What it does

AI News Radar is a GitHub-Actions-powered news pipeline that scrapes RSS, OPML, changelogs, public feeds, and even agent-accessible email digests, then deduplicates and merges related stories into a static web UI. The whole thing runs without API keys, LLM quotas, or a backend server. Fork it, add your feeds, and GitHub Pages hosts your personal 24-hour AI briefing.

The interesting bit

The project’s real engine isn’t the scraper—it’s the “Scout Skill” (伯乐Skill), a built-in agent prompt designed to audit information sources before you commit to them. The idea: let a source run in isolation for a week, then have an agent judge whether it’s genuinely AI-relevant or just noisy. It’s news curation treated as a source-quality problem first, a content problem second.

Key highlights

• Story-line merging: v0.6 clusters the same event across multiple sources (official blog, RSS, X, aggregators) so you don’t read the same model release four times
• Source health dashboard: tracks per-feed AI relevance scores and “AI percentage”—a source that posts 200 items with 5% AI content gets flagged
• Agent-native maintenance: includes structured prompts for Codex/Claude Code to add feeds, debug pipelines, or redeploy without touching credentials in the repo
• Zero-cost default: core pipeline needs no API keys; optional X API, private OPML, and AgentMail integrations live in GitHub Secrets
• Graceful degradation: if the curated daily-brief.json fails to build, the UI falls back to raw signal lists rather than going blank

Caveats

• The “Scout Skill” is essentially a prompt engineering pattern, not an automated service—you still need to run the judgment loop manually or via agent
• X API and email digest features are documented but default-off; the free-tier budget research is dated May 2026, which appears to be a typo or forward-dated filename

Verdict

Worth a look if you’re tired of bloated AI newsletters and want a hackable, self-hosted alternative with actual source hygiene. Skip it if you just want a polished app—this is a configurable pipeline that expects you to tune your own feeds.