deep-researchConvene a multi-vendor research council — drive the user's own logged-in ChatGPT and Gemini web UIs in a browser, cross-examine their answers anonymously, and write a chairman's brief with verified sources. Use when the user asks for deep or outside research on a contested or high-stakes question —
fetch-web-sourcesReach a public web source your fetcher cannot see, before reaching for the browser. Use when a web fetch or curl returns an empty body, a 404 or only navigation chrome on a docs site; when you want a docs page as clean markdown (`.md`, `llms.txt`); when you must find a literal string in public repos
link-rot-checkerCheck whether a batch of URLs is still alive and offer replacements for the dead ones. Use when the user wants to verify links before reusing or forwarding them — "check these links", "are these still working", "which of these listings are gone", «перевір посилання», «чи живі» — or when a saved mess
read-twitterRead ONE X/Twitter post through a logged-in browser and return it as Markdown — the body, the author's self-thread, notable replies, a quoted post, attached images read as pixels, native video, and the real content behind every t.co link. Use when an x.com / twitter.com URL appears and its content m
watch-videoExtract content from a video. A LOCAL VIDEO FILE (a disk path, a task attachment, a phone clip — "analyze /path/clip.mov", «що на цьому відео»): read it offline with ffmpeg + whisper.cpp, no upload. A YOUTUBE URL (youtu.be/…, youtube.com/watch, /shorts/) with intent to extract or discuss — "summaris