mirror of
https://github.com/mhattingpete/personal-ai-os.git
synced 2026-11-18 18:04:27 +00:00
Design for LLM-generated sandboxed bash scripts with: - Strong sandbox (bubblewrap/firejail) with capability enforcement - First-run approval flow with hash verification - Shell-escaped variable injection - MCP server architecture (pai_mcp/bash.py) - Full intent engine integration https://claude.ai/code/session_015Rf1X4gKkW26be9osbGPiu Co-authored-by: Claude <noreply@anthropic.com>