You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Interactive robot work uses the current directory as a persistent project.
A first-use trust screen offers this directory, another directory, or a new
one; --workspace DIR chooses one explicitly and --trust-workspace
accepts it for scripts. Existing project files and directory permissions
are retained. Robot documentation and tools are separate under /robot;
per-run config and robot logs live in the project's .openrua/runs/.
Project guidance helps the agent organize code, observations and handoff
notes. Environment selections are saved per project without credentials.
Opening a project starts a new conversation. --resume and /resume
search only that directory; --all and the picker's Tab switch include
all projects and older records. Confirmed ended project conversations can
continue their native agent conversation on fresh robot resources, retaining
model and permission settings, without replaying old actions or queued work.
The next instruction carries a reminder to inspect the current scene.
Missing project directories keep their history read-only instead of being
recreated during resume.
Upgrade: --workspace on interactive entry points now names the project
directory itself, which can contain existing files. Previously it named
an empty run directory containing a workspace/ child. Existing records
remain readable through --resume --all. Recorded task and benchmark
trials retain their fresh-workspace protocol and original starter template.