web-readlisted

# Web Read ## Core Rule When handed a content URL to read, extract clean markdown with the Defuddle CLI instead of pulling raw HTML through WebFetch — it strips nav, ads, cookie banners, and boilerplate, so the same content costs far fewer tokens. Never silently install a global tool: if Defuddle is missing, surface the one-line install and fall back to WebFetch so the task isn't blocked. ## When to Use When the user gives a URL to read, summarize, or analyze: - Documentation pages, API guides, RFCs, specs - Articles, blog posts, release notes, changelogs - Any standard, content-heavy web page Do NOT use for: - URLs ending in `.md` / `.txt` — already clean; use WebFetch directly - JSON / API endpoints — use WebFetch or `curl` - Pages behind auth — Defuddle fetches anonymously; use an authenticated path instead - Vendoring library docs into `docs/references/` — that's `/references-sync`'s job (it may call this skill to do the extraction step) ## Prerequisites Check once per session: ```bash command -v defuddle >/dev/null && echo ok || echo missing ``` If missing, tell the user the install and ask before running it — a global install is a tool change worth surfacing: ```bash npm install -g defuddle ``` While it's unavailable, fall back to `WebFetch` rather than blocking. ## Process 1. Extract clean markdown: ```bash defuddle parse <url> --md ``` 2. For long pages, save to a file and read only the slice you need instead of inlining the whole thing: