AgentSkillsCN

request_interpreter

阶段 1：将原始的自然语言用户请求映射为结构化的 JSON 场景数据格式。适用场景：(1) 初次启动机器人工作单元设计任务；(2) 处理用户初步输入的机器人任务描述；(3) 将对话式需求转化为结构化数据。系统会填充已知字段，并将未知项标记为 null 或 TBD，以便后续进一步解析。

SKILL.md
--- frontmatter
name: request_interpreter
description: "Stage 1: Takes raw natural language user requests and maps them to structured JSON scene data format. Use when: (1) Starting a new robot workcell design task, (2) Processing initial user input describing robot tasks, (3) Converting conversational requirements into structured data. Fills known fields and marks unknowns as null or TBD for further resolution."
metadata:
  stage: "1"

Request Interpreter

Overview

Converts natural language requests into structured scene data format. This is the ENTRY POINT for translating user requirements into the Golden JSON structure used throughout the system.

When to Use This Skill

  • •User describes a new robot workcell design task
  • •Starting fresh with natural language input
  • •First step of Stage 1 (Interpretation)

Workflow Instructions for Agent

Step 1: Extract User Request

  • •Get the user's natural language description of their workcell needs
  • •If user provides partial info, ask them to describe the complete task

Step 2: Execute Script Run the interpretation script:

bash
python skills/request_interpreter/scripts/interpret_request.py --text "<user_description>"

Step 3: Parse JSON Output The script outputs JSON with structure:

json
{
  "partial_scene_data": {
    "robot_configuration": { "model_id": "franka_panda" or null },
    "workcell_assets": [{"type": "table", "model_id": "TBD", "position": null}],
    "task_type": "pick_place" or null,
    "task_description": "<original text>"
  }
}

Step 4: Present Results Show the user:

  • •What information was successfully extracted
  • •What fields are missing (null or TBD)

Step 5: Next Action

  • •If there are null/TBD fields → Load gap_resolver skill
  • •If everything is complete → Load robot_selector skill

CRITICAL: Save the partial_scene_data - you'll need it for subsequent skills!

Usage

bash
python scripts/interpret_request.py --text "Set up a packing station with a Panda robot"

What It Does

  • •Extracts robot model mentions (Panda, UR5, etc.)
  • •Identifies task type (pick_place, assembly, packing)
  • •Detects workspace furniture (table, workbench)
  • •Recognizes common objects (box, apple, bottle)
  • •Marks unresolved fields as null or TBD

Input Schema

{ "text": "Set up a packing station with a Panda robot." }

Output Schema

{ "partial_scene_data": { "robot_configuration": { "model_id": "franka_panda" }, "workcell_assets": [], "constraints": null } }