feat(go2): add read-only local Wi-Fi diagnostics

[kezaer]

Summary

• Add dimos go2tool doctor for read-only Go2 local Wi-Fi teleop diagnostics.
• Report robot discovery, signal probe state, local interfaces, active DimOS run state, expected UI listeners, suggested LAN URLs, and optional passive color_image receipt.
• Support JSON output, strict mode, custom UI ports, custom image topics, and signal/image timeouts.
• Treat a failed signal-port probe as a warning when live image data proves the WebRTC data plane is active.

Why

Go2 local Wi-Fi failures often look the same from the browser: stale robot IPs, stale operator-machine IPs, loopback-only listeners, missing video, and dead WebRTC sessions all produce blank or unreachable UI states. The doctor gives operators a no-motion way to identify which layer is actually broken.

Safety

• The doctor is read-only and sends no movement commands.
• It is a troubleshooting helper, not part of benchmark success scoring or startup control flow.

Validation

• uv run pytest dimos/robot/unitree/go2/cli/test_doctor.py -q
• uv run ruff check dimos/robot/unitree/go2/cli/go2tool.py dimos/robot/unitree/go2/cli/doctor.py dimos/robot/unitree/go2/cli/test_doctor.py
• Live smoke: dimos go2tool doctor --robot-ip <go2_ip> --no-discover --connect-timeout 0.8 --check-image --image-timeout 1.0 against a running teleop-phone-go2 stack.

No robot motion commands were sent during validation.

[greptile-apps]

Greptile Summary

This PR adds dimos go2tool doctor, a read-only diagnostic command for local Wi-Fi Go2 teleop setups. It introduces doctor.py (robot signal probing, interface enumeration, TCP listener detection with psutil+lsof fallback, passive image receipt) and wires it into go2tool.py with flags for JSON output, strict mode, custom ports, and image topic overrides.

• doctor.py: Two-tier TCP listener detection (psutil first, lsof fill-in for any missing required ports) with LAN-vs-loopback classification; signal-probe FAIL is downgraded to WARN when a live color_image frame proves the WebRTC data plane is active.
• go2tool.py: New doctor sub-command that resolves robot IP from the CLI flag, parent context, or global_config, then delegates to collect_report and exits non-zero under --strict when any WARN/FAIL is present.
• test_doctor.py: Ten unit tests covering discovery, port classification, lsof merge/skip logic, image-subscription cleanup, runtime-evidence adjustment, and URL suggestion.

Confidence Score: 5/5

Safe to merge — the change is entirely additive (new CLI sub-command + tests), sends no robot commands, and all execution paths terminate cleanly.

All three files are new additions with no modifications to existing logic. The diagnostics are read-only, the two-tier TCP listener fallback is well-guarded, and the runtime-evidence adjustment is straightforward. The one cosmetic message inaccuracy does not affect correctness or safety.

No files require special attention.

Important Files Changed

Filename	Overview
dimos/robot/unitree/go2/cli/doctor.py	New 670-line diagnostics module: robot probe, local-interface enumeration, TCP listener detection (psutil + lsof fallback with partial-results merge), port LAN-accessibility classification, passive image subscription check, and report formatting. One minor message accuracy issue when LAN discovery is explicitly disabled; all logic and edge-case handling is otherwise sound.
dimos/robot/unitree/go2/cli/go2tool.py	Adds the doctor subcommand to the existing go2tool Typer app; wires CLI flags (robot-ip, discover/no-discover, timeouts, ui-port overrides, check-image, strict) to collect_report. IP/port resolution logic is correct and the strict/JSON output paths look clean.
dimos/robot/unitree/go2/cli/test_doctor.py	Ten focused unit tests covering discovery, port classification, lsof merging/skipping, image subscription cleanup, robot-check adjustment, and URL suggestion. All patches target the correct module-level names and the assertion coverage matches the addressed review feedback.

Sequence Diagram

sequenceDiagram
    participant CLI as go2tool doctor (CLI)
    participant CR as collect_report
    participant RB as check_robot
    participant LD as LAN discover
    participant SP as signal probe (HTTP)
    participant IF as local_interfaces
    participant RN as check_run
    participant CP as check_ports
    participant PU as psutil.net_connections
    participant LO as lsof fallback
    participant IM as check_image
    participant PS as pubsub subscribe
    participant AD as _adjust_robot_check

    CLI->>CR: collect_report(robot_ip, ...)
    CR->>RB: check_robot(robot_ip)
    RB->>LD: discover(timeout)
    LD-->>RB: "[]|[Go2Device]"
    RB->>SP: GET /con_notify
    SP-->>RB: "ok|timeout"
    RB-->>CR: RobotCheck
    CR->>IF: local_interfaces(resolved_robot_ip)
    IF-->>CR: [LocalInterface]
    CR->>RN: check_run()
    RN-->>CR: RunCheck
    CR->>CP: check_ports(endpoints, probe_hosts)
    CP->>PU: "net_connections(kind=tcp)"
    PU-->>CP: connections
    alt required port missing
        CP->>LO: _lsof_tcp_listeners()
        LO-->>CP: "{port: [hosts]}"
    end
    CP-->>CR: [PortCheck]
    CR->>IM: check_image(topics, timeout)
    IM->>PS: subscribe_pubsub_uri(topic, cb)
    PS-->>IM: "frame | timeout"
    IM-->>CR: ImageCheck
    CR->>AD: _adjust_robot_check_with_runtime_evidence
    AD-->>CR: RobotCheck (possibly FAIL to WARN)
    CR-->>CLI: Go2DoctorReport
    CLI->>CLI: format_report / json.dumps

Loading

_{Reviews (3): Last reviewed commit: "Address Go2 doctor review comments" | Re-trigger Greptile}

[kezaer]

Follow-up stack for turning this doctor into startup automation:

1. feat(core): add blueprint preflight hooks kezaer/dimos#1 - reusable blueprint/hardware preflight infrastructure
2. feat(go2): validate startup IP with LAN discovery kezaer/dimos#2 - Go2 startup autodetect/fail-closed connectivity preflights
3. feat(go2): harden LAN teleop startup and browser control kezaer/dimos#3 - Go2 LAN teleop/browser setup hardening and troubleshooting docs

I kept these as draft stacked PRs in the fork because I do not have permission to push upstream stack base branches to dimensionalOS/dimos. The review order is intended to be #2378 first, then the three follow-ups above.