@tangle-network/agent-bench 0.1.0 → 0.3.6

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (317) hide show
  1. package/CHANGELOG.md +23 -0
  2. package/HARNESS.md +302 -0
  3. package/README.md +39 -2
  4. package/dist/adapters.d.ts +15 -0
  5. package/dist/adapters.js +43 -0
  6. package/dist/adapters.js.map +1 -0
  7. package/dist/benchmarks/_harness.d.ts +125 -0
  8. package/dist/benchmarks/_harness.js +33 -0
  9. package/dist/benchmarks/_harness.js.map +1 -0
  10. package/dist/benchmarks/aec-bench.d.ts +27 -0
  11. package/dist/benchmarks/aec-bench.js +8 -0
  12. package/dist/benchmarks/aec-bench.js.map +1 -0
  13. package/dist/benchmarks/agentbench.d.ts +16 -0
  14. package/dist/benchmarks/agentbench.js +10 -0
  15. package/dist/benchmarks/agentbench.js.map +1 -0
  16. package/dist/benchmarks/appworld.d.ts +37 -0
  17. package/dist/benchmarks/appworld.js +14 -0
  18. package/dist/benchmarks/appworld.js.map +1 -0
  19. package/dist/benchmarks/bfcl.d.ts +18 -0
  20. package/dist/benchmarks/bfcl.js +10 -0
  21. package/dist/benchmarks/bfcl.js.map +1 -0
  22. package/dist/benchmarks/cad-design.d.ts +45 -0
  23. package/dist/benchmarks/cad-design.js +7 -0
  24. package/dist/benchmarks/cad-design.js.map +1 -0
  25. package/dist/benchmarks/cadbench.d.ts +19 -0
  26. package/dist/benchmarks/cadbench.js +8 -0
  27. package/dist/benchmarks/cadbench.js.map +1 -0
  28. package/dist/benchmarks/cadgenbench.d.ts +22 -0
  29. package/dist/benchmarks/cadgenbench.js +8 -0
  30. package/dist/benchmarks/cadgenbench.js.map +1 -0
  31. package/dist/benchmarks/commit0.d.ts +31 -0
  32. package/dist/benchmarks/commit0.js +10 -0
  33. package/dist/benchmarks/commit0.js.map +1 -0
  34. package/dist/benchmarks/crag.d.ts +14 -0
  35. package/dist/benchmarks/crag.js +9 -0
  36. package/dist/benchmarks/crag.js.map +1 -0
  37. package/dist/benchmarks/dabstep.d.ts +18 -0
  38. package/dist/benchmarks/dabstep.js +10 -0
  39. package/dist/benchmarks/dabstep.js.map +1 -0
  40. package/dist/benchmarks/enterpriseops-gym.d.ts +38 -0
  41. package/dist/benchmarks/enterpriseops-gym.js +10 -0
  42. package/dist/benchmarks/enterpriseops-gym.js.map +1 -0
  43. package/dist/benchmarks/finresearchbench.d.ts +15 -0
  44. package/dist/benchmarks/finresearchbench.js +8 -0
  45. package/dist/benchmarks/finresearchbench.js.map +1 -0
  46. package/dist/benchmarks/finsearchcomp.d.ts +49 -0
  47. package/dist/benchmarks/finsearchcomp.js +7 -0
  48. package/dist/benchmarks/finsearchcomp.js.map +1 -0
  49. package/dist/benchmarks/frames.d.ts +59 -0
  50. package/dist/benchmarks/frames.js +13 -0
  51. package/dist/benchmarks/frames.js.map +1 -0
  52. package/dist/benchmarks/hotpotqa.d.ts +48 -0
  53. package/dist/benchmarks/hotpotqa.js +15 -0
  54. package/dist/benchmarks/hotpotqa.js.map +1 -0
  55. package/dist/benchmarks/humaneval.d.ts +53 -0
  56. package/dist/benchmarks/humaneval.js +15 -0
  57. package/dist/benchmarks/humaneval.js.map +1 -0
  58. package/dist/benchmarks/mind2web.d.ts +41 -0
  59. package/dist/benchmarks/mind2web.js +9 -0
  60. package/dist/benchmarks/mind2web.js.map +1 -0
  61. package/dist/benchmarks/nomiracl.d.ts +15 -0
  62. package/dist/benchmarks/nomiracl.js +9 -0
  63. package/dist/benchmarks/nomiracl.js.map +1 -0
  64. package/dist/benchmarks/open-rag-bench.d.ts +14 -0
  65. package/dist/benchmarks/open-rag-bench.js +9 -0
