Play Doom with computer-use agents as an informal benchmark
The OpenClaw team used computer-use agents to play Doom on remote machines, treating game performance as an informal way to improve the computer-use workflow. Task: Suggested implementation: Deploy OpenClaw computer-use agents to play Doom on remote machines via the Crabbox execution environment. Human operators monitor sessions through VNC or SSH to observe agent behavior. The workflow treats game performance as an informal benchmark for computer-use capabilities.; tools: OpenClaw. Human role: The original post does not say how a person stayed involved.; limitations: The source provides no details on setup costs or resource requirements. Crabbox lacks built-in cost limits, requiring manual spending caps for remote instances. Human oversight is necessary to verify agent actions and ensure workflow stability..
Tools
Published on Agent Work Atlas:
Overview
The OpenClaw team used computer-use agents to play Doom on remote machines, treating game performance as an informal way to improve the computer-use workflow.
Workflow
Suggested implementation: Deploy OpenClaw computer-use agents to play Doom on remote machines via the Crabbox execution environment. Human operators monitor sessions through VNC or SSH to observe agent behavior. The workflow treats game performance as an informal benchmark for computer-use capabilities.
Reported result
No specific scores or quantitative results are available for the Doom runs. Vincent Koc reports that computer use currently accounts for approximately 10% of his workload, primarily involving navigation of infrastructure interfaces and verification screens.
Human role
The original post does not say how a person stayed involved.
Limitations
The source provides no details on setup costs or resource requirements. Crabbox lacks built-in cost limits, requiring manual spending caps for remote instances. Human oversight is necessary to verify agent actions and ensure workflow stability.
Try it
Act as an AI agent operator. I want to pilot a computer-use benchmark using OpenClaw agents on remote machines. Context: We are testing agent stability and responsiveness in a controlled gaming environment. Inputs: [Replaceable: OpenClaw agent configuration], [Replaceable: Crabbox remote machine credentials], [Replaceable: Doom game executable path]. Steps: 1. Provision a remote machine via Crabbox and install Doom. 2. Configure the OpenClaw agent to connect to the machine's VNC/SSH interface. 3. Launch the agent and instruct it to play Doom for a 5-minute session. 4. Monitor the session via VNC to verify the agent's actions and responsiveness. 5. Record any errors, latency issues, or successful interactions. Output: A brief report detailing session stability, agent response times, and any observed failures. Quality Check: Ensure the agent completes basic navigation tasks without crashing. Safeguards: Do not execute any commands outside the game environment. Stop the session immediately if the agent attempts to access system files. Require human confirmation before starting the actual run. If inputs are missing, ask for specific configuration details.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.