T3 Code / agent-device lifecycle — real built server
Commit 5be6bc99c234b0db92ffb0b3ebd8746fd579112b
Environment: Linux x86_64; Node 26.12.1; fresh isolated state
No copied database. Android discovery tools; no booted device.
=== A. Graceful quit after an earlier server was killed ===
First server: real server PID 54971, authenticated RPC ready
RPC device.configure: local support ON, agent access ON
RPC device.list retry local host: initialize real agent tools
Environment daemon PID 55327, agent-device 0.21.31, HTTP health=True
SIGKILL server PID 54971: exited -9
Recorded daemon PID 55327 survives abrupt exit: alive=True, health=True
Restart, same state: real server PID 55462, authenticated RPC ready
Restarted server uses NO device RPC or agent tools
SIGTERM restarted server PID 55462: exited 130
After graceful quit: SAME daemon PID 55327: alive=True, HTTP health=True
=== B. Disable agent access after an earlier server was killed ===
Third server: real server PID 55788, authenticated RPC ready
RPC device.configure: local support ON, agent access ON
RPC device.list retry local host: initialize real agent tools
Environment daemon PID 56055, agent-device 0.21.31, HTTP health=True
SIGKILL server PID 55788: exited -9
Recorded daemon PID 56055 survives abrupt exit: alive=True, health=True
Restart, same state: real server PID 56191, authenticated RPC ready
Restarted server has not used device tools
RPC device.configure: agent access OFF (same operation as Devices settings)
Settings response: agentAccessEnabled=False
After disabling access: SAME daemon PID 56055: alive=True, HTTP health=True
DONE. Observations saved; cleanup follows.
Cleanup: all recorded server, hub and daemon processes stopped.