spec_version: 1
name: suspect-x-env
display_name: "Suspect X — AI Interrogation Room"
description: >
  Two-agent adversarial RL environment. An Interrogator LLM tries to extract
  hidden facts from a Suspect LLM. Reward is 100% deterministic — no LLM judge.
  Supports multi-agent self-play and curriculum via n_facts parameter.
type: space
runtime: fastapi
port: 7860
themes:
  - multi-agent
  - self-improvement