OpenRig’s new release separately controls how much automatic input reaches each agent session and how far other agents may intervene. Held input remains readable and is not replayed automatically. The session view shows permissions for the next fresh launch. The release also stops rewriting project permissions when starting managed Claude sessions, while leaving files changed by earlier versions for manual restoration.
OpenRig, a tool for running coding-agent teams in terminal sessions, introduces separate typing and intervention controls in version 0.7.1. Its release notes give each session an automatic-input policy: disabled, urgent input from OpenRig only, urgent only, or hold. The urgency rules determine which incoming messages qualify for automatic typing.[1]
The hold policy keeps incoming material readable without replaying it into the terminal later. A separate terminal policy controls peer agents: no peer interventions, help within instructions, or full control. Compaction and model changes are stated exceptions to that boundary. Existing sessions retain full control rather than being silently switched to a tighter policy.[1]
The status view shows the permissions that apply to the next fresh launch, together with where those permissions came from. Managed Claude launches also stop rewriting the local defaultMode setting to acceptEdits. Files already changed by older OpenRig versions are left untouched, so the release does not automatically restore their earlier contents.[1]
Team skills now live in each session’s own storage rather than project directories, and assigning team roles no longer overwrites shared project guidance. Managed compaction can resume after a daemon restart. The release thus includes changes to stored instructions and running-session continuity alongside the new intervention policies.[1]