  66. package/dist/benchmarks/open-rag-bench.js.map +1 -0
  67. package/dist/benchmarks/programbench.d.ts +38 -0
  68. package/dist/benchmarks/programbench.js +10 -0
  69. package/dist/benchmarks/programbench.js.map +1 -0
  70. package/dist/benchmarks/rag-shared.d.ts +42 -0
  71. package/dist/benchmarks/rag-shared.js +39 -0
  72. package/dist/benchmarks/rag-shared.js.map +1 -0
  73. package/dist/benchmarks/ragbench.d.ts +16 -0
  74. package/dist/benchmarks/ragbench.js +9 -0
  75. package/dist/benchmarks/ragbench.js.map +1 -0
  76. package/dist/benchmarks/simpleqa.d.ts +64 -0
  77. package/dist/benchmarks/simpleqa.js +11 -0
  78. package/dist/benchmarks/simpleqa.js.map +1 -0
  79. package/dist/benchmarks/swe-bench.d.ts +56 -0
  80. package/dist/benchmarks/swe-bench.js +14 -0
  81. package/dist/benchmarks/swe-bench.js.map +1 -0
  82. package/dist/benchmarks/t2-ragbench.d.ts +14 -0
  83. package/dist/benchmarks/t2-ragbench.js +9 -0
  84. package/dist/benchmarks/t2-ragbench.js.map +1 -0
  85. package/dist/benchmarks/tau-bench-shared.d.ts +26 -0
  86. package/dist/benchmarks/tau-bench-shared.js +10 -0
  87. package/dist/benchmarks/tau-bench-shared.js.map +1 -0
  88. package/dist/benchmarks/tau2-bench.d.ts +7 -0
  89. package/dist/benchmarks/tau2-bench.js +11 -0
  90. package/dist/benchmarks/tau2-bench.js.map +1 -0
  91. package/dist/benchmarks/tau3-banking.d.ts +15 -0
  92. package/dist/benchmarks/tau3-banking.js +9 -0
  93. package/dist/benchmarks/tau3-banking.js.map +1 -0
  94. package/dist/benchmarks/terminal-bench.d.ts +24 -0
  95. package/dist/benchmarks/terminal-bench.js +8 -0
  96. package/dist/benchmarks/terminal-bench.js.map +1 -0
  97. package/dist/benchmarks/toollm.d.ts +16 -0
  98. package/dist/benchmarks/toollm.js +10 -0
  99. package/dist/benchmarks/toollm.js.map +1 -0
  100. package/dist/benchmarks/trata-hedge.d.ts +32 -0
  101. package/dist/benchmarks/trata-hedge.js +7 -0
  102. package/dist/benchmarks/trata-hedge.js.map +1 -0
  103. package/dist/benchmarks/types.d.ts +107 -0
  104. package/dist/benchmarks/types.js +1 -0
  105. package/dist/benchmarks/types.js.map +1 -0
  106. package/dist/benchmarks/webarena-verified.d.ts +16 -0
  107. package/dist/benchmarks/webarena-verified.js +10 -0
  108. package/dist/benchmarks/webarena-verified.js.map +1 -0
  109. package/dist/chunk-2PVVP7GN.js +197 -0
  110. package/dist/chunk-2PVVP7GN.js.map +1 -0
  111. package/dist/chunk-2XU6OGEN.js +170 -0
  112. package/dist/chunk-2XU6OGEN.js.map +1 -0
  113. package/dist/chunk-53UPUNBZ.js +325 -0
  114. package/dist/chunk-53UPUNBZ.js.map +1 -0
  115. package/dist/chunk-5SBJCB6W.js +144 -0
  116. package/dist/chunk-5SBJCB6W.js.map +1 -0
  117. package/dist/chunk-7WSD27QQ.js +118 -0
  118. package/dist/chunk-7WSD27QQ.js.map +1 -0
  119. package/dist/chunk-C7T7WEK2.js +103 -0
  120. package/dist/chunk-C7T7WEK2.js.map +1 -0
  121. package/dist/chunk-CKUVRZ2T.js +251 -0
  122. package/dist/chunk-CKUVRZ2T.js.map +1 -0
  123. package/dist/chunk-HBSWHQNJ.js +30 -0
  124. package/dist/chunk-HBSWHQNJ.js.map +1 -0
  125. package/dist/chunk-HHXFIHXC.js +116 -0
  126. package/dist/chunk-HHXFIHXC.js.map +1 -0
  127. package/dist/chunk-IFAV6KEM.js +276 -0
  128. package/dist/chunk-IFAV6KEM.js.map +1 -0
  129. package/dist/chunk-INNOYXCP.js +387 -0
  130. package/dist/chunk-INNOYXCP.js.map +1 -0
  131. package/dist/chunk-J3KDJNX2.js +182 -0
