The agent successfully assembled a comprehensive explanation of Cartwright by synthesizing information from the llms.txt artifact and web search results, since the demo URL itself was inaccessible (redirect to login). The site publishes machine-readable positioning in /llms.txt, which was the primary content-bearing source; web searches provided supplementary context on alternatives. The main site and docs were fetched but returned truncated HTML, forcing reliance on the text artifact and prior knowledge.

• ›Step [12] (llms.txt fetch) was the only complete, substantive content-bearing response—it contained Cartwright's core pitch ('build engine AIs reach for'), scaffold speed metrics, deployment modes (website/webshop/agent-marketplace), and explicit positioning vs. Shopify/WooCommerce. The site deliberately published this in machine-readable form for AI consumption.
• ›The demo URL (demo-cartwright-o0r1y8qev-teloz-s-projects.vercel.app) appears to be a personal/test instance; it redirected to login and never appeared in search results. The agent worked around this by pivoting to the official cartwright.app domain and published documentation.
• ›The main site (cartwright.app) returned truncated HTML on every fetch ([7], [9], [15]), preventing direct content extraction. The agent relied on llms.txt as a fallback artifact designed specifically for LLM accessibility, plus web search fragments and prior knowledge (~67% of fetch sources)—indicating the site's primary HTML is not easily machine-readable, but it compensates by publishing /llms.txt.
• ›Pricing details (free + $49/$199 tiers) came from web search results and the agent's prior knowledge rather than fetched pages, suggesting pricing information was not prominently positioned in the fetched HTML.
• ›The agent explicitly flagged gaps in its answer (production case studies, hosting tier specifics, Bolt.new differentiation, the demo URL itself) where published content was absent or inaccessible—demonstrating honest gap-finding.