The agent assembled a complete explanation of book.vue.tw by combining direct site fetches (which only returned metadata) with web search results that surfaced key facts: Kuro Hsu's authorship, the CC BY-NC-SA license model, the existence of a paid physical book, and comparisons to alternatives. The site itself was poorly navigable for programmatic discovery—it rendered as a client-side Vue app with inaccessible content—so the agent worked around this using prior knowledge and search-sourced artifacts to fulfill the task.

• ›Steps [1] and [4] (homepage fetches) returned only HTML skeleton and metadata (title: '重新認識 Vue.js | Kuro Hsu', meta description: '008 天絕對看不完的 Vue.js 3.0 指南'), but no actual content or navigation structure readable by the agent. The site is a client-side rendered Vue app, making initial discovery difficult.
• ›Steps [9] and [11] (web searches) yielded the actual sources cited in the final response: Creative Commons Taiwan's article confirming the CC BY-NC-SA license, Kuro Hsu's iT 邦幫忙 profile (establishing credibility), and Tenlong bookstore listing (proving a paid physical edition exists). These search results contained the factual foundation for the answer.
• ›Steps [5], [6], and [8] (attempted direct page fetches to /intro/what-is-vuejs, /about, /ch1_getting-started) all returned 404s, indicating either non-existent routes or dynamic routing that the agent could not probe. The site offers no machine-readable sitemap or API endpoint for discovery.
• ›The agent had to rely on 60% prior knowledge (about Vue.js ecosystem alternatives like Vue Mastery and Vue School) and infer Kuro Hsu's standing from community presence, since the site itself does not present comparative positioning or market context.
• ›Critical gap: the agent explicitly noted it 'couldn't read the actual lesson content' and could not find the exact price of the physical book—information would be visible to a human browser but not to fetch-based navigation.