The agent successfully explained what dollarpe does, who it's for, pricing, and differentiation by retrieving two GitHub README files that contained the core documentation. The main website (dollarpe.cristianbgp.com) was nearly empty—just a typewriter animation—forcing the agent to rely entirely on GitHub repositories to answer the task, which it cited explicitly as its sources.

• ›Steps [6] and [7] returned the actual README files from dollarpe-cli and dollarpe-api repositories, which contained all substantive information: that dollarpe aggregates USD→PEN exchange rates from 9 Peruvian exchanges plus SUNAT, offers a CLI and REST API, is free and open-source, and serves a Peru-specific use case.
• ›The agent had to guess the GitHub URLs via prior knowledge (steps [3] and [4] were fetches to github.com pages that returned only HTML boilerplate, not the rendered README content). The agent then directly fetched the raw markdown files ([6], [7]), which contained the actual documentation the main site should have surfaced.
• ›The main website is extremely poor for agent discoverability—it renders almost no visible content to convey what the product is, who it's for, or how to use it. The site acts as a portal to GitHub repositories rather than self-documenting. An agent without prior knowledge of the GitHub repos or without the ability to infer file paths would fail to complete this task.