  132. package/dist/chunk-J3KDJNX2.js.map +1 -0
  133. package/dist/chunk-JRWWGMK7.js +148 -0
  134. package/dist/chunk-JRWWGMK7.js.map +1 -0
  135. package/dist/chunk-JTHWEDEW.js +32 -0
  136. package/dist/chunk-JTHWEDEW.js.map +1 -0
  137. package/dist/chunk-KDIKRJGB.js +120 -0
  138. package/dist/chunk-KDIKRJGB.js.map +1 -0
  139. package/dist/chunk-LRRD7NAG.js +301 -0
  140. package/dist/chunk-LRRD7NAG.js.map +1 -0
  141. package/dist/chunk-ODT47UAY.js +221 -0
  142. package/dist/chunk-ODT47UAY.js.map +1 -0
  143. package/dist/chunk-PA2ZKHJC.js +230 -0
  144. package/dist/chunk-PA2ZKHJC.js.map +1 -0
  145. package/dist/chunk-PPYSEKFM.js +182 -0
  146. package/dist/chunk-PPYSEKFM.js.map +1 -0
  147. package/dist/chunk-PUIRNYI7.js +189 -0
  148. package/dist/chunk-PUIRNYI7.js.map +1 -0
  149. package/dist/chunk-R36V2VP7.js +169 -0
  150. package/dist/chunk-R36V2VP7.js.map +1 -0
  151. package/dist/chunk-R67DFVLO.js +142 -0
  152. package/dist/chunk-R67DFVLO.js.map +1 -0
  153. package/dist/chunk-SEVJPLZC.js +260 -0
  154. package/dist/chunk-SEVJPLZC.js.map +1 -0
  155. package/dist/chunk-SYDW647C.js +318 -0
  156. package/dist/chunk-SYDW647C.js.map +1 -0
  157. package/dist/chunk-TBKU5XQI.js +228 -0
  158. package/dist/chunk-TBKU5XQI.js.map +1 -0
  159. package/dist/chunk-TSWPNOYM.js +147 -0
  160. package/dist/chunk-TSWPNOYM.js.map +1 -0
  161. package/dist/chunk-UAIOHCUK.js +27 -0
  162. package/dist/chunk-UAIOHCUK.js.map +1 -0
  163. package/dist/chunk-UPAMRDX4.js +233 -0
  164. package/dist/chunk-UPAMRDX4.js.map +1 -0
  165. package/dist/chunk-VQRS7VUC.js +342 -0
  166. package/dist/chunk-VQRS7VUC.js.map +1 -0
  167. package/dist/chunk-X3BTXCJ4.js +262 -0
  168. package/dist/chunk-X3BTXCJ4.js.map +1 -0
  169. package/dist/chunk-X5YKXC6V.js +211 -0
  170. package/dist/chunk-X5YKXC6V.js.map +1 -0
  171. package/dist/chunk-Y6O2OCUO.js +130 -0
  172. package/dist/chunk-Y6O2OCUO.js.map +1 -0
  173. package/dist/chunk-YCGY7UIZ.js +208 -0
  174. package/dist/chunk-YCGY7UIZ.js.map +1 -0
  175. package/dist/chunk-Z7ML6L77.js +162 -0
  176. package/dist/chunk-Z7ML6L77.js.map +1 -0
  177. package/dist/chunk-ZEWMTR5M.js +136 -0
  178. package/dist/chunk-ZEWMTR5M.js.map +1 -0
  179. package/dist/index.d.ts +355 -0
  180. package/dist/index.js +1908 -0
  181. package/dist/index.js.map +1 -0
  182. package/fixtures/aec-bench.json +18 -0
  183. package/fixtures/agentbench-dbbench.json +22 -0
  184. package/fixtures/bfcl.json +45 -0
  185. package/fixtures/commit0.json +72 -0
  186. package/fixtures/crag.json +10 -0
  187. package/fixtures/dabstep.json +22 -0
  188. package/fixtures/enterpriseops-gym.json +103 -0
  189. package/fixtures/finresearchbench.json +21 -0
  190. package/fixtures/finsearchcomp.json +66 -0
  191. package/fixtures/frames.json +26 -0
  192. package/fixtures/hotpotqa.json +182 -0
  193. package/fixtures/nomiracl.json +26 -0
  194. package/fixtures/open-rag-bench.json +16 -0
  195. package/fixtures/pier-agent/no-model-task/environment/Dockerfile +16 -0
  196. package/fixtures/pier-agent/no-model-task/environment/seed/src/status.txt +1 -0
  197. package/fixtures/pier-agent/no-model-task/instruction.md +6 -0
  198. package/fixtures/pier-agent/no-model-task/pre_artifacts.sh +6 -0
  199. package/fixtures/pier-agent/no-model-task/task.toml +35 -0
  200. package/fixtures/pier-agent/no-model-task/tests/Dockerfile +17 -0
