Spaces:
Runtime error
Runtime error
File size: 10,389 Bytes
8c486a8 7fedc25 8c486a8 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 | """End-to-end scripted demo of OpenRange episode lifecycle.
Demonstrates:
1. Environment reset with a snapshot
2. Red agent: recon -> exploit -> flag capture
3. Blue agent: log analysis -> detection -> patching
4. Reward computation for both roles
5. Trajectory logging for SFT export
Runs without Docker (uses mock mode) and without any LLM.
Usage::
python -m examples.demo
# or
uv run python examples/demo.py
"""
from __future__ import annotations
import sys
from pathlib import Path
from typing import Any
from open_range.protocols import (
FlagSpec,
GoldenPathStep,
SnapshotSpec,
TaskSpec,
TruthGraph,
Vulnerability,
)
from open_range.models import RangeAction, RangeObservation
from open_range.server.environment import RangeEnvironment
from open_range.training.trajectory import TrajectoryLogger
# ---------------------------------------------------------------------------
# Scripted agents
# ---------------------------------------------------------------------------
class ScriptedAgent:
"""Agent that follows a fixed command list. No LLM needed."""
def __init__(self, script: list[str], role: str = "red") -> None:
self.script = list(script)
self.role = role
self._step_idx = 0
def reset(self, briefing: str, role: str) -> None:
self.role = role
self._step_idx = 0
def act(self, observation: str) -> str:
if self._step_idx < len(self.script):
cmd = self.script[self._step_idx]
self._step_idx += 1
return cmd
return "echo done"
@property
def exhausted(self) -> bool:
return self._step_idx >= len(self.script)
# ---------------------------------------------------------------------------
# Demo snapshot (matches tier1_basic topology)
# ---------------------------------------------------------------------------
def make_demo_snapshot() -> SnapshotSpec:
"""Build a demo snapshot with a SQL injection vulnerability."""
return SnapshotSpec(
topology={
"hosts": [
"attacker", "firewall", "web", "mail",
"db", "files", "ldap", "siem",
],
"zones": {
"external": ["attacker"],
"dmz": ["web", "mail"],
"internal": ["db", "files"],
"management": ["ldap", "siem"],
},
"tier": 1,
},
truth_graph=TruthGraph(
vulns=[
Vulnerability(
id="vuln-sqli-1",
type="sqli",
host="web",
service="php",
injection_point="/search?q=",
vulnerable_code=(
"$conn->query(\"SELECT * FROM products "
"WHERE name LIKE '%$_GET[q]%'\");"
),
root_cause="Unsanitized input in SQL query",
blast_radius="Full database read access",
remediation="Use parameterized queries",
),
],
),
flags=[
FlagSpec(
id="flag-1",
value="FLAG{test_sqli_123}",
path="/var/flags/flag1.txt",
host="db",
),
],
golden_path=[
GoldenPathStep(step=1, command="nmap -sV 10.0.1.0/24",
expect_in_stdout="80/tcp open http",
description="Port scan reveals web server"),
GoldenPathStep(step=2, command="curl http://web/",
expect_in_stdout="Welcome",
description="Enumerate web application"),
GoldenPathStep(step=3, command="curl 'http://web/search?q=test'",
expect_in_stdout="products",
description="Test search functionality"),
GoldenPathStep(step=4,
command="curl 'http://web/search?q=test' UNION SELECT flag FROM flags--'",
expect_in_stdout="FLAG{test_sqli_123}",
description="Exploit SQL injection"),
GoldenPathStep(step=5, command="submit_flag FLAG{test_sqli_123}",
expect_in_stdout="Correct",
description="Submit captured flag"),
],
task=TaskSpec(
red_briefing=(
"Target network has a web application with database backend. "
"Find vulnerabilities and capture the flag."
),
blue_briefing=(
"Monitor SIEM for suspicious activity. "
"Investigate alerts, identify attacks, and apply patches."
),
),
)
# ---------------------------------------------------------------------------
# Episode runner
# ---------------------------------------------------------------------------
def run_demo(
env: RangeEnvironment | None = None,
max_steps: int = 20,
quiet: bool = False,
) -> dict[str, Any]:
"""Run a complete demo episode with scripted Red and Blue agents.
Args:
env: Optional pre-configured environment. If None, creates one
in mock mode (no Docker required).
max_steps: Maximum steps before forced termination.
quiet: If True, suppress printed output.
Returns:
Dict with episode results and trajectory logger.
