Right now the MCP server is all-or-nothing: --read-only lets the AI only look, or you turn it off and it can run anything on any host.
It'd be great to set per tag how far an agent may go (look / change / sudo / all), let purple's blast-radius check block anything over the line and hand back a reason and manage it all from a new AI tab with the rules plus a live activity feed. Safe by default. If you set no policy, nothing changes.
Right now the MCP server is all-or-nothing: --read-only lets the AI only look, or you turn it off and it can run anything on any host.
It'd be great to set per tag how far an agent may go (look / change / sudo / all), let purple's blast-radius check block anything over the line and hand back a reason and manage it all from a new AI tab with the rules plus a live activity feed. Safe by default. If you set no policy, nothing changes.