← ClaudeAtlas

forgelisted

Learn from a video or podcast and turn it into a Claude Code skill when it deserves one. Use when the user shares a YouTube, Instagram, TikTok, Vimeo, podcast, or local media link and wants to learn from it, extract a procedure, or "make a skill from this". Also use for "what did I learn from", "forge this", or "/forge".
villekaresto/skillforge · ★ 0 · Data & Documents · score 72
Install: claude install-skill villekaresto/skillforge
# /forge — learn from media, forge a skill You study a video or podcast the way a good apprentice would, then decide whether what you learned deserves to become a reusable skill. If it does, you write one. If it doesn't, you file a short learning note. Either way the user ends up smarter. Scripts live next to this file in `scripts/`. Run them with the Python 3.10+ interpreter available in the session. ## Workflow ### 0. Read the user's profile ``` python3 scripts/forge.py --profile ``` This prints `~/.skillforge/profile.md`: the user's role, tools with versions, core work, context, and topics that are not relevant. Read it before judging and distilling. If there is none, carry on without one; you can mention once that copying `references/profile.example.md` there makes the judge and `apply_to_me` sharper. Never write profile contents anywhere except `apply_to_me`, and never into a repository. ### 1. Ingest ``` python3 scripts/ingest.py "<url or file>" ``` This fetches metadata and native captions (no download needed for most YouTube videos). It prints a JSON summary with `work_dir`, `transcript_md` and an `errors` list. Read `transcript.md`. Check that the transcript is in the language actually spoken: YouTube's metadata language is often wrong. Ingest prefers the original speech-recognised track over a machine translation and lets Whisper detect the language unless you pass `--lang`. Long transcripts (podcasts, talks over an hour) can run to hundreds of KB. Never