Tried mvanhorn/last30days-skill: A Practical Research Upgrade for AI Agents

mvanhorn/last30days-skill is an AI agent skill for researching what is happening around a topic right now. Instead of relying on a single search engine, it checks Reddit, X, YouTube, Hacker News, Polymarket, and the broader web, then synthesizes the findings into a grounded summary.

That explains why the repository is gaining traction: it currently has +272 GitHub stars today. Developers are increasingly asking agents questions like:

• What changed in a framework during the last month?
• Which AI tools are people actually using?
• What are developers saying beyond polished launch posts?
• Is a trend real, or just social media noise?

The useful part is the multi-source workflow. An agent can collect recent signals, compare conflicting opinions, and provide links or evidence instead of producing a generic model-generated answer.

Quick gateway setup

For a custom OpenAI-compatible relay, I would configure the skill through the gateway’s standard API variables:

export OPENAI_BASE_URL="https://b-lost.com/v1"
export OPENAI_API_KEY="$B_LOST_API_KEY"
export OPENAI_MODEL="claude-fable-5"

A JSON-style IDE configuration could look like this:

{
  "provider": "openai-compatible",
  "baseUrl": "https://b-lost.com/v1",
  "apiKey": "${B_LOST_API_KEY}",
  "model": "claude-fable-5",
  "temperature": 0.2
}

This setup works well for agent workflows in tools such as Cursor, Cline, Roo Code, Windsurf, Aider, NextChat, and LibreChat, provided the client supports a custom OpenAI or Anthropic-compatible base URL.

For long research prompts, prompt caching can make a noticeable difference. B-Lost supports native Anthropic /v1/messages caching, with 90% discounts on cache hits, which is useful when the same instructions, source filters, or research framework are reused across many queries. Its relay also advertises 20% off official list pricing, though actual cost still depends on model usage and source volume.

My quick take: this skill is compelling because it turns an AI agent into a repeatable “what changed recently?” research assistant rather than another chat window.