How eugeneyan/news-agents works
Project type and purpose
• This is an open-source, terminal-based news aggregation system.
• It leverages Amazon Q CLI as an agent orchestration harness, the Model Context Protocol (MCP) for parsing RSS feeds, and tmux to run multiple agents in parallel.
• Its goal is to fetch articles from various sources, summarize them with AI agents, and produce both individual feed summaries and one consolidated digest.
File structure
• Top level
– feeds.txt: list of feed URLs
– pyproject.toml, uv.lock: Python dependency and lockfiles
– context/: markdown files containing instructions/prompts for each agent role
– src/: Python modules
· main.py: entrypoint, spawns and supervises agents
· ainews.py, hackernews.py, techcrunch.py, wired.py, wsj.py: feed-specific fetch & parse logic
– summaries/: output directory for per-feed and main summaries
• Patterns
– Orchestration vs. worker code are separated: main.py vs feed modules
– Context files decouple prompt definitions from code
README for additional insights
• Architecture overview section describes:
• Main Agent reads feeds.txt, splits URLs into three chunks
• Three Sub-Agents run in parallel (in tmux panes) to process chunks
• Each Sub-Agent fetches, parses, summarizes, and writes to summaries/
• Main Agent collects all sub-summaries and produces main-summary.md
• Technology stack: Amazon Q CLI, MCP, tmux, Python with httpx and mcp[cli]