sundarsub/execwall — reverse-engineered prompt

Build me Execwall, a secure execution firewall for AI agents. I want it to sit between an agent like OpenClaw and the computer, so commands only run if they match a policy, dangerous stuff gets blocked, and every attempt is logged clearly.

It should have an interactive shell where I can type commands and see whether they are allowed or denied, plus an API mode so another agent or bot can send requests. Add a locked down Python runner for untrusted code with timeouts, resource limits, no network by default, and useful error output. Policies should be editable in a YAML file, with per user rules, audit mode, rate limits, and JSON audit logs.

Please include setup scripts and default profiles so I can deploy it on a Linux VM, especially Oracle Cloud Free Tier. Also include guidance for running it with OpenClaw and connecting it to WhatsApp or Telegram. Look up current docs online if you need to.