A reusable Claude Cowork skill for building your own personalized intelligence feed — an auto-updating aggregator that collects, filters, scores, and presents the most relevant news for any topic you care about.
Built from the architecture of NeuralField, a live AI+Sports/Gaming intelligence site that runs entirely for free on GitHub Actions + Cloudflare Pages.
Point Claude at any topic — AI in finance, climate policy, biotech, gaming, whatever you follow — and it will guide you through building a production-ready personal feed from scratch:
- Find your feeds — curated direct RSS sources and Google News search queries for your domain
- Filter the noise — multi-layer relevance filtering (URL blocklist → domain gate → topic gate) so only truly on-topic articles make it through
- Score and rank — source authority tiers, recency decay, and keyword relevance combine into a single score per article
- Deduplicate — TF-IDF title similarity + URL normalization so the same story never appears twice
- Archive daily — rolling 7-day live feed with a 22-day daily snapshot archive
- Generate audio briefs — optional daily spoken summary via OpenAI TTS (~$7/year)
- Deploy for free — single static HTML file baked with all data, hosted on Cloudflare Pages via GitHub Actions
Anyone who wants a signal-to-noise ratio they actually control. If you've ever opened a news aggregator and found it full of off-topic stories, or missed important developments in a niche area because no one covers it well, this is the pattern for you.
No backend required. No subscription. No monthly bill.
NeuralField — AI × Sports & Gaming intelligence, updated nightly.
This is a Claude Cowork skill — a structured prompt that gives Claude the context and step-by-step methodology to guide you through building the system.
- Copy the
personal-intelligence-skill/folder into your Cowork workspace at.claude/skills/ - Start a new Cowork session and ask: "Help me build a personal intelligence feed for [your topic]"
- Claude will read the skill and guide you through the full process
The skill files are written to be readable on their own — browse SKILL.md for the full architecture guide, and the references/ folder for deeper dives into specific areas.
SKILL.md ← Main guide: domain definition, feeds, filtering,
scoring, dedup, static site, audio briefs, deployment
references/
├── pipeline.md ← Annotated nightly pipeline with ordering explanation
│ and common pipeline bugs
├── frontend.md ← Static site UI architecture (BAKED_DATA, filter chips,
│ archive accordion, audio player, CSS tokens)
└── feeds.md ← RSS feed discovery by domain, Google News query
formula, feed quality evaluation, staleness detection
A few things learned the hard way building NeuralField that are encoded into this skill:
Run filters before AND after fetching. If you blocklist an article and then call fetch_feeds(), the live RSS feed will re-ingest the blocked article. Always run retroactive_ai_cleanup() post-fetch too.
Generate the audio brief before the static site. The has_brief flag gets baked into the HTML at build time — if you generate the brief afterward, today's archive card won't show the listen link.
Use the user's local timezone, not UTC. A pipeline running at 23:00 UTC is 07:00 the next day in UTC+8. datetime.now() gives the wrong date for title, filenames, and display.
Direct RSS + Google News is a better combination than either alone. Direct feeds give you reliable high-quality coverage from known publications. Google News queries surface long-tail stories those publications don't cover. Both go stale in different ways — rotate Google News queries, add direct feeds when a category goes quiet.
| Layer | Tool | Cost |
|---|---|---|
| Pipeline runner | GitHub Actions | Free (2000 min/month) |
| Static hosting | Cloudflare Pages | Free (unlimited requests) |
| Data persistence | SQLite committed to git | Free |
| Audio briefs | OpenAI TTS (tts-1) |
~$7/year |
| Feed resolution | Python httpx |
Free |
| AI assistant | Claude via Cowork | — |
This skill is released under the MIT License. You're free to use it, fork it, adapt it for any topic, and share it — attribution appreciated but not required.
Created by Leon Chen as part of the Borderless Creators project — building tools at the intersection of AI and human curiosity.
Questions, improvements, or your own feed examples are welcome via GitHub Issues.