This is an early release preview. You may encounter bugs.
Last30days Skill logo
Unclaimed

Tool research

Last30days Skill

AI-agent skill that researches a topic across social platforms and the web and synthesizes an engagement-ranked brief

A+ 94/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
Discussion

01 / Summary

About Last30days Skill.

last30days is an agent skill that researches a topic as discussed over the past 30 days across social platforms, prediction markets, and the web, then has the agent synthesize the results into one cited brief. A topic can be a person, company, product, technology, or an "X vs Y" comparison, and results are ranked by engagement signals such as upvotes, likes, views, and market odds rather than by search-engine relevance.

A run resolves the relevant handles, repositories, subreddits, hashtags, and channels first, then searches every source in parallel with multi-query expansion and scores results by engagement, relevance, and freshness. It pulls full YouTube transcripts, top Reddit comments with scores, TikTok captions, and Polymarket odds, merges coverage of the same story across sources into one cluster, and writes a brief that the agent session can then answer follow-up questions from.

Reddit, Hacker News, Polymarket, and GitHub work without keys. Other sources, including X, YouTube, TikTok, Instagram, LinkedIn, Perplexity, and Bluesky, use your own API keys or browser sessions, and a doctor check reports which credentials or helper tools are missing. The skill runs in Claude Code, as a native Codex plugin, in Grok Build, Claude Desktop (as an MCP bundle), claude.ai, OpenClaw, and 50+ other Agent Skills hosts such as Cursor, Copilot, and Gemini CLI.

  • Engagement-ranked search: results scored by upvotes, likes, views, comments, and prediction-market odds across sources
  • Sources: Reddit, X, YouTube, TikTok, Instagram Reels, Hacker News, Polymarket, GitHub, arXiv, Techmeme, Digg, LinkedIn, Bluesky, Threads, Pinterest, StockTwits, Perplexity, and the web
  • Pre-research resolution: the agent identifies the right handles, repos, subreddits, and channels before any search runs
  • Story clustering: coverage of the same event across platforms merges into one cluster
  • Discovery mode: a topic-less query returns 5 to 10 velocity-ranked trending topics with cross-source numbers
  • Comparisons: single-pass "X vs Y" research with side-by-side tables and auto-discovered competitors
  • Person and company modes: GitHub activity for people and hiring signals read from a company's job pages
  • Take scoring: comments scored for humor and virality alongside relevance
  • HTML briefs: shareable self-contained HTML output
  • Doctor check: runs every source and reports missing keys, tools, or expired cookies
  • Multi-host: Claude Code plugin, Codex plugin, Grok Build plugin, Claude Desktop MCP bundle, OpenClaw, and Agent Skills hosts

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

04 / Build

Build with Last30days Skill.

Browse the catalog for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Last30days Skill →

Browse the catalog