Screenshot to Working App in Hermes: One-Prompt Clone Test with Fugu Ultra

Key Takeaways

• The workflow: attach a screenshot in Hermes Agent → give it one prompt to "build an exact clone and run it locally" → Hermes drives the whole build and launches the app on the local system.
• The model under test is Sakana's Fugu Ultra — an orchestration model that doesn't answer directly but routes every task across a pool of frontier models, claiming to stand shoulder-to-shoulder with the top models.
• Benchmarks skipped on purpose. Fugu reportedly "beats everything" on the vendor's own numbers; Fahd ignores those and judges by one real end-to-end task, on the view that flashy launches usually fall short of the hype.
• Hermes as a general harness: the demo shows Hermes taking a visual input and producing a running local app — a reminder that Hermes is an agent runtime, not just a chatbot.
• Practical lesson: treat tall orchestration-model claims skeptically and validate with a single real prompt before trusting the leaderboard.