[mme]

Summary

The check workflow fails on the macOS runner in verify.service.spec.ts ("Stop ends a check and its descendant before removing the scratch worktree"): Error: kill EPERM, then a 5 s timeout. It failed twice in a row on #5, which was merged before this fix.

Cause: after a managed process group's leader exits, the runner sends a safety SIGKILL to the group from the close handler. macOS answers EPERM for a group whose leader is already gone, not ESRCH. The runner only tolerated ESRCH, so the throw escaped the close handler, the run promise never resolved, and the stop never finished.

 kill(signal)
   process.kill(-child.pid, signal)
-  ignore ESRCH
+  ignore ESRCH and EPERM   # nothing left to signal either way

Evidence

• Before: check runs 41055118166 and its rerun on fix(computer-use): tell Linux users that Hyprland is read-only #5, both failing on that test with kill EPERM and Test timed out in 5000ms.
• After: the process-runner test passes locally (Linux, where the group kill answers ESRCH); this PR's own check run is the macOS verification.

Merge Danger

Door: two-way

Blast Radius: small

One error code added to an ignore list in the shared process runner. A real permission problem signalling a group would now be silent; that cannot happen for groups the backend itself spawned.