last30days v3.11.1: Agent search skill update across Reddit, X, YouTube and more
Decision Brief
From v3.3 to v3.11.1 there were 15 releases and 175 merged PRs, 122 of them from 52 community contributors. Codex becomes a first-class plugin, and a renderer that understands citations makes output read like a briefing instead of a pile of URLs; arXiv, Techmeme, and Digg are free with no API key, auto-installing and activating the CLI on first run. After Reddit's public .json API broke, it switched to RSS plus shreddit scraping, preserving real upvote counts and top comments via arctic-shift, and added a relevance floor to keep off-topic viral posts out of briefings. X search was rebuilt from scratch: FROM and ABOUT channels split a person's own posts from discussion about them, with person-aware subquery disambiguation, engagement-signal ranking, and automatic backend failover. For developers doing research daily in Claude Code, Cursor, Copilot, or Gemini CLI, they can now have the Agent search a dozen-plus platforms in parallel and score by real engagement; LinkedIn is wired in via ScrapeCreators, and StockTwits auto-activates when the topic is a ticker or crypto. A new doctor command health-checks each source and gives concrete fixes, like which key is missing, which CLI isn't on PATH, or which cookie expired. Teams using X and Reddit signals for meeting prep or competitor research get clearly better coverage and traceability than relying on a single platform.
Sources
- Skill Radar(GitHub 趋势)
Trending hands-on MCP servers, agent skills, and AI-coding tools discovered from GitHub search momentum.
- GitHub:mvanhorn/last30days-skill
留言
登入后即可留言,和其他 builder 交换实测心得。
还没有留言,抢头香。