DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to spot what is exploding in a topic before it peaks with Last30days Skill

Ask an open question and the skill switches to discovery mode, sweeping fast-moving sources for rising topics.

Short answer: Install the skill as above; then ask an open question such as /last30days what's exploding in AI agents?; let the agent judge the nominations it collects from Reddit, Hacker News, Digg and X when authenticated. All steps.

  • Research
  • Beginner
  • 15 minutes
  • From the README
  • Claude Code
  • Codex
  • Cursor
  • Gemini CLI

Who it is forA creator, founder or investor who wants the next big topic in a field before everyone covers it.

The problemBy the time a trend reaches the news, the early window is gone.

The outcomeA short list of fast-rising topics with the posts behind them and suggested angles.

What you need first

  • Claude Code, Codex, Cursor, Copilot, Gemini CLI or another Agent Skills host
  • Optional keys or sessions for extra sources such as X, YouTube and TikTok

Steps

  1. Install the skill as above
  2. Ask an open question such as /last30days what's exploding in AI agents?
  3. Let the agent judge the nominations it collects from Reddit, Hacker News, Digg and X when authenticated
  4. Pick a topic and run a normal /last30days query on it for the full brief

Prompt for your agent

Use mvanhorn/last30days-skill in discovery mode for <field>. Return the 5 fastest rising topics with one supporting post count each, then run a full 30 day brief on the top one.

Caveat. X results need your own authenticated session, so discovery is narrower without it.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for mvanhorn/last30days-skill and "Spot what is exploding in a topic before it peaks"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer