morning-briefing

name	morning-briefing
description	Generate a daily AI research + market + news briefing. Use when setting up automated morning briefings, research digests, or daily knowledge feeds. Covers arxiv papers, tech blogs, stock watchlist, industry news, and paper recommendations. Outputs Kindle PDF + channel message. Configurable topics, sources, stocks, and delivery schedule. Optionally uses Amazon Bedrock for AI-powered synthesis and summarization.
inputs	{"config_path":{"type":"string","required":true,"description":"Path to configuration YAML file"},"dry_run":{"type":"boolean","default":false,"description":"Generate briefing without sending email"}}
outputs	{"pdf_path":{"type":"string","description":"Path to generated PDF file"},"markdown_path":{"type":"string","description":"Path to generated markdown file"},"status":{"type":"object","description":"Run status (papers_found, blogs_found, stocks_fetched, news_found, errors, elapsed_seconds)"}}
triggers	{"schedule":"50 6 * * *","manual":"generate my morning briefing"}
requires	{"python":">=3.10","env":{"required":["FINNHUB_API_KEY","BRAVE_API_KEY"],"optional":["GMAIL_USER","GMAIL_APP_PASSWORD","AWS_ACCESS_KEY_ID","AWS_SECRET_ACCESS_KEY","AWS_REGION"]}}

Morning Briefing Skill

Generates a comprehensive morning briefing covering AI/ML papers (arxiv), tech blogs, stock watchlist, industry news, and paper recommendations for reproduction. Outputs as Kindle-optimized PDF with optional email delivery.

Enhanced with Amazon Bedrock for intelligent summarization, cross-section synthesis, and semantic paper scoring. Falls back gracefully to deterministic mode when Bedrock is unavailable.

Prerequisites

Required

• Python 3.10+ (python3 --version)
• pip (pip3 --version)

API Keys (all free tier)

Service	Purpose	Sign Up	Free Tier
Finnhub	Stock market data	finnhub.io	60 calls/min
Brave Search	News aggregation	brave.com/search/api	2000 queries/mo
Gmail App Password	Kindle email delivery	myaccount.google.com	Free

Optional

• AWS credentials -- For Amazon Bedrock intelligence features (~$2.45/month)
• Kindle Scribe/device -- For PDF delivery via email. See references/kindle_setup.md
• CJK fonts -- For Chinese/Japanese/Korean support in PDFs:

  # Ubuntu/Debian
  sudo apt install fonts-noto-cjk
  # macOS
  brew install font-noto-sans-cjk
  

System Dependencies

# Ubuntu/Debian
sudo apt install python3-venv python3-pip

# macOS
brew install python3

Features

Deterministic (no LLM required)

• ArXiv Paper Scanning: Tracks new papers on configured topics
• Blog Feed Monitoring: Aggregates updates from RSS feeds
• Stock Watchlist: Fetches market data for configured tickers (Finnhub API)
• News Aggregation: Collects top AI/tech headlines (Brave Search API)
• Paper Scoring: Ranks papers by reproduction value (code availability, topic match, recency)
• Cross-Section Deduplication: Removes duplicate content between news and blogs
• Kindle-Optimized PDF: 6x8 inch format with CJK support
• Email Delivery: Send directly to Kindle via SMTP
• Config Validation: Catches configuration errors at startup
• Status Reporting: Generates status.json for monitoring

Intelligence Layer (Amazon Bedrock, optional)

• Topic Expansion: Suggests related search queries using Nova Lite
• Paper Summarization: 1-2 sentence takeaways for each paper using Nova Pro
• Semantic Scoring: Relevance scoring using LLM understanding (beyond TF-IDF)
• Stock-News Correlation: Links stock movements to news drivers
• Reproduction Assessment: Evaluates compute, data, and feasibility for top papers
• Cross-Section Synthesis: Finds themes across papers, news, and blogs
• Editorial Intro: Opens briefing with today's key insight
• Market Trend Summary: 2-sentence market analysis with key drivers

Setup

1. Install Dependencies

It's recommended to use a virtual environment:

cd atlas-morning-briefing
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Or install as a package (enables morning-briefing CLI command):

pip install -e .

2. Configure API Keys

Set environment variables:

export FINNHUB_API_KEY="your_finnhub_key"
export BRAVE_API_KEY="your_brave_search_key"
export GMAIL_USER="[e-mail]"
export GMAIL_APP_PASSWORD="your_app_password"

For Bedrock features, configure AWS credentials:

export AWS_ACCESS_KEY_ID="your_key"
export AWS_SECRET_ACCESS_KEY="your_secret"
export AWS_REGION="us-east-1"

Or use an IAM role / AWS profile (recommended for EC2/Lambda).

3. Configure Topics and Sources

Edit config.yaml:

arxiv_topics:
  - "Agent Evaluation"
  - "Multi-Agent Systems"

blog_feeds:
  - name: "Anthropic"
    url: "https://www.anthropic.com/rss.xml"

stocks:
  - AMZN
  - GOOGL

news_queries:
  - "AI artificial intelligence"

kindle_email: "[e-mail]"
sender_email: "[e-mail]"

bedrock:
  enabled: true
  models:
    heavy: "us.anthropic.claude-sonnet-4-20250514-v1:0"
    medium: "amazon.nova-pro-v1:0"
    light: "amazon.nova-lite-v1:0"

See references/config_guide.md for full configuration options.

4. Set Up Kindle Email Delivery

See references/kindle_setup.md for instructions on configuring your Kindle email address.

