Spaces:
Sleeping
Sleeping
File size: 15,697 Bytes
8c391c7 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 | from __future__ import annotations
from typing import Iterable
import pytest
from pydantic import ValidationError
from code_security_auditor_env.models import CodeSecurityAction
from code_security_auditor_env.server.security_environment import CodeSecurityAuditorEnvironment
def _action(**kwargs) -> CodeSecurityAction:
return CodeSecurityAction(**kwargs)
def _run_actions(task_id: str, actions: Iterable[CodeSecurityAction]) -> tuple[float, list[float]]:
env = CodeSecurityAuditorEnvironment(default_task_id=task_id)
obs = env.reset(task_id=task_id)
rewards: list[float] = [float(obs.reward or 0.0)]
for action in actions:
obs = env.step(action)
rewards.append(float(obs.reward or 0.0))
if obs.done:
break
if not obs.done:
obs = env.step(_action(action_type="submit_final_report"))
rewards.append(float(obs.reward or 0.0))
return float(obs.reward or 0.0), rewards
@pytest.mark.parametrize(
"task_id,expected_file_count",
[
("easy", 3),
("medium", 3),
("hard", 4),
],
)
def test_reset_exposes_task_specific_observation_space(task_id: str, expected_file_count: int) -> None:
env = CodeSecurityAuditorEnvironment(default_task_id=task_id)
obs = env.reset(task_id=task_id)
assert obs.task_id == task_id
assert len(obs.available_files) == expected_file_count
assert obs.steps_remaining > 0
assert obs.file_excerpt == ""
assert obs.focused_file is None
assert 0.0 <= float(obs.score_hint) <= 1.0
def test_action_space_validation_rejects_invalid_values() -> None:
with pytest.raises(ValidationError):
_action(action_type="not_valid")
with pytest.raises(ValidationError):
_action(action_type="submit_finding", confidence=1.5)
with pytest.raises(ValidationError):
_action(action_type="submit_finding", line_start=0)
def test_inspect_file_returns_numbered_excerpt() -> None:
env = CodeSecurityAuditorEnvironment(default_task_id="easy")
env.reset(task_id="easy")
obs = env.step(_action(action_type="inspect_file", filename="app/routes.py"))
assert obs.focused_file == "app/routes.py"
assert " 1:" in obs.file_excerpt
assert "SELECT id, email, role" in obs.file_excerpt
def test_partial_progress_reward_for_near_miss_finding() -> None:
env = CodeSecurityAuditorEnvironment(default_task_id="easy")
env.reset(task_id="easy")
obs = env.step(
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=11,
line_end=11,
vuln_type="sql_injection",
severity="high",
confidence=0.8,
evidence="nearby SQL line",
summary="line slightly off",
)
)
assert 0.0 < float(obs.reward or 0.0) <= 0.2
assert "Partial progress" in obs.last_feedback
def test_easy_task_high_quality_trajectory_scores_high() -> None:
actions = [
_action(action_type="inspect_file", filename="app/routes.py"),
_action(action_type="inspect_file", filename="app/config.py"),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=8,
vuln_type="sql_injection",
severity="high",
confidence=0.8,
evidence="f-string SQL query with request arg",
summary="SQL injection",
),
_action(
action_type="submit_finding",
filename="app/config.py",
line_start=5,
vuln_type="hardcoded_secret",
severity="high",
confidence=0.85,
evidence="secret embedded in config",
summary="hardcoded secret",
),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=15,
vuln_type="weak_authentication",
severity="medium",
confidence=0.65,
evidence="static token auth bypass",
summary="weak authentication",
),
_action(action_type="submit_final_report"),
]
score, rewards = _run_actions("easy", actions)
assert score >= 0.75
assert all(0.0 <= r <= 1.0 for r in rewards)
def test_reward_hacking_by_spam_and_duplicates_is_penalized() -> None:
strong_actions = [
_action(action_type="inspect_file", filename="app/routes.py"),
_action(action_type="inspect_file", filename="app/config.py"),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=8,
vuln_type="sql_injection",
severity="high",
confidence=0.8,
evidence="sql injection",
summary="sql injection",
),
_action(
action_type="submit_finding",
filename="app/config.py",
line_start=5,
