last30days-glim

A 30-day social + web research brief for any topic. The skill fans out across Reddit, X, YouTube, GitHub, Hacker News, Polymarket, Bluesky, TikTok, Instagram, Threads, and the open web in parallel, ranks results by real engagement (upvotes / likes / dollar-backed odds), deduplicates across platforms, and synthesizes a brief grounded in primary sources.

This skill is a port of mvanhorn/last30days-skill (MIT) at SHA 5b87cca. All real-world data retrieval and LLM-judge calls are routed through the glim MCP / glim v2 HTTP API. Credit to @mvanhorn and @j-sperling for the v3 engine architecture, planner, judge prompts, and synthesis voice contract.

Powered by glim

One API key. One balance. Reddit, HN, Polymarket, GitHub work without it (free baseline via direct HTTP); X, YouTube, Bluesky, TikTok, Instagram, Threads, Pinterest, web search, and LLM judges all route through glim when GLIM_API_KEY is set. Glim takes the role of upstream's seven separate keys (xAI / ScrapeCreators / Brave / OpenRouter / Apify / X browser cookies / yt-dlp install). When direct HTTP fails for the free baseline (rate-limit, anti-bot), glim is also the resilience fallback.

Setup

Get a glim API key at https://surf.cascade.fyi/app.
Top up the embedded balance: load USDC into a Tempo wallet at https://wallet.tempo.xyz, then transfer that USDC to your glim wallet (address visible in the glim dashboard).
Set GLIM_API_KEY in your env (or in ~/.config/last30days-glim/.env / .claude/last30days-glim.env).
Run /last30days-glim <topic> (or invoke the skill by name).

If you don't set GLIM_API_KEY, the skill still runs but only against the free-baseline sources (Reddit, HN, Polymarket, GitHub) and without LLM-judge reranking. Brief quality drops materially. The agent should surface this state to the user.