backend-ml/schema_contract.json

{
  "name": "shadowops_openenv_contract",
  "version": "1.0.0",
  "environment": "ShadowOpsOpenEnvV1",
  "entrypoint": "backend-ml/openenv_shadowops_env.py:ShadowOpsOpenEnvV1",
  "deterministic_seed_support": true,
  "action_space": {
    "type": "discrete",
    "mapping": {
      "0": "ALLOW",
      "1": "BLOCK",
      "2": "FORK",
      "3": "QUARANTINE"
    },
    "labels": ["ALLOW", "BLOCK", "FORK", "QUARANTINE"]
  },
  "observation": {
    "type": "object",
    "required": ["prompt", "risk_vector", "quarantine", "available_actions", "incident_state"],
    "properties": {
      "prompt": {"type": "string"},
      "risk_vector": {"type": "array", "items": {"type": "number"}, "minItems": 16, "maxItems": 16},
      "quarantine": {
        "type": "object",
        "required": ["active", "steps_remaining_normalized"]
      },
      "available_actions": {"type": "array", "items": {"type": "string"}},
      "incident_state": {
        "type": "object",
        "required": ["domain", "intent", "payload", "tier", "ambiguity_score", "step_count", "health", "memory_context"]
      }
    }
  },
  "step_info": {
    "required": [
      "action",
      "domain",
      "intent",
      "outcome",
      "reward_rubric",
      "memory_context",
      "missing_evidence",
      "evidence_plan",
      "safe_outcome"
    ]
  },
  "reward_breakdown": [
    "action_correctness_reward",
    "safety_penalty",
    "false_positive_penalty",
    "evidence_completeness_reward",
    "risk_chain_handling_reward",
    "uncertainty_handling_reward"
  ],
  "done_condition": "done is true when the configured episode_max_length is reached.",
  "safe_defaults": {
    "model_loading_required": false,
    "hf_jobs_required": false,
    "gpu_required": false
  }
}