One question, every platform
last30days is a research skill that answers one thing fast: what are people actually saying about X right now. Instead of opening fifteen tabs, you ask once. It fans out across Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, and Threads at the same time, then hands back one synthesized brief. It scores by real engagement, upvotes, likes, real money on prediction markets, not by what an editor decided to feature.
- One query fans out to 9 live sources at once.
- Ranked by real engagement, not editorial curation.
- Returns a synthesis you read, not a wall of raw posts.
Resolve, fan out, score, synthesize
Under the hood it runs four steps. First it RESOLVES the topic into the right handles, subreddits, and hashtags to search. Then it FANS OUT, querying every source in parallel. It SCORES and de-dupes, merging the same story across platforms and ranking by engagement. Finally it SYNTHESIZES the evidence into a brief plus a Best Takes section that surfaces the wittiest and most viral quotes.
- Resolve: find the real accounts, subs, and tags for the topic.
- Fan out: all sources searched at the same time.
- Score + dedupe: merge duplicate stories, rank by engagement.
- Synthesize: one brief + Best Takes, with citations.
Nine sources, free and paid
Four sources work the second you install it with zero keys: Reddit, Hacker News, Polymarket, and GitHub. The rest unlock with the keys we already own. X comes through the xAI API, not cookie scraping, so there is no account-ban risk. TikTok, Instagram, and Threads come through ScrapeCreators. YouTube needs yt-dlp for transcripts. You can include or exclude any source per query.
- Free, no key: Reddit, Hacker News, Polymarket, GitHub.
- X via xAI API (safe), TikTok/IG/Threads via ScrapeCreators.
- YouTube needs yt-dlp; web search via Brave/Exa (optional).
- Pick sources per query with --search.
Your first query
The simplest use is one line: the skill name and a topic. In Claude Code type a slash command, or run the script directly on Hetzner. No keys needed for the free sources, so the very first query works out of the box. Add --search to narrow which platforms it hits, and --quick or --deep to trade speed for recall.
- Slash command in Claude Code, or run the script on Hetzner.
- Free sources answer instantly, no setup.
- --quick for speed, --deep for higher recall.
/last30days nvidia earnings reactionpython3 ~/.claude/skills/last30days/scripts/last30days.py "AI video tools" \
--search reddit,x,youtube --emit compact --quickpython3 ~/.claude/skills/last30days/scripts/last30days.py --diagnoseModes and flags worth knowing
A few flags do most of the work. --emit html turns the research into a self-contained, shareable brief you can hand a client. --competitors auto-discovers and compares peer products. --store persists findings to a local SQLite store so you can track a topic over time and run digests. eli5 rewrites the synthesis in plain language on demand. Mix and match per query.
- --emit html: a shareable, self-contained brief.
- --competitors: auto-find and compare peers.
- --store: track a topic over time, run digests.
- eli5 on: plain-language rewrite of the synthesis.
/last30days [client niche] reviews and complaints --emit=html/last30days [our product] --competitors/last30days [topic] --storeRaven's wide-research lane
Inside the fleet, last30days is wired to Raven, our research and intel lead. When a job needs current sentiment, competitor chatter, or voice-of-customer before content, Raven reaches for it instead of opening tabs. It runs on Hetzner per remote-default routing, so nothing heavy or scrapey touches the home machine. The output feeds straight into content, SEO, and strategy work.
- Raven (research lead) owns it as a wide-research tool.
- Runs on Hetzner, off the home IP, behind our keys.
- Feeds content, competitor intel, and VoC research.
Have Raven run last30days on [topic] and give me the synthesis + the 3 best takes, not the raw dump.I need current sentiment on [topic] before we write. What exact question should we hand the research tool, and which platforms matter most here?Wired safe by default
The setup honors the house rules. Keys live in a chmod 600 env file, never in the repo. X runs through the xAI API instead of scraping a real account's cookies, so there's no ban risk from a datacenter IP. The skill is zero-dependency Python with no telemetry, vetted before it shipped. It lives in the canonical skills repo as the source of truth and is wired into both the Hetzner and local runtimes.
- Keys in a chmod 600 env file, never committed.
- X via xAI API, not cookie scraping. No ban risk.
- Zero-dep, no telemetry, vetted before shipping.
- Canonical merlino-skills repo = source of truth.