  201. package/fixtures/pier-agent/no-model-task/tests/seed/src/status.txt +1 -0
  202. package/fixtures/pier-agent/no-model-task/tests/test.sh +19 -0
  203. package/fixtures/programbench.json +17 -0
  204. package/fixtures/ragbench.json +21 -0
  205. package/fixtures/simpleqa.json +121 -0
  206. package/fixtures/t2-ragbench.json +13 -0
  207. package/fixtures/tau2-bench.json +16 -0
  208. package/fixtures/tau3-banking.json +16 -0
  209. package/fixtures/toollm.json +28 -0
  210. package/fixtures/webarena-verified.json +20 -0
  211. package/package.json +60 -20
  212. package/pier_agents/__init__.py +18 -0
  213. package/pier_agents/candidate_contract.py +755 -0
  214. package/pier_agents/process_boundary.py +321 -0
  215. package/pier_agents/tangle_candidate.py +907 -0
  216. package/pier_agents/workspace_boundary.py +368 -0
  217. package/scripts/appworld_driver.py +359 -0
  218. package/scripts/cadbench_prepare.py +22 -0
  219. package/scripts/cadgenbench_hard_parts.py +48 -0
  220. package/scripts/clbench_codebase_judge.py +73 -0
  221. package/scripts/commit0_judge.py +170 -0
  222. package/scripts/dabstep_judge.py +42 -0
  223. package/scripts/enterpriseops_gym_judge.py +281 -0
  224. package/scripts/programbench_judge.py +120 -0
  225. package/scripts/render-gate-chart.mjs +176 -0
  226. package/scripts/run-package-tests.mjs +56 -0
  227. package/scripts/terminate-pier-trial.mts +66 -0
  228. package/scripts/trata-hedge/README.md +56 -0
  229. package/scripts/trata-hedge/run.sh +60 -0
  230. package/scripts/trata-hedge/solve.py +83 -0
  231. package/scripts/verify-packed-consumer.mjs +235 -0
  232. package/scripts/verify-pier-agent.mts +715 -0
  233. package/scripts/verify-pier-pair.mts +74 -0
  234. package/scripts/verify-pier-recovery.mts +139 -0
  235. package/src/adapters.ts +26 -0
  236. package/src/benchmarks/_harness.test.mts +178 -0
  237. package/src/benchmarks/_harness.ts +239 -16
  238. package/src/benchmarks/agentbench.ts +163 -0
  239. package/src/benchmarks/appworld.test.mts +15 -9
  240. package/src/benchmarks/bfcl.ts +346 -0
  241. package/src/benchmarks/crag.ts +137 -0
  242. package/src/benchmarks/dabstep.test.mts +70 -0
  243. package/src/benchmarks/dabstep.ts +212 -0
  244. package/src/benchmarks/external-adapters.test.mts +150 -0
  245. package/src/benchmarks/finresearchbench.ts +269 -0
  246. package/src/benchmarks/humaneval.ts +20 -8
  247. package/src/benchmarks/nomiracl.ts +180 -0
  248. package/src/benchmarks/open-rag-bench.ts +153 -0
  249. package/src/benchmarks/rag-benchmarks.test.mts +138 -0
  250. package/src/benchmarks/rag-shared.ts +327 -0
  251. package/src/benchmarks/ragbench.ts +171 -0
  252. package/src/benchmarks/swe-bench.test.mts +61 -0
  253. package/src/benchmarks/swe-bench.ts +201 -19
  254. package/src/benchmarks/t2-ragbench.ts +166 -0
  255. package/src/benchmarks/tau-bench-shared.ts +214 -0
  256. package/src/benchmarks/tau2-bench.ts +30 -0
  257. package/src/benchmarks/tau3-banking.ts +29 -0
  258. package/src/benchmarks/terminal-bench.test.mts +33 -0
  259. package/src/benchmarks/terminal-bench.ts +23 -8
  260. package/src/benchmarks/toollm.ts +254 -0
  261. package/src/benchmarks/types.ts +42 -0
  262. package/src/benchmarks/webarena-verified.ts +200 -0
  263. package/src/commit0-prereqs.sh +0 -0
  264. package/src/coordination-mcp-container-reach.mts +181 -0
