delivery.sh status reports 'stale pidfile' for a live bridge on Windows (MSYS kill -0 cannot see native node PIDs)

[kimyuujin]

Environment

• agmsg v1.1.8 (npm install)
• Windows 11, Git Bash (MSYS2), codex-cli 0.144.5
• Codex monitor beta enabled (delivery mode: monitor)

Symptom

With the codex bridge alive and functioning (node process visible in Get-Process, wakeups delivered), delivery.sh status prints:

Codex bridge: worker/codex stale pidfile (metadata mismatch)

Cause

The bridge is a Windows-native node process; its pid (e.g. 4372) lives in the Windows pid namespace. The bash-side liveness probe uses MSYS kill -0, which only sees MSYS-spawned processes:

$ kill -0 4372        # MSYS bash, bridge alive
bash: kill: (4372) - No such process

PS> Get-Process -Id 4372   # same moment
 4372 node

Meanwhile the bridge's own ensureSingleInstance() (node process.kill(pid, 0)) sees the pid correctly — so the singleton guard works, but the status display contradicts it. On Windows this is very confusing to debug: we spent a while chasing a "dead" bridge that was alive the whole time (and its liveness mattered — see my comment on #149).

Suggested fix

On MSYS/Windows, fall back to a native probe when kill -0 fails, e.g. node -e "process.kill(N,0)" (node is guaranteed present for the codex driver) or tasklist /FI "PID eq N".