What problem does it solve? Engineering blogs from AI labs publish far too much to read manually, and only a handful of posts per week actually explain how autonomous agents work. This Skill monitors 15 sources (Anthropic, OpenAI, Cursor, LangChain, Amp, Cognition, DeepMind and others), filters out product announcements and model releases, and delivers a digest of fundamental posts on agent memory, long-running loops, orchestration, run economics, and evals. ## Core Features & Use Cases - Deterministic collection: A stdlib-only Python script scrapes RSS feeds and HTML indexes over a date window (default: last full week), deduplicates URLs, and enriches titles and dates via JSON-LD metadata. - Rubric-based triage: A written relevance rubric with three buckets and a 4-axis scoring matrix separates fundamental architecture posts from feature announcements, with a canon of landmark publications for novelty assessment. - Structured digest output: Produces a Markdown report with a top-3 must-read section including a "what to take away" line per post, a worth-a-glance list, rejection counts, and explicit reporting of silent or failed sources. - Use Case: Ask for a weekly digest of agent research; the Skill collects candidates, reads the shortlist, and writes a report file you can archive or forward to Telegram. ## Quick Start Ask the agent to run the radar and show what fundamental posts about AI agents came out in the blogs last week.