  265. package/src/decoder-live.mts +1 -1
  266. package/src/examples/README.md +103 -39
  267. package/src/examples/benchmark-matrix.mts +101 -0
  268. package/src/examples/lean-proof-gate.README.md +77 -0
  269. package/src/examples/lean-proof-gate.mts +162 -0
  270. package/src/examples/lean-verify.ts +95 -0
  271. package/src/examples/lean.Dockerfile +12 -0
  272. package/src/examples/math-demo.mts +9 -7
  273. package/src/examples/strategy-demo.mts +10 -12
  274. package/src/gate.ts +3 -2
  275. package/src/hev-eval.mts +69 -0
  276. package/src/hev-improve.mts +169 -0
  277. package/src/hev-structural.mts +688 -0
  278. package/src/index.ts +73 -0
  279. package/src/mbpp-structural.mts +662 -0
  280. package/src/pier-agent.test-fixtures.mts +19 -0
  281. package/src/pier-agent.test.mts +363 -0
  282. package/src/pier-agent.ts +657 -0
  283. package/src/pier-result-grader.mjs +30 -0
  284. package/src/pier-result-grader.test.mts +62 -0
  285. package/src/pier-result-grader.ts +108 -0
  286. package/src/pier-task-outcome.test.mts +117 -0
  287. package/src/pier-task-outcome.ts +240 -0
  288. package/src/pier-trial-controller.test.mts +412 -0
  289. package/src/pier-trial-controller.ts +858 -0
  290. package/src/pier-trial-supervisor.mjs +352 -0
  291. package/src/resolve-client.ts +25 -2
  292. package/src/run-benchmarks-cli.mts +72 -0
  293. package/src/run-benchmarks-report.ts +66 -0
  294. package/src/run-benchmarks.test.mts +231 -0
  295. package/src/run-benchmarks.ts +589 -0
  296. package/src/smoke-structural-rollout.mts +393 -0
  297. package/src/swe-bench-env.test.ts +207 -0
  298. package/src/swe-bench-env.ts +554 -0
  299. package/src/swe-jail.ts +293 -0
  300. package/src/swe-self-improve.mts +84 -0
  301. package/src/swe-structural-judge-policy.test.ts +117 -0
  302. package/src/swe-structural-judge-policy.ts +133 -0
  303. package/src/swe-structural-policy.test.ts +124 -0
  304. package/src/swe-structural-policy.ts +132 -0
  305. package/src/swe-structural-provenance.test.ts +93 -0
  306. package/src/swe-structural-provenance.ts +138 -0
  307. package/src/swe-structural.mts +1260 -0
  308. package/src/swe-temp.ts +14 -0
  309. package/src/tb-container-executor.mts +234 -0
  310. package/src/tb-container-executor.test.mts +99 -0
  311. package/src/tb-supervisor-sidecar.mts +222 -0
  312. package/src/trata-gepa.mts +1 -1
  313. package/steerers/eops-itsm-population.json +1 -0
  314. package/tb_agents/opencode_refine_agent.py +117 -0
  315. package/tb_agents/opencode_router_agent.py +406 -0
  316. package/tb_agents/opencode_supervisor_agent.py +239 -0
  317. package/tb_agents/script_agent.py +66 -0
@@ -0,0 +1,321 @@
1
+ """Evaluator-owned process isolation for one Pier candidate invocation."""
2
+
3
+ from __future__ import annotations
4
+
5
+ import secrets
6
+ import shlex
7
+ import tempfile
8
+ from dataclasses import dataclass
9
+ from pathlib import Path
10
+ from typing import Any, Protocol
11
+
12
+ from pier.environments.base import BaseEnvironment
13
+
14
+ from .candidate_contract import sha256_bytes
15
+
16
+
17
+ class ProcessBoundaryHost(Protocol):
18
+ async def _exec(
19
+ self, environment: BaseEnvironment, command: str, **kwargs: Any
20
+ ) -> Any: ...
21
+
22
+ async def _ensure_real_directory(
23
+ self, environment: BaseEnvironment, path: str, *, anchor: str, mode: int = 0o755
24
+ ) -> None: ...
