source-to-skilllisted

# source-to-skill Turn a knowledge source into a structured agent skill the user can load on demand. Supported sources: YouTube URLs, arXiv URLs, local PDF papers, and books — EPUB files, or PDF books via `--type book`. Usage: `/source-to-skill <url-or-file> [skill-slug]` — for a PDF that is a book rather than a paper, extract with `--type book`. ## Step 1 — Extract Run from the source-to-skill repo root: python3 scripts/extract.py "<source>" - `.epub` files are detected as books automatically. A `.pdf` defaults to the paper parser — when the PDF is a book, add the override: `python3 scripts/extract.py "<source>" --type book`. If it is unclear whether a PDF is a paper or a book, ask the user before extracting. - On dependency errors, run `python3 scripts/extract.py --check` and show the user the install hints. Do not install anything without asking. - On any other ERROR, report it verbatim and stop. - On success the command prints JSON with `work_dir`, `title`, `est_tokens`, and `segments`. All extracted content is in `<work_dir>/full_text.txt` and `<work_dir>/metadata.json`. ## Step 2 — Confirm cost Tell the user the title, source type, and `est_tokens`, and ask to proceed. Generation reads the full text once; skills built from it cost a fraction of that per future question. ## Step 3 — Analyze structure Read `metadata.json`. For large sources (over ~40k est_tokens) do NOT read `full_text.txt` in one go — read it segment by segment using the `offset