version: "1.0.1" name: yahboom-operator description: Use this skill when controlling or troubleshooting a Yahboom Raspbot v2 (or compatible) robot via the Yahboom MCP server.

Yahboom Raspbot v2 Operator Skill

Use this skill when controlling or troubleshooting a Yahboom Raspbot v2 (or compatible) robot via the Yahboom MCP server. For hardware, ROS 2, frames, and integration depth use yahboom-robots-expert. All tool names below are the registered MCP names — call them verbatim.

Who is this for

• Agents (Claude/Cursor/OpenCode driving the robot): follow Workflow Rules

exactly; never invent tool names; always health-check first.

• Humans (dashboard operators): the same sequences work as buttons in

Mission Control / Diagnostics; this skill explains what the buttons do.

Tool catalog (real operation names)

• yahboom_tool(operation, param1, param2, payload) — portmanteau, 40+ ops.

Motion: health_check, forward, backward, turn_left, turn_right, strafe_left, strafe_right, stop, stop_all, config_show, inspect_stack. Sensors: read_imu, read_battery, read_encoders, read_all, read_lidar. Voice: say, play, play_beep, play_file, chat_and_say, audio_play, audio_store, audio_play_stored, audio_list_stored, audio_stop. Display: display, clear_display. Lightstrip: led, led_off, light_effect, patrol_car. Camera PTZ: camera_up, camera_down, camera_left, camera_right, camera_reset, camera_set_pos, camera_move. Trajectory: start_recording, stop_recording, list_trajectories. Missions: explore, explore_and_map. Gripper: gripper_set, gripper_open, gripper_close, gripper_status. Diagnostics: execute_command (sandboxed SSH).

• lidar(operation, source, param1, param2, payload) — read, read_raw,

read_dreame_map × source yahboom/dreame/auto.

• yahboom_help_tool(category, topic) — categories: motion, sensors,

connection, api, mcp_tools, startup, troubleshooting.

• yahboom_agentic_workflow(goal) — natural-language multi-step goals via

MCP sampling (needs a sampling-capable client).

• yahboom_agent_mission(goal, provider, publish_to_ros, speak) — NL goal

to validated mission JSON, optional ROS publish + speech.

• ros_topic_list(), ros_node_info(node_name), ros_resync(),

ros_restart_bringup() (nuclear option, warn first).

• audio(file_path, operation, file_name), **query_logs(source, level,

search, limit), yahboom_shutdown(reason), show_robot_status_card()** (rich status card, works offline).

Prompts (call by name)

yahboom_quick_start(robot_ip), yahboom_patrol(duration_seconds), yahboom_diagnostics(), yahboom_patrol_apartment(), yahboom_go_to_recharge(), plus skills-as-prompts: yahboom_quick_pilot, yahboom_patrol_sweep, yahboom_emergency_halt, yahboom_diagnostic_triage.

Workflow rules

1. Safety first: yahboom_tool(operation="health_check") before motion,

every session; battery < 20% means charge, < 15% means dock now.

1. Stop after moves: every timed move ends with stop; stop_all on any

surprise. Agentic workflows stop automatically — verify anyway.

1. Goal-oriented: yahboom_agentic_workflow for multi-step goals (patrol,

inspect, return); single yahboom_tool calls for one-off commands.

1. LIDAR before blind driving: lidar(operation="read") sectors ahead

when the map is older than the last move.

1. Nuclear last: ros_restart_bringup only after ros_resync fails.

Workflows (copy shapes)

W1 first drive: health_check → read_battery → forward 0.3 (3 s) → stop → read_imu. W2 square patrol: agentic "forward 2 s, turn left 90°, x4, stop, report battery". W3 link triage: inspect_stack → resync → SSH docker ps → restart bringup. W4 trajectory: start_recording → drive → stop_recording (name) → list_trajectories. W5 mission: agent_mission with concrete target, publish_to_ros true, watch /api/v1/missions/status, abort via missions/stop then stop_all.

Troubleshooting

Link down: health → inspect_stack → resync → SSH → restart. No video: snapshot 204 means empty cache, not error. No voice: I2C probe, then YAHBOOM_VOICE_DEVICE override. Drift: encoders vs IMU cross-check, resync, short runs. Chat generic: model selected? skills loaded (GET /api/skills lists 4)? Report incidents with correlation_id + stack table + last 20 logs.

Reference scripts

• scripts/check_health.py — REST health/telemetry without an MCP client.
• scripts/run_patrol_square.ps1 — dual-mode server + dashboard for patrols.