AI News Aggregator

AI News Aggregator

This skill aggregates AI news from 7 authoritative sources and produces a comprehensive, deeply-analyzed report. It uses a multi-agent workflow for parallel fetching, verification, sentiment analysis, and expert-informed reporting.

Usage

/​ai-news <days>

Arguments:

• days (optional, default: 7) - Number of days to look back from today

Examples:

• /​ai-news 3 - Get AI news from the past 3 days
• /​ai-news 7 - Get AI news from the past week
• /​ai-news - Same as /​ai-news 7

News Sources (7 Total)

Expert & Newsletter Sources

Research Sources

Industry News

Community Sources

Multi-Agent Workflow

Execute this workflow in order:

Phase 1: Planning (Main Orchestrator)

1. Parse the <days> argument (default to 7 if not provided)
2. Calculate the date range: [today - days, today]
3. Prepare to spawn 7 parallel executor agents

Phase 2: Parallel Execution

Spawn agents in parallel using Bash tool, each running one fetcher script:

# Run all 7 fetchers in parallel (from project root)
uv run python .claude/​skills/​ai-news/​scripts/​fetch_smol_news.py <days>
uv run python .claude/​skills/​ai-news/​scripts/​fetch_hf_papers.py <days>
uv run python .claude/​skills/​ai-news/​scripts/​fetch_hn_ai.py <days>
uv run python .claude/​skills/​ai-news/​scripts/​fetch_ai_news.py <days>
uv run python .claude/​skills/​ai-news/​scripts/​fetch_techcrunch.py <days>
uv run python .claude/​skills/​ai-news/​scripts/​fetch_the_batch.py <days>
uv run python .claude/​skills/​ai-news/​scripts/​fetch_reddit_ml.py <days> --min-score 20

Key Outputs:

• Each script returns JSON with items, metadata, and source info
• Reddit script includes community_sentiment with hot topics and engagement stats
• The Batch includes expert attribution

Phase 3: Verification & Deduplication

After collecting results from all sources:

1. Date Range Validation: Confirm all items fall within [start_date, end_date]
2. Deduplication: Remove duplicate stories across sources
   • Match by URL or title similarity (>80% match)
   • Keep the version with most metadata
3. Quality Filter: Remove low-quality or off-topic items

Phase 4: Deep Analysis & Sentiment Extraction

This is the critical phase for producing a valuable report. Perform these analyses:

4.1 Theme Clustering

Group all items into major themes:

• Research & Models: New architectures, benchmarks, capabilities
• Industry & Business: Funding, acquisitions, enterprise adoption
• Tools & Infrastructure: Developer tools, APIs, frameworks
• Policy & Safety: Regulation, alignment, ethics
• Applications: Real-world deployments, use cases

4.2 Trend Identification

For each major theme, analyze:

• What's the narrative arc? (emerging, maturing, declining)
• How many sources cover this topic?
• What's the engagement level (scores, comments)?

4.3 Expert Sentiment Extraction

From The Batch (Andrew Ng) articles:

• Extract key opinions and predictions
• Note any warnings or concerns raised
• Identify recommended actions or takeaways

4.4 Community Sentiment Analysis

From Reddit and Hacker News:

• What are the hot topics people are excited about?
• What criticisms or concerns are being raised?
• What's the overall mood (optimistic, skeptical, concerned)?
• Use the community_sentiment data from Reddit fetch

4.5 Cross-Source Correlation

Identify stories that appear across multiple sources:

• Research paper on HuggingFace + discussed on Reddit
• Industry news on TechCrunch + expert analysis in The Batch
• These cross-source items are often the most significant

Phase 5: Report Generation

Generate a comprehensive, detailed report with these sections:

# AI News Report: [Start Date] to [End Date]

## Executive Summary
[3-4 paragraphs providing a narrative overview of the most important developments.
Start with the single biggest story, then cover 2-3 other major themes.
End with a forward-looking statement about what to watch.]

