last30days
An AI agent-led search engine that searches people, not editors—aggregating Reddit, X, YouTube, and more, scored by real engagement.
- Source repo
- mvanhorn/last30days-skill
- Stars
- ★ 65k
- Last updated
- today
- License
- MIT
- Primary language
- Python
- FA score
- 56/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude Code · Claude.aiChatGPT (Partial support)
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A salesperson prepares for a meeting by researching a prospect's recent activity and community sentiment.
- Not a fit if
- Windows users wanting Claude Desktop (support deferred)
- Users unwilling to maintain yt-dlp, Node.js and other external tools
- Source review
- 56/100 · Major gaps
What does this agent do, and when should you use it?
last30days is an AI agent skill that researches any topic across Reddit, X, YouTube, Hacker News, Polymarket, and the web, then synthesizes a grounded summary. It scores results by what real people engage with (upvotes, likes, prediction-market odds) and merges cross-source stories into clusters. The repo provides the skill spec (skills/last30days/SKILL.md), a Python engine (skills/last30days/scripts/last30days.py), install scripts, and configuration docs. Installation is via Claude Code plugin marketplace, the Agent Skills CLI (supporting 50+ hosts), or manual symlink. Reddit, Hacker News, Polymarket, and GitHub work immediately; X, YouTube, TikTok, and others require API keys or browser cookies. MIT licensed, with 2,700+ tests, built on Python 3.12+ with yt-dlp and Node.js.
It takes a topic (person, company, product, or 'X vs Y'), resolves relevant X handles, GitHub repos, subreddits, and hashtags via a pre-research brain. Then it queries sources in parallel: Reddit (via RSS and shreddit scraping), Hacker News, Polymarket, GitHub, X (via browser cookies or API using a vendored Node.js Bird client), YouTube (via yt-dlp for transcripts), TikTok and Instagram (via ScrapeCreators), arXiv, Techmeme, Digg (via CLI), Perplexity (via API), and web search (via Brave API). Results are scored by engagement, relevance, and freshness. Same stories across sources are merged into clusters, and an AI judge synthesizes a brief with citations, top comments, and 'Best Takes' scoring. Direct engine use is via CLI: python3 skills/last30days/scripts/last30days.py "AI coding agents" --emit=json.
- A salesperson prepares for a meeting by researching a prospect's recent activity and community sentiment.
- A job seeker uses /last30days to gauge a company's hiring signals and strategic focus.
- A content creator discovers trending topics before they peak, using discovery mode.
- A traveler checks theme park wait times, ride closures, and community reviews before a trip.
- A developer compares tools (e.g., OpenClaw vs Hermes) with live GitHub star data.
- An analyst tracks geopolitical or market events with Polymarket odds as a signal.
How do you install or deploy this agent?
For Claude Code: /plugin marketplace add mvanhorn/last30days-skill then /plugin install last30days. For Codex, Cursor, etc.: npx skills add mvanhorn/last30days-skill -g. You can also download last30days.skill for claude.ai or use a .mcpb bundle for Claude Desktop. Requires Python 3.12+. API keys or browser cookies are needed for X, YouTube, etc.
How do you use this agent?
Run /last30days <topic> on a supported host, e.g., /last30days Peter Steinberger. First run launches a setup wizard to unlock additional sources. For direct engine use: python3 skills/last30days/scripts/last30days.py "AI coding agents" --emit=json. Use --discover for topic discovery, --store for trend monitoring, --output to specify output file, and --preflight to review planned writes.
What are this agent's strengths and limitations?
- Searches deeply across platforms, including Reddit comments, YouTube transcripts, and TikTok engagement, which major search engines can't.
- Ranks results by real user interactions (upvotes, likes, prediction-market odds) rather than SEO.
- Free and open source, with Reddit, HN, Polymarket, and GitHub working with zero config.
- Multiple installation paths, including Claude Code plugin marketplace, Agent Skills CLI, and manual symlink.
- X, TikTok, Instagram, and YouTube comments require API keys or browser cookies, which involve setup and potential ongoing costs.
- Windows support for Claude Desktop is deferred; requires Python 3.12+ on PATH.
- No deduplication across install methods; having both the marketplace plugin and npx skills copy active shows duplicate entries.
