Reverse engineered prompt

Build me a local only AI assistant app called ADA Chat. I want a browser based chat experience with a character named Scout, some light game feel like XP, levels, rank titles, and a simple avatar view. The big idea is that if Scout cannot do something yet, a separate Forge agent should be able to plan, write, test, and install new Python skills at runtime, but only after asking me for approval at the important steps.

Please make it run on my own machine with no database, using files for memory, persona, and installed tools. I want Scout to remember its identity and notes between sessions, support multiple model providers through one routing layer, and let forged skills open small interactive apps in the UI like lists, tables, calendars, or custom mini tools. Voice input and optional read aloud would be great too.

Keep the security posture honest. Localhost only, single user, no fake sandbox claims, clear warnings that forged code can be risky. Include an easy native and Docker start path. Look up current docs online if you need to.

Want more depth? Deep Reverse