PixelCraftLab/sysadmin_troubleshooter

SysAdmin Troubleshooting Environment

A real-world OpenEnv environment where an RL agent acts as a Junior System Administrator. The agent interacts with a mock Linux shell to identify and resolve common server issues.

Environment Details

Action Space

SysAdminAction: A single shell command string.

• —command (str): The shell command to execute (e.g., ps aux, systemctl start nginx, kill 1024).

Observation Space

SysAdminObservation: Returns the result of the command and the current system state.

• —stdout (str): Standard output from the command.
• —stderr (str): Standard error from the command.
• —exit_code (int): Return code of the command.
• —system_state (dict): Summary of active services and running processes.
• —tasks_status (dict): Boolean status of the three target tasks.
• —reward (float): Partial progress reward (0.0 to 1.0 cumulative).
• —done (bool): True if all tasks are complete or max steps reached.

Tasks

The environment includes three tasks of increasing difficulty:

1. 1.[Easy] Rogue Process Cleanup: Identify a high-CPU process (rogue_app) using ps and terminate it using kill or killall.
2. 2.[Medium] Service Recovery: The nginx service is currently inactive. The agent must identify this and start the service using systemctl start nginx.
3. 3.[Hard] Configuration Fix: The nginx configuration has a typo (liten instead of listen). The agent must read the config file (cat /etc/nginx/nginx.conf), fix the typo (e.g., using sed), and restart the service.

Reward Function

The reward is based on task completion:

• —Task 1: +0.2
• —Task 2: +0.3
• —Task 3: +0.5 Total potential reward: 1.0.

Quick Start

1. Build and Start the Environment

# Build the Docker image
docker build -t sysadmin-env:latest -f server/Dockerfile .

# Run the container
docker run -p 8000:8000 sysadmin-env:latest

2. Run Inference

Ensure you have the required environment variables set:

export API_BASE_URL="your-api-endpoint"
export MODEL_NAME="your-model-name"
export HF_TOKEN="your-hf-token"

python inference.py

Spec Compliance

This environment implements the full OpenEnv spec:

• —Typed Pydantic models for Actions and Observations.
• —Standard step(), reset(), and state() endpoints.
• —Valid openenv.yaml manifest.
• —reproducible inference.py with mandatory [START], [STEP], and [END] logging.