Automated Research Scouter

A single-file Python scanner that pulls hot posts from your configured subreddits, the top 30 Hacker News stories, and top GitHub repos matching your search — then scores each item, drops what is below threshold, and deduplicates against every previous run via a local seen_hashes.json. You only ever see new things.

Scoring is +2 per keyword match plus an engagement bonus capped at 5. Output is JSON for feeding an agent or Markdown for reading. The high-leverage last step is handing that output to your agent with a prompt like "give me a setup angle, a money angle, and an action trigger for each item" — EXAMPLE-OUTPUT.md shows the briefing format that produces.

Honest note carried from the pack: Reddit returns 403 on its .json endpoints for unauthenticated clients as of 2026, regardless of User-Agent. scanner.py tries JSON first and falls back to Reddit’s public Atom feed, which still works — feed items carry no score or comment counts, so they are judged on keywords alone. HN and GitHub are unaffected.

• Zero dependencies beyond the Python standard library — and $0 to run
• No API keys needed; optional GitHub token only for higher rate limits
• Cross-run dedup, so a scheduled scan never repeats itself
• Bonus scripts/advanced/ — the original production scripts from the Helio system