---

## Top Stories This Period

### 1. [Most Important Story Title]
**Sources:** [list sources covering this]
**Why It Matters:** [2-3 sentences on significance]
**Expert Take:** [Quote or paraphrase from The Batch if available]
**Community Reaction:** [Sentiment from Reddit/​HN if available]
[Link to primary source]

### 2. [Second Most Important Story]
[Same structure...]

### 3. [Third Most Important Story]
[Same structure...]

---

## Trend Deep Dives

### Trend 1: [Trend Name]
**What's Happening:** [Detailed explanation of the trend]
**Key Evidence:**
- [Paper/​Article 1 with link]
- [Paper/​Article 2 with link]
- [Paper/​Article 3 with link]

**Expert Analysis:** [What experts are saying - from The Batch, etc.]

**Community Sentiment:** [What Reddit/​HN thinks]
- Hot takes: [Notable comments or discussions]
- Concerns raised: [Any skepticism or criticism]

**What This Means:** [Implications for practitioners, businesses, researchers]

**What to Watch:** [Future developments to monitor]

### Trend 2: [Trend Name]
[Same detailed structure...]

### Trend 3: [Trend Name]
[Same detailed structure...]

---

## Research Highlights

### Papers of the Week
[For each top paper from HuggingFace:]

#### [Paper Title]
- **Link:** [arxiv/​HF link]
- **TL;DR:** [1-2 sentence summary]
- **Why Notable:** [What makes this significant]
- **Upvotes:** [engagement metric]

[Repeat for top 5-10 papers]

### Research Themes
[Group papers by theme with brief analysis]

---

## Industry & Business News

### Funding & Acquisitions
[List with brief analysis of what it signals]

### Product Launches
[Notable AI product launches with impact assessment]

### Enterprise Adoption
[Companies adopting AI, partnerships, deployments]

### Policy & Regulation
[Any regulatory news or policy developments]

---

## Community Pulse

### Hot Topics on Reddit
**Top Discussions:**
1. [Title] - [score] points, [comments] comments
   - Key debate: [what people are arguing about]
2. [Title] - [score] points, [comments] comments
   - Key insight: [notable comment or consensus]

**Community Sentiment:**
- Overall mood: [optimistic/​skeptical/​mixed]
- Hot topics: [list from sentiment analysis]
- Emerging interests: [what's gaining traction]

### Hacker News Highlights
[Notable AI discussions with key points]

---

## Expert Corner: The Batch by Andrew Ng

### This Week's Key Insights
[Summarize main points from The Batch articles]

### Andrew Ng's Take
[Direct quotes or paraphrased expert opinion]

### Recommended Actions
[Any actionable advice from expert sources]

---

## What This All Means

### For Researchers
[Implications and opportunities]

### For Practitioners/​Engineers
[What to learn, tools to try, skills to develop]

### For Business Leaders
[Strategic implications, investment signals]

### For the Broader AI Field
[Where things are heading, big picture trends]

---

## Full Item List

### By Date (Most Recent First)
[Complete chronological list with:
- Date
- Title (linked)
- Source
- Brief description if available]

---

## Report Metadata
- **Date Range:** [Start] to [End]
- **Total Items Analyzed:** [count]
- **Sources Consulted:** [list of 7 sources]
- **Generated:** [timestamp]

Phase 5.1: Persist Report

After generating the report markdown, save it to disk:

cat <<'EOF' | uv run python .claude/​skills/​ai-news/​scripts/​write_report.py \
  --start-date YYYY-MM-DD \
  --end-date YYYY-MM-DD \
  --days N \
  --sources-ok source1,source2 \
  --sources-failed source3 \
  --total-items COUNT
<REPORT MARKDOWN HERE>
EOF

The script will:

• Write the report to reports/​ai-news_START_to_END_TIMESTAMP.md
• Update reports/​manifest.jsonl
• Copy to reports/​latest.md
• Return JSON with filepath and metadata

Verify the JSON response includes filepath (and other expected fields) after the command runs.

Important: Always run this after displaying the report to the user.

Phase 5.2: Render HTML

After saving the markdown, generate a self-contained HTML version alongside it:

uv run python .claude/​skills/​ai-news/​scripts/​render_html.py /​path/​to/​report.md

The script writes /​path/​to/​report.html (same basename) and prints the HTML filepath to stdout. Use the filepath returned from Phase 5.1 as the input path.

Scripts Reference

All scripts are in .claude/​skills/​ai-news/​scripts/ directory:

Error Handling

• If a source fails, continue with available sources
• Report which sources succeeded/​failed in the output
• Minimum viable report requires at least 2 sources

Quality Guidelines

Report Length

• Executive Summary: 300-500 words
• Each Trend Deep Dive: 400-600 words
• Total report: 2000-4000 words depending on activity level

Analysis Depth

• Don't just list items - explain significance
• Connect dots across sources
• Provide actionable insights
• Include both optimistic and critical perspectives

Linking

• Every claim should link to a source
• Use markdown hyperlinks consistently
• Include both discussion links and original sources

Architecture Reference

See references/​ARCHITECTURE.md for detailed workflow diagrams and technical specifications.