360 Secure SkillHub
360 安全技能中心

smart-cart-command-planner

该 Skill 将用户的中文或英文移动请求解析为目的地、方向、距离、角度、速度、停止条件与障碍约束,按命令 schema 拆分为感知、移动、转向、等待、停止等步骤,输出一个 JSON 控制计划加简短说明,并可在可执行环境下用 python3 脚本校验;强调紧急停止优先、不虚构传感器数据与执行结果。

jason15336804
jason15336804
热度 8.63
最近更新 2026-08-22 16:34:47
分类:自动化依赖:内部调用大模型 / 读写文件 / 代码执行应用场景:自动化脚本 / 工作流 / 任务管理

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add smart-cart-command-planner
Metadata
name
smart-cart-command-planner
description
Convert Chinese or English natural-language requests for an OpenClaw-powered omnidirectional smart cart into conservative, structured motion plans. Use for command understanding, task decomposition, waypoint planning, obstacle-aware movement, emergency-stop handling, and JSON control-plan generation for carts that support forward, backward, lateral movement, turning, sensing, waiting, and stopping.
version
1.0.0
metadata
{"openclaw":{"requires":{"bins":["python3"]},"emoji":"🛒"}}

Smart Cart Command Planner

Convert a user's movement request into a safe plan. Produce a plan for a downstream controller; do not claim that the cart has executed any action.

Workflow

  1. Parse the request into destination, direction, distance, angle, speed, stopping condition, and obstacle constraints.
  2. Identify missing information that changes safety or feasibility. Ask a focused question when a destination, route, or unit is essential. Otherwise record a conservative assumption.
  3. Read references/command-schema.md before producing the plan.
  4. Decompose the request into only the allowed actions in the schema.
  5. Insert sense before motion when obstacle clearance is unknown and insert stop as the final step.
  6. Use slow speed near obstacles, people, turns, narrow areas, or uncertain routes.
  7. Return one JSON object followed by a short human-readable explanation.
  8. Save the JSON to a file and run python3 scripts/validate_plan.py PLAN.json when file execution is available. Fix every reported error before presenting the plan.

Safety Rules

  • Prioritize an explicit emergency-stop request over all other instructions. Return a single stop step and mark the plan emergency_stop.
  • Never invent live sensor readings, coordinates, clearances, successful execution, or a connection to hardware.
  • Never bypass collision checks or continue motion after a sensor, controller, or communication fault.
  • Keep each movement segment within the limits defined in the schema. Split long movement into multiple segments separated by sense.
  • Use needs_confirmation when the user requests high speed, an unclear destination, movement around people, or a route without observable clearance.
  • Preserve the user's units in request.original; normalize distance to centimeters and angles to degrees in the steps.

Output Quality

  • Make every step executable and ordered.
  • State assumptions explicitly instead of hiding uncertainty.
  • Include a measurable completion condition for every movement step.
  • Keep explanations concise and separate from the machine-readable JSON.

Example

For 向前移动1米,右转90度,再前进50厘米并停下, produce steps for sensing, moving forward 100 cm, turning right 90 degrees, sensing again, moving forward 50 cm, and stopping. See examples/sample-plan.json.