货量、占比、市场份额、融资金额、注册资本等)。
* 保持文字简洁、行业研报级表达。
3. 风格要求:
* 参考以下格式风格:
"【公司名】关键动作 + 场景落地/战略意义;关键数据(出货/市占率/营收/利润等)。"
同时要遵守:
- 不编造数字
- 不增加无依据判断
- 公司名、政策名、股票代码、技术术语必须保留准确`;
Output Format
# 每日日报|YYYY-MM-DD HH:mm
## 一、政策
- 【机构名】关键动作 + 场景落地/战略意义;关键数据(出货/市占率/营收/利润等)。
- 【机构名】关键动作 + 场景落地/战略意义;关键数据(出货/市占率/营收/利润等)。
## 二、上市公司
- 【公司名】关键动作 + 场景落地/战略意义;关键数据(出货/市占率/营收/利润等)。
- 【公司名】关键动作 + 场景落地/战略意义;关键数据(出货/市占率/营收/利润等)。
## 三、非上市公司 / 产业动态
- 【公司名/机构名】关键动作 + 场景落地/战略意义;关键数据(出货/市占率/营收/利润等)。
- 【公司名/机构名】关键动作 + 场景落地/战略意义;关键数据(出货/市占率/营收/利润等)。
Manual Triggers
daily-news-brief run- 立即执行一次daily-news-brief test- 测试模式运行(不发送推送)daily-news-brief config- 查看当前配置
Implementation Requirements
- Source Fetchers: 每个fetcher输出统一结构 {title, source, published_at, url, summary, full_text}
- Normalizer: 统一不同来源的数据格式
- Deduper: URL去重 + 标题规范化去重 + 语义相似去重预留
- Classifier: 显式规则分类,避免复杂NLP
- Rewriter: 固定prompt,先写占位函数
- Ranker: 按分类优先级、主题相关性、数据完整性、时间新鲜度排序
- Scheduler: 支持Asia/Shanghai时区的三个定时任务
- Deliverer: Telegram发送函数
安全使用建议
This skill is functionally coherent for collecting and pushing news, but exercise caution before installing:
- Telegram send uses a shell-invoked curl command that directly interpolates your bot token and chat id into a shell string. If those config values are ever attacker-controlled or contain unexpected characters, they could allow shell injection. Prefer a version that uses a proper HTTP client (axios is already a dependency) or ensure tokens are strictly validated/escaped.
- The skill requests env:read and file-system read/write permissions. Confirm the platform's sandboxing and that only the skill's own config is writable; avoid supplying other sensitive environment variables to this skill.
- The installer script expects OPENCLAW_SKILLS_DIR to be set and the upload docs reference a local developer path — treat those as developer leftovers, not instructions to expose local files. Do not point the skill at directories containing sensitive data.
- Review and set Telegram credentials only after confirming they are stored securely by your OpenClaw platform (do not paste them into shared logs). Consider running the skill in test mode first (daily-news-brief test) to verify behavior before enabling scheduled runs.
If you want higher assurance, request a small change from the author: replace child_process.exec + curl with a safe axios POST that sends JSON to the Telegram API and ensure all user-configurable strings are validated/escaped. With that change and removal/justification of unnecessary env reading, this would look coherent and lower-risk.
功能分析
Type: OpenClaw Skill
Name: daily-news-report
Version: 1.0.2
The skill bundle is a functional news aggregator designed to fetch, process, and deliver financial and tech news to Telegram. The code implements standard logic for deduplication, classification, and summarization, with news sources pointing to legitimate domains like cls.cn. While the Telegram delivery module (deliverer/telegram.ts) uses child_process.exec to invoke curl, it properly sanitizes the message content using encodeURIComponent. No evidence of data exfiltration, unauthorized persistence, or malicious prompt injection was found; the instructions in Skill.md are strictly aligned with the stated purpose of news processing.
能力评估
Purpose & Capability
Name/description describe a news aggregator + Telegram delivery and the code implements fetchers, normalizer, deduper, classifier, rewriter, ranker, scheduler, and a Telegram deliverer — all consistent with the stated purpose. Permissions like internet and file-system read/write are reasonable for this functionality. Minor oddities: package.json lists a 'crypto' npm dependency even though Node provides crypto built-in, and some upload docs reference a local developer path (/Users/lezi) which appears to be leftover developer instructions.
Instruction Scope
SKILL.md and code keep to the news pipeline, but the TelegramDeliverer uses child_process.exec to run a curl command built by interpolating the bot token and chat id directly into a shell string. Bot token/chat id are configurable values; inserting them without shell-escaping can lead to shell injection if those values are malicious or incorrectly formatted. No other instructions request broad system data, but install.sh requires OPENCLAW_SKILLS_DIR to be set which is expected for installation.
Install Mechanism
There is no external download/install spec; the package is instruction/code-based and uses an install.sh that copies files into the platform skills directory. No remote arbitrary URL downloads or archive extraction were observed. This is lower risk from an install-source perspective.
Credentials
The skill declares permission 'env:read' in skill.json but the runtime code does not read process.env for credentials (install.sh checks OPENCLAW_SKILLS_DIR during install). Telegram credentials are intended to be set via skill config. Requesting env:read may be unnecessary. Requested config keys for Telegram are proportionate, however file-system write/read permissions are broad but plausibly needed for config and local caching/logs.
Persistence & Privilege
The skill is not always-enabled and has normal autonomous invocation settings. It does not attempt to modify other skills or global agent settings. Its scheduler registers tasks inside its own scope. No excessive persistence privileges were requested beyond typical file-system and network access.
如何使用
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install daily-news-report - 安装完成后,直接呼叫该 Skill 的名称或使用
/daily-news-report触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.2
- No changes detected in files for version 1.0.2.
- Functionality, configuration, and documentation remain unchanged.
v1.0.1
- Initial release of the daily-news-report skill.
- Added full implementation for automated news fetching, normalization, deduplication, classification, rewriting, ranking, and reporting.
- Supports scheduled tasks, full/targeted modes, and Telegram output.
- Manual triggers for run, test, and config inspection are available.
v1.0.0
Daily News Brief v1.0.0
- 初始版本,支持定时多源新闻抓取、去重、分类、改写和推送每日结构化新闻摘要。
- 提供全量模式与重点领域监控,灵活配置新闻源、时间、输出数量及关键词过滤。
- 实现显式分类(政策、上市公司、产业动态),规范化新闻表达,保留核心数据。
- 支持一键手动触发、测试及配置查看,输出Markdown格式并可自动推送至Telegram。
- 明确流程分工:采集、标准化、去重、过滤、分类、改写、排序、输出与推送等数据处理环节。
元数据
常见问题
Dailynewsreport 是什么?
定时抓取并整合多源新闻,智能去重分类,改写成行业研报级一句话新闻,输出结构化日报并支持Telegram推送。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 384 次。
如何安装 Dailynewsreport?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install daily-news-report」即可一键安装,无需额外配置。
Dailynewsreport 是免费的吗?
是的,Dailynewsreport 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
Dailynewsreport 支持哪些平台?
Dailynewsreport 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 Dailynewsreport?
由 FrankLe1117(@frankle1117)开发并维护,当前版本 v1.0.2。
推荐 Skills