Repository navigation
Machine Control desktop 0.7.0
- Give Desktop and Browser independent Off, Ask first and Automatic modes on
Windows, macOS and Linux, using one shared TypeScript policy engine and a
privately bundled Node runtime. - Add predictable takeover warnings, native approval, Start now, per-task
deferral, timed or manual pauses, Resume and Stop. Deliberate human input
interrupts desktop work; mouse movement alone does not. - Support explicitly enabled, named Chrome profiles across desktop platforms.
Approval andbrowser --profileselect exact profiles; connection loss,
stale references and Stop cannot redirect work to another profile. - Include supported CDP and page JavaScript in Browser permission. Keep
Desktop permissions independent, including browser access while locked. - Update Settings and native task controls for the shared permissions model.
Restart restores access choices but no live control authority; explicitly
enable the intended surfaces and browser profiles after upgrading. Named
profiles require the matching Chrome extension 0.9.0 or later. - Keep the Python CLI and SDK, automatically renewing supported live owners
under their original consent. Use--no-renewfor an intentionally finite
task; Stop, disconnect, claim loss and consent expiry still end access. - Preserve the actual refusal reason when browser profile discovery meets a
session transition, without dispatching or replaying the requested action. - Keep Linux warnings visible while Settings has focus, add an explicitly
installed read-only physical-activity observer, and reap the policy child
even when its native parent crashes. Ordinary effects remain AT-SPI/portal.
Targets all six desktop architectures. Bounded native ARM64 appliance evidence
for Windows, macOS and GNOME Wayland Linux is documented separately from
signed candidate acceptance. Physical human-input classification, other Linux
compositors and broader display/hardware configurations require qualification.