Follow-up from #2971 review. Arms differ in how subject-owned background processes survive agent shutdown vs framework timeout (see the DSH PTY-tree teardown found in the 89-task run). Define one policy, apply it to every arm, and test it. (Raised by @M4n5ter in the approving review.)
Follow-up from #2971 review. Arms differ in how subject-owned background processes survive agent shutdown vs framework timeout (see the DSH PTY-tree teardown found in the 89-task run). Define one policy, apply it to every arm, and test it. (Raised by @M4n5ter in the approving review.)