vuln_type="hardcoded_secret",
severity="high",
confidence=0.85,
evidence="hardcoded secret",
summary="hardcoded secret",
),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=15,
vuln_type="weak_authentication",
severity="medium",
confidence=0.65,
evidence="static token",
summary="weak auth",
),
_action(action_type="submit_final_report"),
]
spam_actions = [
_action(action_type="inspect_file", filename="app/routes.py"),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=8,
vuln_type="sql_injection",
severity="high",
confidence=0.8,
evidence="sql injection",
summary="sql injection",
),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=8,
vuln_type="sql_injection",
severity="high",
confidence=0.95,
evidence="duplicate #1",
summary="duplicate #1",
),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=8,
vuln_type="sql_injection",
severity="high",
confidence=0.99,
evidence="duplicate #2",
summary="duplicate #2",
),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=2,
vuln_type="xss",
severity="critical",
confidence=1.0,
evidence="intentional false positive",
summary="intentional false positive",
),
_action(action_type="submit_final_report"),
]
strong_score, _ = _run_actions("easy", strong_actions)
spam_score, _ = _run_actions("easy", spam_actions)
assert strong_score > spam_score
assert spam_score < 0.6
def test_medium_and_hard_tasks_support_successful_completion() -> None:
medium_actions = [
_action(action_type="inspect_file", filename="service/webhook.py"),
_action(action_type="inspect_file", filename="service/export.py"),
_action(action_type="inspect_file", filename="service/serializers.py"),
_action(
action_type="submit_finding",
filename="service/webhook.py",
line_start=10,
vuln_type="weak_authentication",
severity="medium",
confidence=0.65,
evidence="timing unsafe compare",
summary="signature compare",
),
_action(
action_type="submit_finding",
filename="service/webhook.py",
line_start=22,
vuln_type="weak_authentication",
severity="high",
confidence=0.8,
evidence="debug bypass",
summary="debug bypass",
),
_action(
action_type="submit_finding",
filename="service/export.py",
line_start=8,
vuln_type="command_injection",
severity="critical",
confidence=0.92,
evidence="os.system with user input",
summary="command injection",
),
_action(
action_type="submit_finding",
filename="service/serializers.py",
line_start=4,
vuln_type="insecure_deserialization",
severity="high",
confidence=0.83,
evidence="yaml.Loader unsafe",
summary="unsafe yaml load",
),
_action(action_type="submit_final_report"),
]
hard_actions = [
_action(action_type="inspect_file", filename="api/auth.py"),
_action(action_type="inspect_file", filename="api/files.py"),
_action(action_type="inspect_file", filename="api/fetcher.py"),
_action(action_type="inspect_file", filename="api/storage.py"),
_action(
action_type="submit_finding",
filename="api/auth.py",
line_start=12,
vuln_type="weak_authentication",
severity="critical",
confidence=0.9,
evidence="alg=none token acceptance",
summary="jwt none alg",
),
_action(
action_type="submit_finding",
filename="api/files.py",
line_start=11,
vuln_type="weak_authentication",
severity="high",
confidence=0.8,
evidence="tenant param controls authorization",
summary="idor cross tenant",
),
_action(
action_type="submit_finding",
filename="api/fetcher.py",
line_start=4,
vuln_type="ssrf",
severity="high",
confidence=0.8,
evidence="requests.get arbitrary URL",
summary="ssrf",
),
_action(
action_type="submit_finding",
filename="api/storage.py",
line_start=6,
vuln_type="path_traversal",
severity="critical",
confidence=0.9,
evidence="path join without normalization",
summary="path traversal",
),
_action(action_type="submit_final_report"),
]
medium_score, medium_rewards = _run_actions("medium", medium_actions)
hard_score, hard_rewards = _run_actions("hard", hard_actions)
assert medium_score >= 0.7
assert hard_score >= 0.7
assert all(0.0 <= r <= 1.0 for r in medium_rewards)
assert all(0.0 <= r <= 1.0 for r in hard_rewards)
def test_confidence_miscalibration_reduces_partial_progress_rewards() -> None:
