web-researchlisted

# Web Research Use this skill for web questions and URL verification. ## Tool sequence 1. Discover - Start with provider-native web search for candidate sources. 2. Read - For **GitHub URLs** (`github.com/...`, `gist.github.com/...`), use `gh` CLI instead of defuddle. It hits the API directly: cleaner output, fewer tokens, and it works on private repos. See the GitHub URL mapping below. - For **YouTube URLs** (`youtube.com/watch`, `youtu.be/...`, `youtube.com/shorts/...`), use `defuddle parse '<url>' --md`. Defuddle returns the video description plus a markdown transcript with timestamps and chapters when captions exist. Strip `&t=` / `#t=` from the URL if parsing fails. - For all other URLs, use `defuddle parse <url> --md`. This is the default for the rest of the web; it strips clutter and reduces token usage. - If defuddle returns empty or stub content (JS-rendered pages, SPAs, login-gated content, dashboards, YouTube videos without captions), fall back to Scrapling when it is installed (`command -v scrapling`): 1. `scrapling extract get '<url>' /tmp/page.md --impersonate chrome` — fast HTTP fetch with browser impersonation, then read the output file. 2. If still empty or stub, `scrapling extract fetch '<url>' /tmp/page.md` — renders the page in a headless browser, so it handles JS-only content. Slower; use it only after step 1 fails. - If Scrapling is not installed or both attempts fail, say so clearly and ask the user to paste the content or try from a machine whe