clipsmith-weblisted

# Clipsmith Web Capture Use this skill for generic web pages when no platform-specific Clipsmith skill matches. Run the bundled browser extractor first: ```bash cd <clipsmith-repo>/skills/clipsmith-web npx tsx scripts/run.ts \ --url "<url>" \ --output_dir "$HOME/Downloads/clipsmith-web" ``` Use `--preserve_full_raw` only when debugging a difficult capture or when the user explicitly asks for a fuller raw archive. The default output intentionally keeps raw evidence small. The script creates a draft bundle directory containing: - `capture.json` - `post.md` - `summary.md` - `raw/source.html` - `raw/rendered.txt` - `raw/metadata.json` ## AI Normalization Step After the script finishes, read `prompts/extract-article.md`, then inspect the bundle's `raw/source.html`, `raw/rendered.txt`, and `raw/metadata.json`. Use the current AI session to rewrite `post.md` into clean, complete article Markdown: - remove site chrome, navigation, cookie banners, newsletter prompts, related links, social sharing labels, and footer text - preserve the full article body in order - preserve article headings, lists, quotes, code blocks, tables, and meaningful links - improve `title`, `canonical_url`, `author`, and `published_at` in `capture.json` only when the raw evidence supports the change - keep raw audit files declared in `capture.json.assets` Do not invent article text. If the raw files contain only a bot challenge, login wall, error page, or a short/ambiguous fragment, keep th