Medium Research

Research a topic on Medium without scraping JS-rendered pages or paying per-article. Phase 1 finds candidates from RSS tag feeds plus a content search. Phase 2 fans out one subagent per selected article to fetch the full text through the Freedium mirror, in parallel. The parent (caller) gets a compact structured report — one summary card per article — plus the full body text saved to disk for selective deep-reading.

Inputs

• <topic>: Required. The research topic, e.g. "claude code skills", "AI agents in production", "GraphRAG". Free-form phrase; the script derives tag slugs and search queries from it.
• --days=N: Optional. Recency window in days. Default 90. "Trendy" topics work well at 14–30; broader surveys want 90–180.
• --top-n=N: Optional. How many articles to fetch in full. Default 5. Capped at 10 to keep subagent fanout reasonable.

Goal

Return a single structured markdown document to the caller containing:

1. A one-line summary of search totals and slugs used
2. For each of the top N articles: title, author, date, paywall flag, key bullets, link, on-disk path to full body
3. A short table of other surfaced candidates that weren't deep-fetched

The caller then synthesizes themes/trends. Synthesis is not this skill's job — staying out of the synthesis lane keeps this skill reusable across very different research goals (article writing, due diligence, sentiment, competitive scan).