belschak
UserTwo Claude Code / agent skills that watch videos for you: Whisper transcript plus one page of findings with timestamps, for YouTube, podcasts and local recordings. An optional server pipeline turns a YouTube playlist into transcripts in your inbox.
Categories
Indexed Skills (6)
video-watch
Transcribe YouTube videos and podcast episodes in high quality (Whisper via a configurable API endpoint) and distill compact findings with timestamps. Use whenever the user wants a video "watched", summarized, transcribed or analyzed, drops a YouTube or podcast URL together with a question, asks "what does he say about X in that video", or follows up on an earlier video folder, even when the word "transcribe" never appears. For visual questions, frames can be extracted on top. NOT for local audio files (voice memos, recordings); that is the audio-transcribe skill.
publish-gate
Mandatory security, legal, and quality gate BEFORE anything goes public. Always run when a GitHub repository is about to be created, made public, or pushed; when something is submitted to a package registry or marketplace (npm, PyPI, a skills directory); when a blog post, gist, or artifact containing code or config is published; or when the user says "make it public", "publish", "push the repo", "launch", "release", "go live". Also trigger when publishing is only a sub-step of a larger task, when it seems urgent, or when someone claims "it was already checked". Nothing is published without a complete gate run and an explicit human go.
audio-transcribe
Transcribe local audio files (voice memos, meeting recordings, dictations, interviews; m4a/mp3/wav/opus/ogg/flac, also the audio track of local video files) in high quality via Whisper (configurable API endpoint) and place the transcript as a dated .md next to the source file. Use whenever the user drops an audio file, mentions a path to a recording, or wants its content as text: "transcribe this", "what do I say in the recording", "turn this into text", "summarize the memo", even without the word "transcribe". NOT for YouTube or podcast URLs; that is the video-watch skill.
parallel-streams
Split a larger project into independent work streams and build them in several parallel headless claude -p terminals at once. Limit-tolerant: when the subscription window is exhausted, every stream pauses automatically and resumes until its done marker exists. Use when the user says "in multiple terminals", "build in parallel", "streams", "build through the night", "split it up and run simultaneously", "overnight run", "keep going even when the limit hits", or when a build job is clearly too big for one session (3+ independent work packages). Also use when the user wants to watch running streams live, steer one ("tell stream B that..."), pause one, or take over interactively (steer file / takeover). Not for small tasks (<3 packages: subagents or sequential work).
web-research-cascade
A robust web-fetch cascade so that blocked primary sources still get read instead of being silently skipped. Use this skill whenever a URL is fetched, a page needs to be researched, or WebFetch/WebSearch returns a 403/401/429/"unable to fetch"/empty body/CAPTCHA. Also use it for any analysis or comparison research where primary sources matter (not just secondary blogs), and on "read this page", "what does X say", "research Y", "look at this". Cascade: WebFetch, then Jina Reader, then a Firecrawl script (sparingly), then Chrome (browser MCP). Also covers Reddit (threads, subreddits, search) and single X/Twitter tweets, which need their own route instead of the generic cascade. Core rule: a good but blocked source is opened via the cascade, NEVER swapped for a weaker error-free one.
repo-audit
Security audit of any third-party repo, package, skill, plugin, or MCP server BEFORE it is installed or executed. Use this skill whenever the user asks "is this safe?", "audit/vet/check this repo", "should I install X", pastes a GitHub/npm/PyPI/marketplace link with intent to install, or whenever you yourself are about to recommend or install third-party code, even if nobody says the word "audit". Covers maintainer reputation, typosquatting, install-time hooks, obfuscation, exfiltration and secrets-harvesting patterns, dependency and CI risks, and prompt injection in agent-facing files (skills, agent configs, MCP tool descriptions). Produces a SAFE / SAFE-WITH-CONDITIONS / UNSAFE verdict with evidence, transparent red flags, and conditions such as version pinning.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.