Skip to content

last30days-skill

What it is

last30days-skill is a highly optimized AI agent skill and search engine extension for Claude Code, OpenClaw, and custom command-line workflows. It functions as a specialized research assistant designed to prioritize real-time social signals (including Reddit upvotes, X engagement rates, YouTube transcripts, Polymarket odds, and Hacker News sentiment) over traditional search results, with native support for FastMCP 3.1.

What problem it solves

Standard search engines often surface stale, generic editorial content or SEO-manipulated web results. In the rapidly evolving AI and software ecosystem, critical updates, bug reports, and novel methodologies first appear within developer communities. /last30days bridges these disconnected communication platforms, allowing frontier models like Claude 5.1, GPT-5.5, Gemini 4.0 Pro, and Llama 4 to search, rank, and synthesize authentic community discussions and technical trends from the last 30 days.

Where it fits in the stack

Category: AI Assistants & Knowledge / Claude Code Skills. It functions as a FastMCP 3.1 server or native skill for terminal development environments, integrating with Model Context Protocol (FastMCP 3.1) for dynamic resource retrieval.

Typical use cases

  • Deep Tool Comparison: Querying /last30days OpenClaw vs Hermes to analyze real-world developer experience reports and commit velocities rather than marketing pages.
  • Pre-Meeting Briefings: Instantly compiling a person's or company's technical and social contributions over the previous 30 days.
  • Outage and Bug Identification: Finding immediate workarounds for newly introduced library bugs or service degradation (e.g., /last30days vllm cuda 12.6 out of memory).
  • Git History Synthesis: Summarizing the last 30 days of issues, pull requests, and commits to onboard an AI agent to a codebase.

Strengths

  • Social Signal Integration: Ranks and filters results based on authentic developer engagement and sentiment analysis rather than standard keyword optimization.
  • Parallel Platform Ingestion: Simultaneously queries GitHub, Hacker News, Reddit, and technical blogs using entity-aware subprocesses.
  • Smart Pre-Research: Translates natural language queries into platform-optimized search syntax (hashtags, subreddits, user handles) automatically.
  • Modern Export Formats: Emits interactive, responsive HTML reports, JSON structures, or clean Markdown files suitable for direct consumption by downstream models.

Limitations

  • Token Consuming: Synthesizing raw streams from multiple concurrent sources can quickly consume input tokens if limits are not strictly configured.
  • Rate-Limiting Susceptibility: Strongly dependent on the API limits of external platforms (X, Reddit, GitHub), requiring robust caching mechanisms.
  • Recency Bias: Intentionally overlooks mature documentation or long-standing guides in favor of information from the immediate 30-day window.

When to use it

  • When researching bleeding-edge software updates, frameworks, or newly released open-source models.
  • When you need a "vibe check" on community reception or unexpected performance quirks of a new tool.
  • When tracking live outages, active community-driven workarounds, or breaking API changes.

When not to use it

  • For historical academic research or reviewing stable, long-established APIs and concepts.
  • When authoritative, official documentation is the primary requirement for production deployment.
  • When building safety-critical systems where unverified social-media reports could introduce non-deterministic bugs.

Getting started

Installation (Claude Code Plugin)

# Add the skill via the Claude Code plugin marketplace
/plugin marketplace add mvanhorn/last30days-skill

Installation (OpenClaw)

clawhub install last30days-official

Hello-World

# Research a specific technical topic
/last30days "Claude Code MCP servers"

CLI examples

Deep Tool Comparison with HTML Export

/last30days "OpenRouter vs DeepSeek V4" --emit=html --output=comparison.html

/last30days "vLLM PagedAttention bugs" --sources=github,reddit

Periodic Activity Summary

/last30days "Anthropic API updates" --frequency=weekly --summarize=bulleted

API examples

Integration via OpenClaw Skill API (Python)

Using the modern Python SDK with strict Pydantic v2 schemas for robust input validation.

from pydantic import BaseModel, Field, field_validator
from openclaw import SkillRunner
from typing import List, Optional

class SearchConfig(BaseModel):
    query: str = Field(..., min_length=3)
    platforms: List[str] = Field(default=["github", "reddit", "hacker-news"])
    max_results: Optional[int] = Field(default=15, ge=1)

    @field_validator('query')
    @classmethod
    def validate_query(cls, v: str) -> str:
        if not v.strip():
            raise ValueError("Query string cannot be blank")
        return v

# Initialize the skill
skill = SkillRunner("last30days-skill")

# Validate input schema
config = SearchConfig(
    query="Llama 4 Maverick performance benchmarks",
    platforms=["github", "reddit"]
)

# Execute research query
brief = skill.execute(
    query=config.query,
    depth="detailed",
    platforms=config.platforms
)

print(f"Summary: {brief.summary}")

Programmatic Webhook Trigger (JavaScript)

Triggering a last30days research task programmatically from an external monitoring tool.

// Post search request to local skill server
fetch('http://localhost:3000/skills/last30days/run', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    query: 'GPT-5.5 release dates and features',
    depth: 'comprehensive'
  })
})
.then(response => response.json())
.then(data => console.log('Research brief initialized:', data.task_id));

FastMCP 3.1 Tool Schema (Agentic)

An agent using the Model Context Protocol (FastMCP 3.1) can call the skill using this standard JSON schema:

{
  "tool": "last30days_search",
  "arguments": {
    "query": "Model Context Protocol FastMCP 3.1 updates",
    "sources": ["reddit", "hacker-news"],
    "limit": 10
  }
}

Sources / references

Contribution Metadata

  • Last reviewed: 2027-01-07
  • Confidence: high