mvanhorn/last30days-skill

⭐ 25,343  ·  Python  ·  GitHub Repo

AI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary

ai-prompts ai-skill bluesky claude claude-code clawhub deep-research hackernews

1-Sentence Summary

Parallel social search engine for AI agents that surfaces recency-ranked, engagement-scored content from walled gardens.

🔥 Key Capabilities & USP

• Multi-Platform Parallel Search: Simultaneously queries 15+ sources (Reddit, X, YouTube, TikTok, Hacker News, Polymarket, GitHub, Instagram, Bluesky, Threads, Pinterest, Digg, Perplexity, open web) in a single /last30days command. Pain point solved: Eliminates manual tab-switching and fragmented search across platforms that block traditional search engines.
• Engagement-Based Scoring: Ranks results by real human signals—upvotes, likes, views, and Polymarket money odds—not SEO manipulation or editorial curation. Pain point solved: Buried signal from genuine human interest surfaces above keyword-stuffed content.
• Intelligent Pre-Research Brain (v3): A Python-based entity resolver identifies specific people, subreddits, channels, and hashtags before searching, enabling discovery of content a naive keyword search would miss entirely. Pain point solved: Keyword searches fail on platforms where context matters (e.g., finding a CEO's mentions on Reddit without knowing the right subreddit).
• Shareable HTML Briefs: Generates self-contained, dark-mode, offline-friendly HTML files with inline CSS and zero JavaScript—ready for Slack, email, or Notion embedding. Pain point solved: Raw AI chat output is unshareable; these briefs are polished artifacts for decision-makers.
• Zero-Config Core + Optional API Wizard: Reddit, HN, Polymarket, and GitHub work immediately. A ~30-second setup wizard unlocks X, YouTube, TikTok, and others with user-provided keys. Pain point solved: Low friction for immediate value, with optional depth for power users.

Technical Architecture

Component	Implementation
Runtime	AI agent skill deployable via Claude Code (/plugin marketplace add), OpenClaw (clawhub install), or Hermes (sync.sh)
Search Engine	v3 pipeline: Python pre-research brain (by @j-sperling) → entity resolution → parallel multi-platform queries
Scoring Layer	Engagement-based ranking (upvotes, likes, views, Polymarket odds) with recency weighting
Output Format	Markdown synthesis in chat + optional self-contained HTML file (dark mode, inline CSS, no JS)
Storage	$LAST30DAYS_MEMORY_DIR/{topic}-brief.html (defaults to ~/Documents/Last30Days/)
Data Integrity	Degraded-run warnings kept in engine logs; never leaked into shareable artifacts

Quick Start Guide

# Option 1: Claude Code (recommended)
/plugin marketplace add mvanhorn/last30days-skill

# Option 2: OpenClaw
clawhub install last30days-official

# Option 3: Hermes (auto-deploy via sync.sh)
# Or manually copy to ~/.hermes/skills/research/last30days/

# Basic usage
/last30days Peter Steinberger

# With HTML brief export
/last30days OpenClaw --emit=html

# Natural language variants
/last30days Cursor IDE for slack
/last30days Anthropic earnings export as html

Pros, Cons & Use Cases

Pros

• Unlocks walled-garden content that no single AI or search engine can access (Reddit comments + X posts + YouTube transcripts simultaneously)
• Real-world signal via engagement scoring—prioritizes what people actually care about, not what SEO optimizers want you to see
• Polished, portable artifacts—HTML briefs are shareable, offline-friendly, and require zero infrastructure
• Community-extensible—new platform support via contributor PRs means the surface area grows over time
• Recency-first design—solves the "stale results" problem endemic to traditional search

Cons

• API key friction—X, YouTube, TikTok require user-provided keys or browser sessions; not fully zero-config
• Synthesis quality is AI-dependent—the final summary quality relies on the AI agent judge's capabilities (Claude, Hermes, etc.)
• Platform fragility—community-maintained platform connectors may break when APIs change; no SLA guarantees
• No offline mode—requires network access to all platforms; can't pre-cache results

Who should NOT use this?

• Users who need deep, historical analysis (months/years)—this is explicitly a "last 30 days" tool, not a research archive
• Teams requiring guaranteed uptime/SLAs—community project with no commercial support
• Users unwilling to provide API keys—core platforms work, but X/YouTube/TikTok require setup
• Enterprise compliance teams—no audit trail, no data retention controls, no SSO

Ideal Use Cases

• Pre-meeting intelligence—CEO background checks, competitor moves, partner reputation scans
• Sales intelligence—rapidly understand a prospect's recent activity, public sentiment, and market position
• Product decisions—gauge real user sentiment about features, tools, or competitors across all social platforms
• Journalism & analysis—track breaking stories, public reaction, and prediction market odds simultaneously
• Travel & event planning—get the "ground truth" about destinations, conferences, or local trends from real people

Community & Activity

With 25,343 stars and an active update as recent as May 2026, this project has clearly struck a nerve. The v3 pre-research brain (community contribution by @j-sperling) signals healthy contributor momentum. The badge ecosystem (GitHub Trending #1, Trendshift.io) confirms this isn't just passive interest—developers are actively installing and using this skill. The broad topic coverage (Bluesky, Threads, Polymarket, TikTok) shows the maintainer is keeping pace with platform shifts, not resting on a stale feature set.