Spaces:
Sleeping
Sleeping
File size: 6,922 Bytes
543a85f 63c7e0c 543a85f 63c7e0c 543a85f 63c7e0c 543a85f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 | """
PrivilegeDesk Environment — OpenEnv-compatible wrapper.
Exposes the WorldState over the standard OpenEnv Environment interface.
"""
from typing import Any, ClassVar, Optional
from uuid import uuid4
from openenv.core.env_server.interfaces import Environment
from openenv.core.env_server.types import State
try:
from ..models import PrivilegeDeskAction, PrivilegeDeskObservation
except ImportError:
from models import PrivilegeDeskAction, PrivilegeDeskObservation
from env.world_state import WorldState
class PrivilegeDeskEnvironment(Environment[PrivilegeDeskAction, PrivilegeDeskObservation, State]):
"""
Zero-Standing-Privilege Ops Environment for training AI agents on IAM tasks.
The agent is dropped into a synthetic enterprise and must handle:
Task 1 (easy): Access Decision — approve/deny a single access request
Task 2 (medium): JIT Escalation — route through approval chains, set TTL
Task 3 (hard): Access Review — audit entitlements, revoke risky grants
Each episode is procedurally generated from a seed, ensuring no two
episodes are identical while remaining deterministically gradable.
"""
SUPPORTS_CONCURRENT_SESSIONS: bool = True
# Class-level reference to the most recently active WorldState.
# Updated on every reset() and step() so /grader can read the real
# live state without framework hacks.
_active_world: ClassVar[Optional["WorldState"]] = None
def __init__(self):
super().__init__()
self._world = WorldState()
self._state = State(episode_id=str(uuid4()), step_count=0)
self._last_episode_score: Optional[float] = None
# ── OpenEnv API ───────────────────────────────────────────────────────────
def reset(
self,
seed: Optional[int] = None,
episode_id: Optional[str] = None,
**kwargs: Any,
) -> PrivilegeDeskObservation:
"""Reset to a new episode.
Kwargs:
task_id (str): "access_decision" | "jit_escalation" | "access_review"
difficulty_level (int): 1-3 (scales entity counts)
"""
task_id = kwargs.get("task_id", "access_decision")
difficulty_level = kwargs.get("difficulty_level", 1)
obs_dict = self._world.reset(
seed=seed,
task_id=task_id,
difficulty_level=difficulty_level,
)
self._state = State(
episode_id=episode_id or str(uuid4()),
step_count=0,
)
self._last_episode_score = None
# Update class-level reference so /grader can always find this world
PrivilegeDeskEnvironment._active_world = self._world
return self._build_observation(obs_dict, reward=0.0, done=False, tool_result=None)
def step(
self,
action: PrivilegeDeskAction,
timeout_s: Optional[float] = None,
**kwargs: Any,
) -> PrivilegeDeskObservation:
"""Execute one tool call."""
action_dict = {
"tool_name": action.tool_name,
"arguments": action.arguments,
}
obs_dict, step_reward, terminated, truncated, info = self._world.step(action_dict)
self._state.step_count = self._world.step_count
# Keep class reference current so /grader always sees the latest state
PrivilegeDeskEnvironment._active_world = self._world
done = terminated or truncated
if done:
self._last_episode_score = info.get("episode_score")
return self._build_observation(
obs_dict,
reward=step_reward,
done=done,
tool_result=info.get("tool_result"),
metadata={
"step": info.get("step"),
"terminated": terminated,
"truncated": truncated,
"step_reward": info.get("step_reward", 0.0),
"episode_reward": info.get("episode_reward", 0.0),
"episode_score": info.get("episode_score"),
},
)
@property
def state(self) -> State:
return self._state
def get_episode_score(self) -> Optional[float]:
"""Return the last episode's graded score."""
if self._last_episode_score is not None:
return self._last_episode_score
# If episode is still running, compute current score
score_dict = self._world.compute_episode_score()
return float(score_dict.get("score", 0.10))
def get_metadata(self):
from openenv.core.env_server.interfaces import EnvironmentMetadata
return EnvironmentMetadata(
name="PrivilegeDesk",
description=(
"Zero-Standing-Privilege Ops environment where an AI agent handles "
"enterprise access requests, JIT privilege escalation, approval routing, "
"and periodic access reviews with audit-based revocation."
),
version="1.0.0",
)
def close(self) -> None:
pass
# ── Helpers ───────────────────────────────────────────────────────────────
def _build_observation(
self, obs_dict: dict, reward: float, done: bool,
tool_result: Optional[dict], metadata: dict = None
) -> PrivilegeDeskObservation:
return PrivilegeDeskObservation(
# Core fields
task_id=obs_dict.get("task_id", ""),
task_goal=obs_dict.get("task_goal", ""),
step=obs_dict.get("step", 0),
max_steps=obs_dict.get("max_steps", 25),
current_time=obs_dict.get("current_time", ""),
available_tools=obs_dict.get("available_tools", []),
# Org
users=obs_dict.get("users", {}),
org_graph=obs_dict.get("org_graph", {}),
resources=obs_dict.get("resources", {}),
policies=obs_dict.get("policies", {}),
groups=obs_dict.get("groups", {}),
# Access state
entitlements=obs_dict.get("entitlements", {}),
pending_requests=obs_dict.get("pending_requests", {}),
approval_chains=obs_dict.get("approval_chains", {}),
workflows=obs_dict.get("workflows", {}),
# Objectives
objectives=obs_dict.get("objectives", []),
audit_log=obs_dict.get("audit_log", []),
notifications=obs_dict.get("notifications", []),
review_target_user_id=obs_dict.get("review_target_user_id"),
# OpenEnv base fields
done=done,
reward=reward,
tool_result=tool_result,
metadata=metadata or {},
)
|