"""
if env is None:
env = RangeEnvironment(docker_available=False, max_steps=max_steps)
snapshot = make_demo_snapshot()
# Build scripted agents
red = ScriptedAgent(
script=[
"nmap -sV 10.0.1.0/24",
"curl http://web/",
"curl 'http://web/search?q=test'",
"curl 'http://web/search?q=test%27+UNION+SELECT+flag+FROM+flags--'",
"submit_flag FLAG{test_sqli_123}",
],
role="red",
)
blue = ScriptedAgent(
script=[
"tail -n 50 /var/log/siem/web_access.log",
"grep -i 'union' /var/log/siem/web_access.log",
"submit_finding SQLi attack detected from attacker targeting /search endpoint",
"patch web /var/www/html/search.php",
],
role="blue",
)
# Trajectory logger
traj = TrajectoryLogger()
def _print(msg: str) -> None:
if not quiet:
print(msg)
# --- Episode start ---
_print("=" * 60)
_print(" OPENRANGE DEMO -- Scripted Red vs Blue Episode")
_print("=" * 60)
obs = env.reset(snapshot=snapshot, episode_id="demo-001")
traj.start_episode(
episode_id="demo-001",
snapshot_id="tier1-sqli-demo",
tier=1,
)
red.reset(briefing=obs.stdout, role="red")
blue.reset(briefing=obs.stdout, role="blue")
_print(f"\n{obs.stdout}\n")
_print("-" * 60)
step = 0
last_obs = obs
while not last_obs.done and step < max_steps:
# Red's turn
if red.exhausted:
break
red_cmd = red.act(last_obs.stdout)
_print(f"\n[Step {step + 1}] RED >> {red_cmd}")
red_obs = env.step(RangeAction(command=red_cmd, mode="red"))
reward = red_obs.reward if red_obs.reward is not None else 0.0
traj.log_turn(
role="red",
observation=last_obs.stdout,
action=red_cmd,
reward=float(reward),
)
_print(f" stdout: {red_obs.stdout[:120]}{'...' if len(red_obs.stdout) > 120 else ''}")
if red_obs.flags_captured:
_print(f" FLAGS CAPTURED: {red_obs.flags_captured}")
_print(f" reward: {reward:.4f} done: {red_obs.done}")
last_obs = red_obs
step += 1
if last_obs.done:
break
# Blue's turn
if blue.exhausted:
continue
blue_cmd = blue.act(last_obs.stdout)
_print(f"\n[Step {step + 1}] BLUE >> {blue_cmd}")
blue_obs = env.step(RangeAction(command=blue_cmd, mode="blue"))
reward = blue_obs.reward if blue_obs.reward is not None else 0.0
traj.log_turn(
role="blue",
observation=last_obs.stdout,
action=blue_cmd,
reward=float(reward),
)
_print(f" stdout: {blue_obs.stdout[:120]}{'...' if len(blue_obs.stdout) > 120 else ''}")
if blue_obs.alerts:
_print(f" alerts: {blue_obs.alerts}")
_print(f" reward: {reward:.4f} done: {blue_obs.done}")
last_obs = blue_obs
step += 1
# Determine outcome
if env.state.flags_found:
outcome = "flag_captured"
elif step >= max_steps:
outcome = "timeout"
else:
outcome = "completed"
episode = traj.end_episode(
outcome=outcome,
metrics={
"steps": step,
"flags_found": list(env.state.flags_found),
"tier": env.state.tier,
},
)
# --- Summary ---
_print("\n" + "=" * 60)
_print(" EPISODE SUMMARY")
_print("=" * 60)
_print(f" Outcome: {outcome}")
_print(f" Steps: {step}")
_print(f" Flags found: {env.state.flags_found}")
_print(f" Red total reward: {episode.total_red_reward:.4f}")
_print(f" Blue total reward:{episode.total_blue_reward:.4f}")
_print(f" Red turns: {len(episode.red_turns)}")
_print(f" Blue turns: {len(episode.blue_turns)}")
_print("=" * 60)
return {
"outcome": outcome,
"steps": step,
"flags_found": list(env.state.flags_found),
"episode": episode,
"trajectory_logger": traj,
"env": env,
}
# ---------------------------------------------------------------------------
# Main
# ---------------------------------------------------------------------------
def main() -> None:
"""Run the demo and optionally export trajectories."""
result = run_demo()
# Export trajectories to JSONL
traj: TrajectoryLogger = result["trajectory_logger"]
out_path = Path("demo_trajectories.jsonl")
count = traj.export_jsonl(out_path, reward_threshold=0.0)
print(f"\nExported {count} trajectory records to {out_path}")
if __name__ == "__main__":
main()
|