# Use line offsets that produce partial (not confirmed) matches so confidence
# calibration impacts component score and therefore shaped reward.
overconfident_actions = [
_action(action_type="inspect_file", filename="app/routes.py"),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=13,
vuln_type="sql_injection",
severity="high",
confidence=1.0,
evidence="near miss with inflated confidence #1",
summary="near miss #1",
),
_action(
action_type="submit_finding",
filename="app/config.py",
line_start=1,
vuln_type="hardcoded_secret",
severity="high",
confidence=1.0,
evidence="near miss with inflated confidence #2",
summary="near miss #2",
),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=20,
vuln_type="weak_authentication",
severity="medium",
confidence=1.0,
evidence="near miss with inflated confidence #3",
summary="near miss #3",
),
_action(action_type="submit_final_report"),
]
calibrated_actions = [
_action(action_type="inspect_file", filename="app/routes.py"),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=13,
vuln_type="sql_injection",
severity="high",
confidence=0.8,
evidence="near miss with calibrated confidence #1",
summary="near miss #1",
),
_action(
action_type="submit_finding",
filename="app/config.py",
line_start=1,
vuln_type="hardcoded_secret",
severity="high",
confidence=0.8,
evidence="near miss with calibrated confidence #2",
summary="near miss #2",
),
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=20,
vuln_type="weak_authentication",
severity="medium",
confidence=0.65,
evidence="near miss with calibrated confidence #3",
summary="near miss #3",
),
_action(action_type="submit_final_report"),
]
overconf_score, overconf_rewards = _run_actions("easy", overconfident_actions)
calibrated_score, calibrated_rewards = _run_actions("easy", calibrated_actions)
assert sum(calibrated_rewards) > sum(overconf_rewards)
assert calibrated_score >= overconf_score
def test_step_limit_stalling_strategy_auto_finalizes_with_low_score() -> None:
env = CodeSecurityAuditorEnvironment(default_task_id="easy")
obs = env.reset(task_id="easy")
# Repeatedly inspect the same non-critical pattern to simulate stalling.
while not obs.done:
obs = env.step(_action(action_type="inspect_file", filename="app/db.py"))
assert obs.done is True
assert 0.0 <= float(obs.reward or 0.0) <= 1.0
assert float(obs.reward or 0.0) < 0.5
assert "Max steps reached" in obs.last_feedback
def test_repeated_duplicate_confirmed_findings_reduce_quality_multiplier() -> None:
env = CodeSecurityAuditorEnvironment(default_task_id="easy")
env.reset(task_id="easy")
first = env.step(
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=8,
vuln_type="sql_injection",
severity="high",
confidence=0.8,
evidence="correct first finding",
summary="correct first finding",
)
)
qm_after_first = float(first.metadata["quality_multiplier"])
second = env.step(
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=8,
vuln_type="sql_injection",
severity="high",
confidence=0.95,
evidence="duplicate second",
summary="duplicate second",
)
)
qm_after_second = float(second.metadata["quality_multiplier"])
third = env.step(
_action(
action_type="submit_finding",
filename="app/routes.py",
line_start=8,
vuln_type="sql_injection",
severity="high",
confidence=1.0,
evidence="duplicate third",
summary="duplicate third",
)
)
qm_after_third = float(third.metadata["quality_multiplier"])
assert qm_after_second < qm_after_first
assert qm_after_third < qm_after_second
assert int(third.metadata["duplicate_submission_count"]) >= 2
|