AI News Digest | AI 科技新闻日报#

Zero-config AI news aggregation. No scripts, no APIs, no setup. From raw search results → polished, insight-ready report in one conversation.

---

Why This Skill Wins#

Problem with other toolsHow we solve itNeed RSS feeds, Python scripts, cron jobsPure prompt-driven, zero setupDumps raw links at youStructured categories with summariesNo deduplicationSmart merge: same event from 5 sources → 1 entryNo rankingSignificance-based prioritization (critical → minor)One-size output4 depth levels + custom topicsEnglish-onlyBilingual (Chinese + English) by default

---

Activation Triggers#

• "AI 新闻" / "科技日报" / "AI 动态"
• "What's new in AI today?"
• "追踪大模型/AI Agent/AI编程赛道"
• "分析 OpenAI/Google/Anthropic 最新动态"
• "AI 周报" / "weekly AI roundup"
• "AI 行业趋势"

---

Workflow: Daily Digest (Default Mode)#

Phase 1 — Discovery (8 parallel searches)#

Use web_search with count: 8 for each. Run all searches before proceeding.

#QueryPurpose1"AI breaking news today {current_year}"General breaking2"LLM model release launch {current_year}"New model releases3"AI startup funding investment {current_year}"Funding rounds4"人工智能 最新消息 今天"Chinese-language news5"AI product launch announcement"Product updates6"AI research breakthrough paper"Academic/research7"OpenAI Google Anthropic Meta AI update"Big tech moves8"AI regulation policy government"Policy changes

Focus mode — If user specified a topic, replace queries 5-8 with topic-specific searches:

TopicReplacement queriesAI Agents"AI agent framework launch", "autonomous agent 2026", "AI agent startup", "MCP tool use agent"AI Coding"AI coding assistant update", "Copilot Cursor Windsurf Claude Code", "AI developer tool", "code generation model"AI Video/Image"AI video generation Sora Runway Kling", "AI image Midjourney DALL-E", "AI creative tool", "generative AI art"AI Hardware"AI chip GPU NVIDIA AMD", "AI inference hardware", "edge AI chip", "AI data center"Robotics"humanoid robot", "AI robotics startup", "autonomous vehicle AI", "embodied AI"

Phase 2 — Extraction#

For each result that looks relevant:

1. Use web_fetch with maxChars: 3000
2. Extract: headline, 2-3 sentence summary, source name, date, significance level
3. Skip if: paywalled with no content, clearly outdated (>7 days), duplicate of already-extracted article, product was already released before the report period (verify launch date before listing as "new release")

Date filtering rules (STRICT — NO EXCEPTIONS):

• 默认只收录 当天（today）的新闻。用户说"今天"就是今天，说"3月31日"就是3月31日。
• 不要包含昨天、前天、上周的新闻。用户如果需要最近几天/几周的新闻，会明确说"最近3天"、"本周"、"最近一周"等。
• 每条新闻发布前必须确认发布日期是今天，不确定的标注"[日期待确认]"，不得放入头条。
• 空板块显示"暂无相关动态"，绝不用旧新闻填充。

Source credibility tiers (prefer higher-tier when multiple sources cover same story):

• 🥇 Tier 1: Official blogs (openai.com, blog.google, anthropic.com, ai.meta.com), arxiv.org
• 🥈 Tier 2: techcrunch.com, theverge.com, arstechnica.com, 36kr.com, theinformation.com
• 🥉 Tier 3: General news, blogs, social media posts

Phase 3 — Deduplication & Ranking#

Deduplication rules:

1. Same event, multiple sources → keep most detailed article, note other sources as "also reported by: ..."
2. Follow-up coverage of ongoing story → merge into one entry, update with latest development
3. Translation of same article (EN/CN) → keep in user's preferred language

Significance ranking:

LevelCriteriaExamples🔴 CriticalIndustry-changing eventNew GPT release, major AI safety incident, billion-dollar deal, government ban🟡 NotableSignificant but expectedProduct update, $10M+ funding, benchmark record, partnership🟢 RoutineNormal activityMinor feature, opinion piece, routine update, rumor

Phase 4 — Output#

Depth levels (user can specify, default: standard):

LevelContentBest forBriefHeadlines + one-liners onlyQuick glance, mobile readingStandardHeadlines + 2-3 sentence summaries + linksDefault daily useDeepFull summaries + context + analysisResearch, decision-makingExecutive3-5 bullet points onlyBusy leaders, standup prep

