Problem

Running the ps1 test suite on a machine that has a live psmux kills the user's running sessions and can clobber their config, with no warning. An unsuspecting would-be contributor who runs the tests for the first time without reading them gets an unpleasant surprise.

This is presumably by-design for the throwaway Docker dev container / CI, but nothing in the scripts prevents — or warns about — a destructive run on a normal dev machine.

Evidence

Process killing (kills ALL psmux processes by image name, not just the test's):

• tests/run_all_tests.ps1 Clean-Server runs Get-Process psmux | Stop-Process -Force before every test (line ~180).
• Individual tests do it independently, e.g. taskkill /f /im psmux.exe (test_bind_pipe_key.ps1:28, test_bug_detection.ps1:18) and Get-Process psmux | Stop-Process (test_agent_teams_e2e.ps1:127, test_cc_iterm2_compat.ps1:100).

Runtime/config files (operates on the real %USERPROFILE%):

• run_all_tests.ps1 Clean-Server deletes ~/.psmux\*.port, *.key (lines ~184-185) and ~/.psmux.conf, ~/.psmuxrc (lines ~187-188) unconditionally between tests.
• Config tests overwrite the real ~/.psmux.conf, e.g. test_bind_key.ps1:284. They attempt to back up and restore it, but the restore is inline (not in a finally), so an error or the runner's 600s Stop-Job timeout between the overwrite and the restore leaves the user's config replaced with test content.

Impact

A contributor who runs the promising-looking tests/run_all_tests.ps1 on their workstation loses their running psmux sessions and risks losing their ~/.psmux.conf.

Ask

Make the suite safe to run on a normal dev machine — at minimum, refuse to run (or otherwise prevent host damage) unless it's clearly in a throwaway environment. Not everyone has the possibility to run Windows containers.