last30days Skill for Claude: Research What People Say Right Now

• last30days <topic>

  Resolves the topic, writes a query plan, runs the engine across sources, adds web search results and writes a cited brief that starts with 'What I learned:'.

  Runs code · Topic research

• last30days A vs B, or --competitors

  Researches each side and writes a Quick Verdict, a section per entity, a head-to-head and a bottom line.

  Runs code · Comparison

• last30days what's exploding in <area>

  Nominates topics from Reddit, Hacker News, Digg and X, has Claude judge them, then finalizes a ranked brief.

  Runs code · Discovery / trending

• --hiring-signals

  Reads current jobs and careers pages as cited evidence of a company's focus.

  Network · Hiring signals

• Ranked list of things to try or buy

  Ranks and renders recommended items from the research evidence.

  Recommendations

• --agent, or --emit=json

  Structured output for other tools, without the conversational footer or invitation.

  Agent mode

• Offline commands on saved briefs

  Searches saved research, builds a local index.html and Atom feed.xml, and works through a queue without new research.

  Write · Library: search, feed, queue

• Questions about research already done

  Answers from existing findings, or re-renders them, without re-running the search unless you ask.

  Follow-ups

• --emit=html, or ask for a shareable brief

  Saves a local HTML file first. It uploads only if you ask for a hosted link, and the SKILL.md says that link may be public.

  Write · HTML export

• Diagnose broken or missing sources

  Reports which sources are configured and healthy, and points to setup repair.

  Runs code · Doctor (source health check)

• Unlock more sources in about 30 seconds

  Asks before reading browser cookies, offers the ScrapeCreators signup and picks sources. Can save keys to a .env file or the macOS Keychain.

  Write · First-run setup wizard

• See what a run would read and write

  An opt-in inspector for config source, planned writes and available sources. It does not read cookie values or run research.

  Read · --preflight

Watch out for

• Its code can decrypt your browser's X login cookies

  If you consent in setup (or set FROM_BROWSER), scripts/lib/chrome_cookies.py copies the browser cookie database and asks the macOS Keychain for the browser's Safe Storage key to read the X cookies auth_token and ct0. scripts/lib/env.py also lists Truth Social's _session_id. A stolen X login cookie acts as your account, so decline unless you want that. github.com

• It sends your search topics to other services

  Queries go to ScrapeCreators, X, Hacker News, Polymarket and, if you set keys, OpenAI, xAI, Perplexity or OpenRouter. Do not research topics you cannot share with those services.

• It can spend credits on paid keys

  With a ScrapeCreators key, TikTok, Instagram and some Reddit and YouTube backfills use credits (10,000 free, then pay as you go). The SKILL.md names settings that limit Reddit backfill.

• First-run setup installs software from outside the skill

  scripts/lib/setup_wizard.py runs `brew install yt-dlp` when yt-dlp is missing and `npx -y @mvanhorn/printing-press-library@0.1.16 install <cli> --cli-only` to fetch the arXiv, Techmeme and Digg CLIs from npm. The setup wizard can also save keys to a .env file or the macOS Keychain. Check what it is about to run before you approve it. github.com

• Hosted sharing is public by default

  Publishing the library to ht-ml.app happens only after you opt in, and pages are public unless you choose password protection.

• 136 bundled scripts, only some read for this page

  The skill bundles 136 files under scripts/, including a vendored X client in JavaScript. We read the SKILL.md, the README and the setup, cookie and subprocess code, not every script. The SKILL.md itself says to review the scripts before first use.