Remote process cleanup proves only original PGID emptiness, not escaped descendants

[Calmingstorm]

Severity: P2

Verified at v4.10.0, 55c15761845b00931ab2ea4aa184a7797dfd0614.

Locations

src/tools/process_manager.py:57-157,187-200,250-266

Failure scenario and affected installations

Remote Linux/macOS jobs whose descendant setsid escapes original group. Supervisor/controller signal/inspect onlyoriginalPGID and can emit empty=true while escaped child survives completion/kill/deadline.

Verification

Independently traced embedded supervision ownership/group checks and absence of ancestry/adoption. Reviewer real embedded code with mocked operations emitted emptytrue/exit0 for absentgroup plus survivingescapedchild.

Correction and regression coverage

Contain/track all owned descendants or report limited group-only evidence honestly; add harmless escaped-descendant model fixture.

Duplicate screening

Checked open/closed inventory and GitHub title/body search; no matching root. Harmless mocked operations/temporary data and complete source traces; no live processes killed, remote calls or service actions.

[Calmingstorm]

Validated as fixed on campaign/v4.10.0-review (PR #629) at 3047c46c: the fix was reviewed against the final code, and its regression tests pass on the branch and in CI. Round-1 correction 2b06c963 (+ 7db8c15a): a verified-empty remote process group now settles as exited/killed with an explicit process_group_only caveat that escaped descendants are unverified, instead of every remote outcome being unknown; a real escaped-process test pins the caveat. It is not merged to master yet. Closing per the maintainer's instruction; reopen if a later change regresses it.