Web fetch — JS-rendered CFP pages

Applies only to interactive CFP-page research (Steps 4, 6, 7). Scheduled mode must not use this fallback chain; follow the scheduled branches in SKILL.md. Many sites return empty SPA shells to plain WebFetch.

• Switch to mcp__nanoclaw__fetch_markdown(url: "<cfp-url>", wait_until: "networkidle") — the NanoClaw host's server-side renderer returns the fully-rendered page as clean markdown ready for the model. A built-in disk cache makes repeat lookups against the same CFP page during a single sweep free.
• Fall back to Cloudflare Browser Rendering via Composio (CLOUDFLARE_BROWSER_RENDERING_TAKE_WEBPAGE_SNAPSHOT for full DOM, ..._SCRAPE_HTML_ELEMENTS for CSS-selector queries) when the fetch_markdown response is empty or indicates a failure (loading shell or renderer error).
• The Composio path needs an account ID and has parameter-name and MULTI_EXECUTE batching caveats — see your Composio Cloudflare Browser Rendering tool configuration.