How it works · how to play · live Space info

Team Neuron — CyberSelfPlay cockpit for the OpenEnv Hackathon (Meta's OpenEnv

Flow

  1. New episode opens a persistent WebSocket to wss://harshitshri026-cyberselfplay-env.hf.space/ws (OpenEnv /ws on your Hugging Face Space). Episode state lives on the server, not in this browser tab beyond the open socket.
  2. Each Execute sends a real step with your CyberAction (actor, tool, target, params). Rewards, observations, and done are computed by CyberSelfPlayEnvironment in the Space.
  3. The auto opponent is a policy in this page (heuristics + randomness). It only chooses the next Red/Blue action; the environment still runs that action on the same remote session.
  4. Plain HTTP POST /step on the Space creates a new env per request (not stateful). This UI uses /ws so turns chain correctly.
  5. Tool names and scenarios below come from GET /info, loaded via same-origin /hf-cyber-env/info (rewrite to the Space).

How to play

  1. Press New episode to open a live session on the Hugging Face Space (wait if the Space is waking up).
  2. Choose SOC · Play as Blue (defender) or Red Cockpit (attacker). You only control that side; the other side moves right after each of your steps (scripted opponent in this UI).
  3. Pick a tool and target host. For Blue execute_instruction, set params.required_tool to match the current playbook step.
  4. Click Execute + auto opponent. Watch the Transcript and Scoreboard; repeat until the episode ends.
  5. The host map is a hint from alerts and recon — Blue does not see the full hidden state (partially observable game).

From Space /info

Loading…

Team Neuron · OpenEnv Hackathon · CyberSelfPlay

Play Blue or Red

Scoreboard (session)
Blue Σ reward
0.00
Red Σ reward
0.00
MTTD-ish
—
MTTR-ish
—
ρ_inst
—
Business
—
Blue tool
Hosts (blue view)
host-00host-01host-02
last action alert-linked other
Transcript
Click "New episode" to open a WebSocket to the Space, then pick a tool.