Usage

Generate Briefing (Dry Run)

python3 scripts/briefing_runner.py --config config.yaml --dry-run

Generate and Send to Kindle

python3 scripts/briefing_runner.py --config config.yaml

Run Individual Scanners

Each scanner can be run independently:

python3 scripts/arxiv_scanner.py --config config.yaml --output papers.json
python3 scripts/blog_scanner.py --config config.yaml --output blogs.json
python3 scripts/stock_fetcher.py --config config.yaml --output stocks.json
python3 scripts/news_aggregator.py --config config.yaml --output news.json

Score Papers for Reproduction

python3 scripts/paper_scorer.py --input papers.json --config config.yaml --output scored_papers.json

Generate PDF Only

python3 scripts/pdf_generator.py --input briefing.md --output Atlas-Briefing.pdf --format kindle

Run Status

After each run, a status.json file is generated:

{
  "timestamp": "2026-03-06T06:50:12",
  "papers_found": 14,
  "blogs_found": 3,
  "stocks_fetched": 5,
  "news_found": 8,
  "intelligence_enabled": true,
  "errors": [],
  "pdf_generated": true,
  "email_sent": true,
  "elapsed_seconds": 7.2
}

Scheduling

Set up a daily cron job:

crontab -e

Add:

0 7 * * * /path/to/atlas-morning-briefing/run_briefing.sh >> /path/to/atlas-morning-briefing/logs/briefing.log 2>&1

Create wrapper script (run_briefing.sh):

#!/bin/bash
cd /path/to/atlas-morning-briefing
source venv/bin/activate
source .env  # Load API keys
python3 scripts/briefing_runner.py --config config.yaml

Cost Estimate

Without Bedrock: $0.00/month

All external APIs (ArXiv, Finnhub, Brave, Gmail) have free tiers sufficient for daily use.

With Bedrock (default: Claude Opus 4.6): ~$12-24/month

Default configuration uses Claude Opus 4.6 for all tiers for maximum quality.

Estimated cost per run: ~$0.40-0.80 | Monthly (30 daily runs): ~$12-24

For lower cost (~$0.08/run, ~$2.45/month), switch to tiered models in config.yaml:

bedrock:
  models:
    heavy: "us.anthropic.claude-sonnet-4-20250514-v1:0"
    medium: "amazon.nova-pro-v1:0"
    light: "amazon.nova-lite-v1:0"

File Naming

Configure in config.yaml:

file_naming: "Atlas-Briefing-{yyyy}.{mm}.{dd}"

Available variables:

• {yyyy}: Year (4 digits)
• {mm}: Month (2 digits)
• {dd}: Day (2 digits)
• {type}: Briefing type (defaults to "Daily")

Example output: Atlas-Briefing-2026.03.06.pdf

Paper Scoring Criteria

Papers are scored based on:

• has_code (weight: 5): Links to open source code repository
• topic_match (weight: 3): Cosine similarity to configured topics (TF-IDF)
• semantic_score (Bedrock): LLM-assessed relevance with explanation
• recency (weight: 2): Days since publication
• citation_count (weight: 1): Number of citations (if available)

Reproduction difficulty is estimated as S/M/L/XL based on:

• Dependencies complexity
• Dataset size
• Compute requirements

When Bedrock is enabled, reproduction assessment includes specific compute estimates and blocker identification.

Troubleshooting

No papers found

• Check arxiv_topics in config.yaml match arxiv categories
• Verify date range is not too narrow

PDF generation fails

• Ensure fonts are installed for CJK support
• Check markdown formatting is valid

Email delivery fails

• Verify GMAIL_USER and GMAIL_APP_PASSWORD are set
• Check sender_email matches GMAIL_USER
• Ensure Kindle email is whitelisted in Amazon account

API rate limits

• Finnhub: Free tier allows 60 calls/minute
• Brave Search: Check your plan limits
• Rate limiting is built-in (0.5s delay between Finnhub calls, 1.0s between Brave calls)

LLM call budget exhausted

• The intelligence layer has a per-run call budget (default: 20 calls)
• With all features enabled and many papers, this can be exceeded
• Increase via bedrock.max_calls_per_run in config.yaml:

  bedrock:
    max_calls_per_run: 30
  

• Typical usage: ~10-15 calls with default settings

Bedrock errors

• Verify AWS credentials are configured
• Check the model ID is available in your region
• Set bedrock.enabled: false to disable and run deterministically

Config validation errors

• The runner validates config at startup and reports specific errors
• Check the error message for the invalid field and expected type

Architecture

briefing_runner.py (orchestrator)
├── [Intelligence] Topic expansion (Bedrock Light)
├── arxiv_scanner.py → papers
├── blog_scanner.py → blogs
├── stock_fetcher.py → stocks
├── news_aggregator.py → news
├── [Dedup] Cross-section deduplication
├── [Intelligence] Paper summarization (Bedrock Medium)
├── [Intelligence] Semantic scoring (Bedrock Medium)
├── [Intelligence] Stock-news correlation (Bedrock Heavy)
├── paper_scorer.py → scored papers
├── [Intelligence] Reproduction assessment (Bedrock Medium)
├── [Intelligence] Cross-section synthesis (Bedrock Heavy)
├── [Generate markdown briefing with editorial content]
├── pdf_generator.py → Atlas-Briefing.pdf
├── kindle_sender.py → Email to Kindle
└── [Save status.json]

Exit Codes

• 0: Success
• 1: Partial failure (some scanners failed but briefing generated)
• 2: Total failure (unable to generate briefing or invalid config)

References

• references/config_guide.md: Full configuration reference
• references/kindle_setup.md: Kindle email setup instructions
• examples/sample-briefing.md: Example generated briefing