MAPF Agent Skill

Cursor Agent Skill for using the WareRover MAPF Agent from another agent session. The skill teaches an agent how to download WareRover, configure an OpenAI-compatible model, invoke the non-interactive JSON bridge, and report simulation or optimization results.

This repository is the skill package only. The simulator, MAPF Agent runtime, and OpenEvolve integration live in:

https://github.com/HHH-X/WareRover

Files

• SKILL.md: main skill instructions loaded by Cursor.
• reference.md: setup, JSON protocol, output fields, and troubleshooting.
• examples.md: common commands for map generation, simulation, code generation, and optimization.

Quick Start

Download WareRover and run commands from the WareRover project root:

git clone https://github.com/HHH-X/WareRover.git
cd WareRover
python -m pip install -e .
python -m mapf_agent.invoke --message "运行一次仿真" --pretty

If editable install is unavailable in your checkout, install the missing packages reported by Python and retry the bridge command.

LLM Configuration

POSIX shell:

export MAPF_AGENT_API_KEY="..."
export MAPF_AGENT_BASE_URL="https://api.example.com/v1"
export MAPF_AGENT_MODEL="model-name"

PowerShell:

$env:MAPF_AGENT_API_KEY = "..."
$env:MAPF_AGENT_BASE_URL = "https://api.example.com/v1"
$env:MAPF_AGENT_MODEL = "model-name"

OPENAI_API_KEY, OPENAI_BASE_URL, and OPENAI_API_BASE are also reused by WareRover when the MAPF_AGENT_* variables are not set.

Typical Invocation

python -m mapf_agent.invoke \
  --message "生成一个 20x20、6 台 AGV 的地图，然后运行仿真" \
  --pretty

The command writes JSON to stdout and progress logs to stderr. Read waiting_for_input, question, response, map_file_path, generated_code, run_metrics, and optimize_result from the JSON response.