/last30days: A Cross-Platform Social Media Deep Research Engine
One-Sentence Summary
Use an AI Agent to search 14+ platforms including Reddit, X, YouTube, TikTok, Hacker News, and Polymarket at the same time, ranking results by real user likes, votes, and real money, rather than by SEO editorial ranking.
The Core Problem
Google aggregates editorial content. But real user opinions, community discussions, and immediate reactions are scattered across closed platforms:
- Reddit: the most authentic user opinions; Google cannot search the comments
- X/Twitter: the fastest immediate reaction; ChatGPT has a partnership but cannot search X
- YouTube: in-depth analysis; Gemini has it but Claude does not
- TikTok: cultural trends; no AI can search it natively
- Polymarket: prediction markets with real money at stake, the most reliable "consensus"
Every platform has its own API, authentication, and format. No single AI can access all of them at once.
/last30days solves exactly this problem: you bring your own API keys and browser sessions, and the AI Agent searches across platforms, cross-scores, and synthesizes a summary for you.
Data Sources (14+ Platforms)
| Platform | Signal Type | Free? |
|---|---|---|
| Likes, comment counts, public JSON | ✅ Zero config | |
| Hacker News | Points, comment counts | ✅ Zero config |
| Polymarket | Real-money betting odds | ✅ Zero config |
| GitHub | PR activity, repo stars, issues | ✅ Zero config |
| Digg | The AI 1000 leaderboard (an aggregation of high-signal X accounts) | ✅ Zero config |
| X / Twitter | Likes, reposts, requires a browser session | 🔑 Requires login |
| YouTube | Full subtitle search, yt-dlp | 🔑 Requires installation |
| TikTok | Views, engagement | 🔑 Requires API |
| Instagram Reels | Speech to text | 🔑 Requires API |
| Threads | The post-Twitter text layer | 🔑 Requires API |
| Bluesky | AT Protocol decentralized social | 🔑 Requires API |
| Perplexity | Sonar Pro deep search + citations | 🔑 Requires API |
| Visual discovery | 🔑 Requires API | |
| Truth Social | An alternative social platform | 🔑 Requires API |
On first run, a setup wizard automatically guides you through unlocking each platform.
Real-World Use Cases
1. Pre-meeting background research
/last30days Peter Steinberger
→ Joining the OpenAI Codex team, 23 PRs merged, fighting Anthropic's third-party Agent ban, building LobsterOS. A Reddit discussion with 569 likes asks whether he is a "hero or a nuisance". Real-time updates that Google cannot find.
2. Hiring signal analysis
/last30days Listen Labs --hiring-signals
→ Infer a company's strategic direction in reverse from its careers page: moving into enterprise security, customer success, and infrastructure expansion.
3. Tool comparison
/last30days OpenClaw vs Hermes vs Paperclip
→ Collect real user reviews across platforms, ranked by community engagement.
4. Trending event tracking
/last30days Kanye West
→ UK visa refusal, sponsor withdrawals, Billboard #2, a Fantano return review with 653K views, an 86% Polymarket probability of posting again. 23 Reddit threads, 17 YouTube videos, 86K likes.
v3 Engine Architecture
User query → Query Planner → Parallel search across 14+ platforms
↓
Engagement Scorer (Likes/Votes/Monetary Weighting)
↓
Deduplication Engine (cross-platform deduplication)
↓
AI Judge Synthesized Summary → Output
Core design principles:
- Community engagement > SEO ranking
- Cross-platform signal cross-validation
- Every conclusion comes with a source link
Installation
OpenClaw (the way we use it)
clawhub install last30days-official
Claude Code
/plugin marketplace add mvanhorn/last30days-skill
Other Agent platforms (Codex, Cursor, Gemini CLI, and 50+ more)
npx skills add mvanhorn/last30days-skill -g
Value to Junze Zhiku
| Scenario | Application |
|---|---|
| M&A due diligence | Search a target company's CEO/founder social activity over the past 30 days, surfacing signals the annual report does not show |
| Industry trends | Track Reddit/HN discussions in industries such as chemicals, data centers, and semiconductors |
| Competitors | Monitor a competitor's GitHub activity and hiring moves |
| Policy risk | Polymarket prediction market odds, a consensus backed by real money |
Tech Stack
- Python 3.12+
- yt-dlp (YouTube subtitles)
- Node.js (Bird client for X search)
- ScrapeCreators API (TikTok/Instagram)
License: MIT | Repo: mvanhorn/last30days-skill
More in Evidence
- A Reality Check on Decision Models: Why They Seem Miraculous Online but We Measured Only 54%: A Full Comparison of JEV / LAYA / KEV / CLM-8B and a Deployment Formula
- The "Non-Text-Generating Model": Jev and the New System One Category, and How Agent Architecture Changes When AI Only Answers Multiple Choice
- WeChat Open Source WeMM-Embedding Deep Dive: The Multimodal Embedding Model Topping MMEB-v2, Can It Run on Your Mac?
- A Source-Level Architectural Dissection of DeepSeek Harness: How an Everything-Is-a-Plugin Agent Framework Is Built