Standard depth output format:

markdown

# 🤖 AI 科技日报 | {YYYY-MM-DD 周X}

---

## 🔴 今日头条

### {Headline}
{2-3 sentences: what happened, why it matters, what's next}
🔗 [{Source}]({url}) | 也被报道: {Other Source 1}, {Other Source 2}

---

## 📰 行业动态
| 事件 | 要点 | 来源 |
|------|------|------|
| {Title} | {One-line summary} | [🔗]({url}) |

## 💰 投融资
- **{Company}** — {金额} {轮次} (领投: {Investor}) → 用途: {use of funds}
  🔗 [{Source}]({url})

## 🛠️ 技术突破
- **{What}** — {Why it matters, technical detail}
  🔗 [{Source}]({url})

## 📱 产品发布
- **{Product}** ({Company}) — {Key feature or change}
  🔗 [{Source}]({url})

## 📜 政策法规
- {What changed} — {Who is affected}
  🔗 [{Source}]({url})

---

💡 **趋势洞察**: {1-2 sentences connecting today's stories to broader patterns}
📋 **明日关注**: {Upcoming events, expected announcements, developing stories}

---

Advanced Modes#

Mode: Topic Deep-Dive#

User: "深入分析 {topic}" / "deep dive into {topic}"

1. Run 10+ focused searches from different angles
2. Fetch top 15 articles
3. Output a research brief:

markdown

# 📊 {Topic} 深度分析 | {Date}

## 背景概述
{What led to this topic becoming relevant — 3-4 sentences}

## 最新进展
{Timeline of recent events, newest first}

## 核心玩家
| 公司/团队 | 动作 | 策略方向 |
|-----------|------|----------|
| {Name} | {What they did} | {Where they're heading} |

## 技术分析
{Deeper look at the technology, 1-2 paragraphs}

## 市场影响
{How this affects the industry, investment landscape, etc.}

## 未来展望
{What to expect in next 3-6 months}

## 完整来源
1. [{Title}]({url}) — {Source}
2. ...

Mode: Competitor Radar#

User: "分析 {Company} 的最新 AI 动态"

markdown

# 🔍 {Company} 竞品雷达 | {Date}

## 近30天动态
| 日期 | 事件 | 类型 |
|------|------|------|
| {Date} | {What happened} | 产品/技术/人事/战略 |

## 产品路线分析
{Where their product is heading based on recent moves}

## 关键人才变动
{Notable hires or departures}

## 研究/专利
{Recent papers, patents, or open-source releases}

## 对我们的启示
{Actionable insights for the user's own strategy}

Mode: Weekly Roundup#

User: "AI 周报" / "weekly AI roundup"

Daily format + these extra sections:

markdown

## 📈 本周趋势
1. **{Trend 1}**: {Evidence from multiple stories}
2. **{Trend 2}**: {Evidence}

## 📊 数据亮点
- 融资总额: ${amount} across {n} rounds
- 新模型发布: {n} 个
- 重要产品: {highlights}

## 📅 下周看点
- {Upcoming event 1}
- {Expected announcement}
- {Ongoing story to follow}

---

Quality Guarantee Checklist#

Before delivering output, verify EVERY item:

• Every claim has a clickable source link
• No duplicate entries for the same event
• All sections present (use "暂无相关动态" for empty ones)
• Critical events are in the 头条 section
• No content older than 7 days
• Objectivity: facts only, no editorializing
• Language matches user preference (Chinese default, English on request)
• Proper nouns kept in English (GPT-4o, not "GPT-4哦")

Error Handling#

SituationActionSearch API fails for a querySkip, continue with remaining searchesArticle fetch returns emptyUse search snippet, mark "[摘要来自搜索结果]"All searches failReturn: "⚠️ 搜索服务暂时不可用，请稍后再试"No news in a categoryShow "暂无相关动态" — never omit the sectionSource is paywalledUse snippet, mark "[付费文章，仅摘要]"

Delivery Options#

TargetHowChat (default)Markdown inlineFeishu docfeishu_doc tool → create → writeFileSave to user-specified pathEnglishUser says "in English" → all labels switch to English

User Satisfaction Tips#

• Speed matters: Run searches in parallel (batch all web_search calls)
• Be concise: Don't pad with filler — every sentence should inform
• Add value: Don't just list news; connect dots in 趋势洞察
• Respect time: If user asks for "简报", don't give them a wall of text
• Follow up: If a story seems incomplete, note it in 明日关注