Show HN: Vibe coded an AI chat app with features I wanted, Poe (github.com)

Very cool, always great to build projects because you need them yourself.

From the screenshots it seems that there's a "tool" in the list of tools provided to the LLM for command line utilities like `rm`, `mkdir`, `ls` and so forth.

Just a small piece of advice: you might want to look into further. You can also expose the command line as a single tool, and most LLMs will be able to provide pretty good formatted commands. You could still filter invalid or allowed and non-allowed commands out within the tool that's actually being called by the LLM.

Just wanted to share that!

would be interested in reading your outlook on the execution model, isolation, guardrails, tool calling. those are some of the baseline things i evaluate before try before trying an agentic env