The AI world reinvents itself every month. This skill keeps you current.
/last30days researches your topic across Reddit, X, YouTube, and the web from the last 30 days, finds what the community is actually upvoting, sharing, and saying on camera, and writes you a prompt that works today, not six months ago.
Three Headline Features
1. Open-class skill with watchlists. Add any topic to a watchlist -- your competitors, specific people, emerging technologies -- and /last30days re-researches it on demand or via cron. Designed for always-on environments like Open Claw. SQLite-backed with FTS5 full-text search.
2. YouTube transcripts as a 4th source. When yt-dlp is installed, /last30days automatically searches YouTube, grabs view counts, and extracts auto-generated transcripts from the top videos. A 20-minute review contains 10x the signal of a single post -- now the skill reads it. Inspired by @steipete's yt-dlp + summarize toolchain.
3. Works in OpenAI Codex CLI. Same skill, same engine, same four sources. Install to ~/.agents/skills/last30days and invoke with $last30days.
Plus: Bundled X search -- vendored Bird GraphQL client (MIT). No external CLI, no npm install, no API keys needed. Just Node.js 22+ and your browser cookies.
Real Results (verified Feb 15)
| Topic | X | YouTube | Web | |
|---|---|---|---|---|
| Nano Banana Pro | -- | 32 posts, 164 likes | 5 videos, 98K views, 5 transcripts | 10 pages |
| Seedance 2.0 access | 3 threads, 114 upvotes | 31 posts, 191 likes | 20 videos, 685K views, 4 transcripts | 10 pages |
| OpenClaw use cases | 35 threads, 1,130 upvotes | 23 posts | 20 videos, 1.57M views, 5 transcripts | 10 pages |
| YouTube thumbnails | 7 threads, 654 upvotes | 32 posts, 110 likes | 18 videos, 6.15M views, 5 transcripts | 30 pages |
| AI generated ads | 12 threads | 29 posts, 101 likes | 3 videos, 83K views, 3 transcripts | 30 pages |
What's New
Added
- Open-class skill with watchlist, briefing, and history modes
- YouTube search + transcript extraction via yt-dlp
- OpenAI Codex CLI compatibility
- Bundled Twitter/X search (vendored Bird GraphQL, MIT)
- Native web search backends (Parallel AI, Brave, OpenRouter/Perplexity Sonar Pro)
--diagnoseflag for source status checking--storeflag for SQLite accumulation- Conversational first-run experience (NUX)
Changed
- Two-phase search architecture (entity-aware drill-down)
- Reddit JSON enrichment for real engagement metrics
- Smarter query construction with auto-retry on 0 results
- Engagement-weighted scoring (relevance 45%, recency 25%, engagement 30%)
--days=Nconfigurable lookback (thanks @jonthebeef)
Fixed
- YouTube/Reddit timeout resilience
- Reddit 429 rate limit fail-fast
- Eager import crash in Codex environments
- X search returning 0 results on popular topics
- Windows Unicode crash (thanks @JosephOIbrahim)
New Contributors
- @JosephOIbrahim -- Windows Unicode fix (#17)
- @levineam -- Model fallback for unverified orgs (#16)
- @jonthebeef --
--days=Nconfigurable lookback (#18)
Credits
- @steipete -- Bird CLI (vendored X search) and yt-dlp/summarize inspiration for YouTube transcripts
- @galligan -- Marketplace plugin inspiration
- @hutchins -- Pushed for YouTube feature
Install
# Claude Code
git clone https://github.com/mvanhorn/last30days-skill.git ~/.claude/skills/last30days
# Codex CLI
git clone https://github.com/mvanhorn/last30days-skill.git ~/.agents/skills/last30days30 days of research. 30 seconds of work. Four sources. Zero stale prompts.