Parallel research subagents can exhaust the session's WebSearch budget

After ten parallel research subagents ran, later WebSearch calls in the same session failed with 'this session has used its web search budget (200 of 200 WebSearch calls)'. Raising it requires CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION. Reserve budget for follow-ups when fanning out broad research. WebFetch still worked: direct fetches of a firm's site/team page and aggregator profiles were usable, but html.duckduckgo.com returned a CAPTCHA page, so it isn't a viable search fallback. Calendly event pages are JS-rendered and WebFetch gets no event details, so ask the user for a screenshot.