smithery.ai

news-fetch

Search and archive significant global events (7-day window). Topics: international conflicts, trade/China policy, and AI/Agent updates. Include entity tracking, semantic dedupe against news.jsonl, and atomic archival via recorder.

First seen Apr 10, 2026

Installation

$ npx skills add https://smithery.ai

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from smithery.ai · top by installs.

npx skills add https://smithery.ai

Browse all from smithery.ai

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 4,368 B
  • docs SUMMARY.md 248 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 1 installs

SKILL.md

News Fetch

用于一次性抓取“最近7天”的重大国际事件/政策/AI Agent 动态,并落库到 news 模块。

工作流(保持简洁)

  1. 动态日期锚定

- 以系统当前日期为 T,窗口为 T-7 至 T(含首尾)

  1. 确认范围

- 默认范围:国际冲突、贸易政策、中国政策、AI/Agent(含公司/实验室/论文) - 默认关注实体:OpenAI、Google/DeepMind、阿里、字节、腾讯、智谱、清华相关实验室、幻方量化 - 若用户指定主题(例如“Agent 论文”),以用户主题为主,并加入 scope.topics - 任务结束后将新增主题写回本技能的默认 topics 清单(保证下次可复用)

  1. 多路并行检索(按 Topic 拆分 Query)

- 必须使用 web.run,时间窗限定最近 7 天 - 每个 topic 单独构建查询,避免热点霸屏 - AI 领域优先:公司官方博客/研究发布 + arXiv - 政策领域优先:政府官网/央行/监管机构/权威媒体

  1. 候选列表 → 精选集合(List + Set)

- 先将检索结果放入“待看 list”(仅收集标题/日期/来源链接) - 逐条阅读并判断是否纳入“提炼 set” - set 采用语义去重(AI 判定是否同一事件)

  1. 语义去重(Smart Dedupe)

- 读取 workspace/records/news/news.jsonl 历史记录 - 由 AI 判断是否为同一事件(允许标题/日期轻微差异) - 来源权威度更高者优先保留 - 写入 dedupe 字段(策略+剔除数量)

  1. 原子化写入(Atomic Storage)

- 每条新闻单独写一条 news 记录(便于检索/RAG) - 不写 newsdigest,仅保留原子化条目 - 使用 recorder 脚本:recordjsonl.py --record-type news - 推荐通过 --extra 写入结构化内容

  1. News Fetch 自用 TODO

- 每次调用 News Fetch,都先基于模板更新 workspace/records/news/news-fetchtodolist.md - 清单至少包含:分 Topic 检索、去重、写入 news、补充 coveragegap、产出摘要 - 完成后将 TODO 归档到 workspace/records/news/newsarchive/YYYY/MM/DD/news-fetch_todo.md

  1. List / Set 文件位置

- 待看 list:workspace/records/news/stagingcandidates.jsonl - 提炼 set:workspace/records/news/curatedset.jsonl

  1. 归档 List / Set(完成任务后)

- 将 stagingcandidates.jsonlcuratedset.jsonl 归档到: - workspace/records/news/newsarchive/YYYY/MM/DD/stagingcandidates.jsonl - workspace/records/news/newsarchive/YYYY/MM/DD/curatedset.jsonl - 归档完成后再重置两个文件为空白(保留文件)。

最小可执行步骤(PowerShell)

# 1) 初始化/清空 list 与 set
"" | Set-Content -Path "workspace/records/news/staging_candidates.jsonl" -Encoding utf8
"" | Set-Content -Path "workspace/records/news/curated_set.jsonl" -Encoding utf8

# 2) 初始化 TODO
@'
# News Fetch TODO
- 分 Topic 检索
- 语义去重
- 写入 news(原子化)
- 补充 coverage_gap
- 产出摘要(可选)
'@ | Set-Content -Path "workspace/records/news/news-fetch_todo_list.md" -Encoding utf8

推荐记录结构(news)

{
  "date": "YYYY-MM-DD",
  "category": "AI 自动判定(可随主题变化)",
  "summary": "...",
  "tags": ["news", "ai", "agent"],
  "sources": [{"name":"...","url":"..."}],
  "entities": ["..."],
  "source_rank": "official|media|preprint"
}

记录示例(PowerShell)

$extra = @'
{"time_window":{"start":"2026-01-13","end":"2026-01-19","timezone":"local"},"scope":{"topics":["国际冲突","贸易政策","中国政策","AI与Agent"],"entities_watch":["OpenAI","Google/DeepMind","阿里","字节","腾讯","智谱","清华相关实验室","幻方量化"]},"dedupe":{"strategy":"title+source+date","dropped":0}}
'@
python .\\.codex\\skills\\recorder\\scripts\\record_jsonl.py --record-type news --title "单条新闻标题" --summary "单条新闻摘要" --tags "news,ai,agent" --module "news" --source "web" --extra $extra