The agent successfully assembled a comprehensive overview of Interact by combining documentation fetched from the site itself (markdown reference pages) with GitHub repository metadata and README content. The site's interact.combostrap.com domain proved minimally navigable for discovery—it returned 404s on conventional paths like /pricing, /features, /about, and /docs—forcing the agent to guess the /reference/*.md pattern and eventually pivot to GitHub to retrieve the actual product description and license information. The final answer is complete and accurate but required substantial off-site bridging.

• ›The site's homepage (/) loaded but appeared to be a dynamic React app that rendered no discoverable navigation or links in its HTML response; the agent had to extract the README and package.json from GitHub raw content to answer 'what they do' and 'how it's priced' questions.
• ›The /reference/*.md pattern was reverse-engineered after manual 404 discovery; once the pattern was identified (step 9), subsequent reference pages (layout, markdown, conf, component, rsc, mdx, ai, cli, build) all returned 200 with technical content, showing the site does host structured docs but does not expose a discoverable index or navigation.
• ›Pricing information was absent from the site itself; the agent had to inspect package.json on GitHub (step 23) to identify the FSL-1.1-ALv2 license and conclude Interact is open-source and free. The site published no business model, pricing page, or licensing statement.
• ›The /index.md fetch (step 7) was the only successful content retrieval from a non-reference path, confirming the site uses file-based routing but does not advertise it; the agent had to guess and test this pattern.
• ›GitHub search (step 20) was critical for discovery; the web search did not return interact.combostrap.com results directly, so the agent had to infer the GitHub org from meta-search and navigate to the repository for the README—the main product narrative was not served by the site itself.
• ›The site is agent-readable via markdown endpoints (documented in step 26's /reference/ai.md) but the discoverability layer is broken; a human or agent with prior knowledge of the /reference/ pattern can retrieve docs efficiently, but cold discovery requires external lookups.