- The engine depends on several external tools (yt-dlp, Node.js, ScrapeCreators), adding maintenance overhead.
How does this agent compare with similar options?
The README references Google and ChatGPT as alternatives, noting they lack access to Reddit comments, X posts, or TikTok content, which last30days provides.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| last30days This agent | 56 · Major gaps | Agent plugin / skillFree + model costs | ★ 65k | today | Python | Codex · Claude Code · Claude.ai |
| Grok CLI | 52 · Major gaps | CLIFree + model costs | ★ 3.5k | 5mo ago | TypeScript | — |
| browser-search: Web Search & Browsing Skill for AI Agents | 55 · Major gaps | Agent plugin / skillFree | ★ 536 | 1mo ago | JavaScript | Claude Code · OpenAI API |
| wigolo | 54 · Major gaps | CLIFree + model costs | ★ 5.5k | 22d ago | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows preflight and doctor commands providing config source and health checks, but least privilege not explicit; installation requires user confirmation (--trust skips), but default behavior unclear; data flow transparency good with config docs and preflight; sensitive data handling involves browser cookies and API keys but storage/encryption not detailed; dependency security has OSV scan and dependency review but no mitigation details; external effects include network requests and file writes but no restrictions; rollback not explicit; source attribution has citations and source list but authenticity not verified.
Self-consistency good, README and SKILL.md consistent, but no full code review; dependency availability not explicit but multiple install methods; failure messages have doctor and error handling but no concrete examples.
Audience and scenarios broad, covering many users and use cases; capability boundaries described but not limited; trigger precision has command examples but not detailed; environment fit supports many hosts and platforms but no detailed config.
Information architecture clear with README, SKILL.md, CONFIGURATION.md; install notes detailed covering many environments; naming stable with version numbers; examples and FAQ have examples but no FAQ; known limitations not explicit; license MIT; versioning changelog has CHANGELOG and towncrier; maintenance responsibility has community contributions and release process.
Output usability has HTML and JSON formats but no examples; marginal value high with multi-source search and synthesis; cost benefit not explicit but free sources and optional APIs.
Claims traceable with sources and citations; cross-source corroboration with multiple sources; fact/inference separation has intent recognition and source labeling but not explicit.
- Least privilege not explicit, may request excessive permissions.
- Sensitive data handling details insufficient, need to ensure secure storage of cookies and API keys.
- Dependency security has scans but no mitigation details.
- External effects not restricted, may cause unexpected network requests or file writes.
- Rollback mechanism not described, may not recover from failed updates.
- Source attribution not verified, may contain false citations.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
What's Changed * docs: add security policy and private vulnerability reporting by @iliaal in https://github.com/mvanhorn/last30days-skill/pull/1257 * docs: align CONFIGURATION.md with the engine, and add Opera to the cookie scan by @iliaal in https://github.com/mvanhorn/last30days-skill/pull/1263 * test: isolate the suite from the developer's last30days config by @iliaal in https://github.com/mvanhorn/last30days-skill/pull/1265 * fix(security): require https for the hosted API and keep local state owner-only by @iliaal in https://github.com/mvanhorn/last30days-skill/pull/1267 …View on GitHub ↗ -
Release notes excerpt
What's Changed * docs: explain Claude Code reattach preview flattening citation links by @mvanhorn in https://github.com/mvanhorn/last30days-skill/pull/1254 * chore(release): bump version to 3.27.3 by @github-actions[bot] in https://github.com/mvanhorn/last30days-skill/pull/1255 Full Changelog: https://github.com/mvanhorn/last30days-skill/compare/v3.27.2...v3.27.3View on GitHub ↗ -
Release notes excerpt
What's Changed * fix(skill): preserve literal topics on macOS Bash 3.2 by @Tivonsico in https://github.com/mvanhorn/last30days-skill/pull/1240 * fix(grounding): keep undated keyed-web results instead of dropping the whole response (#928) by @terry-li-hm in https://github.com/mvanhorn/last30days-skill/pull/1038 * fix(x): sanitize the xurl search query and name invalid-request failures by @clairernovotny in https://github.com/mvanhorn/last30days-skill/pull/1179 …View on GitHub ↗