intel

Turn a scraped App Store reviews JSON into a concise, data-driven Chinese analysis doc plus charts — most-liked and most-disliked patterns ranked by frequency, and the top feature requests. Use when asked to analyze/分析 an app's App Store reviews, summarize what users love and hate, or extract feature requests from a reviews dataset. NOT for scraping reviews (the JSON must already exist) or for non-review market research.

Turn a long-form source — an article, a podcast transcript, a YouTube video, a PDF (whitepaper, filing, deck), a page that only offers audio — into durable, searchable highlights. Use for "/digest <url-or-file>", "highlights of <url>", "what did <source> say about X", "summarize this article/episode/paper", "search my notes". Handles ordinary article and transcript pages, YouTube (via subtitles), PDFs (URL or local file), and audio pages (via the transcribe skill). NOT for: general web research across many pages (use agent-reach), or evaluating a tool or vendor (use evaluate).

Search Anna's Archive for EPUB books, compare result download counts, extract fast and slow download links, download the selected book to ~/Downloads, and convert it into a dark-themed English PDF. Use when asked to find a book's download options, inspect this site's EPUB search results, download a book, or turn a book into a PDF.

Brainstorm brand names for a product and return the ones whose domain is actually registrable — short coined words (and metaphor words) distilled from a theme you give, checked live on .xyz/.ai/.fun via Namecheap's official API, with same-name collisions against anything popular filtered out. Use for "find me a domain", "name this product", "brainstorm a brand name and check the domain", "什么域名还能注册". NOT for checking one specific domain you already have (just run scripts/check.mjs), and NOT for logo/visual identity.

Turn audio into plain-text words, transcribed locally with whisper — a finite file or URL, or a live stream captured as it plays. Use when you have recorded or ongoing audio (a podcast episode, an interview, a Twitch or YouTube live, an X Space, a radio stream) and need its transcript, and the source offers no text of its own. It also transcribes the audio track of a **video** file or stream directly — ffmpeg extracts the audio, so no manual step is needed. Other skills call this as their audio-to-text step. NOT for: reading a text transcript that already exists (fetch it directly), or anything in a video's picture (it transcribes the audio only — no slide OCR or scene description).

Translate a whole English book PDF into a Chinese PDF that keeps the original's format: the same cover page, page size, chapter structure, running heads, folios, a clickable 目录 page and PDF bookmarks (Cover, 目录, one per chapter). Works from the PDF outline, or from the printed contents page when there is none, and on two-up scans. Chapters are translated in parallel by gpt-6-luna through `codex exec`, one call per chapter, and typeset with headless Chrome in Baskerville + Songti SC. Use for "/translate <book.pdf>", "把这本书翻译成中文", "translate this book", or to re-render an already translated book after editing its Markdown. NOT for a single page or article (just translate it inline), and not for digesting or summarizing a source (use digest).