25
+
26
+ async def _assert_real_directory(
27
+ self,
28
+ environment: BaseEnvironment,
29
+ path: str,
30
+ *,
31
+ label: str,
32
+ user: str | int | None = "root",
33
+ ) -> None: ...
34
+
35
+ async def _write_verified_file(
36
+ self,
37
+ environment: BaseEnvironment,
38
+ source: Path,
39
+ target: str,
40
+ *,
41
+ anchor: str,
42
+ mode: int,
43
+ digest: str,
44
+ byte_length: int,
45
+ ) -> None: ...
46
+
47
+
48
+ @dataclass(frozen=True)
49
+ class ProcessBoundaryConfig:
50
+ uid: int
51
+ gid: int
52
+ evaluator_root: str
53
+ control_root: str
54
+ home: str
55
+ temporary: str
56
+ isolate_user: bool = True
57
+
58
+
59
+ class CandidateProcessBoundary:
60
+ def __init__(self, host: ProcessBoundaryHost, config: ProcessBoundaryConfig):
61
+ self._host = host
62
+ self.config = config
63
+
64
+ async def prepare_identity(
65
+ self, environment: BaseEnvironment, workspace_roots: tuple[str, ...]
66
+ ) -> None:
67
+ config = self.config
68
+ await self._host._ensure_real_directory(
69
+ environment, config.evaluator_root, anchor="/", mode=0o755
70
+ )
71
+ await self._host._exec(
72
+ environment,
73
+ f"chmod 0755 -- {shlex.quote(config.evaluator_root)}",
74
+ user="root",
75
+ )
76
+ for path, mode in (
77
+ (config.control_root, 0o700),
78
+ (config.home, 0o700),
79
+ (config.temporary, 0o700),
80
+ ):
81
+ await self._host._ensure_real_directory(
82
+ environment, path, anchor="/", mode=mode
83
+ )
84
+ await self._host._exec(
85
+ environment,
86
+ f"chmod {mode:o} -- {shlex.quote(path)}",
87
+ user="root",
88
+ )
89
+ if not config.isolate_user:
90
+ return
91
+ await self._host._exec(
92
+ environment,
93
+ "\n".join(
94
+ [
95
+ "set -eu",
96
+ "command -v setsid >/dev/null",
97
+ "command -v setpriv >/dev/null",
98
+ f"! grep -l '^Uid:[[:space:]]*{config.uid}[[:space:]]' "
99
+ "/proc/[0-9]*/status >/dev/null 2>&1",
100
+ f"chown {config.uid}:{config.gid} -- "
101
+ f"{shlex.quote(config.home)} {shlex.quote(config.temporary)}",
102
+ f"chown -R {config.uid}:{config.gid} -- "
103
+ + " ".join(shlex.quote(root) for root in workspace_roots),
104
+ ]
105
+ ),
106
+ user="root",
107
+ )
108
+
109
+ async def stage_wrapper(self, environment: BaseEnvironment) -> tuple[str, str]:
110
+ token = secrets.token_hex(24)
111
+ wrapper = f"{self.config.control_root}/run-{token}.sh"
112
+ marker = f"{self.config.control_root}/timeout-{token}"
113
+ raw = self._runner_script()
114
+ with tempfile.NamedTemporaryFile(delete=False) as handle:
115
+ handle.write(raw)
116
+ local_path = Path(handle.name)
117
+ try:
118
+ await self._host._write_verified_file(
119
+ environment,
120
+ local_path,
121
+ wrapper,
122
+ anchor=self.config.control_root,
123
+ mode=0o700,
124
+ digest=sha256_bytes(raw),
125
+ byte_length=len(raw),
126
+ )
127
+ finally:
128
+ local_path.unlink(missing_ok=True)
129
+ return wrapper, marker
130
+
131
+ async def kill_and_wait(self, environment: BaseEnvironment) -> None:
132
+ """Kill every process in the candidate identity and prove none remain."""
133
+ if not self.config.isolate_user:
134
+ return
135
+ command = """set -eu
136
+ candidate_pids() {
137
+ for status in /proc/[0-9]*/status; do
138
+ test -r "$status" || continue
139
+ pid=${status#/proc/}
140
+ pid=${pid%/status}
141
+ while IFS=: read -r key value; do
142
+ if test "$key" = Uid; then
143
+ set -- $value
144
+ if test "${1:-}" = "$TARGET_UID"; then
145
+ printf '%s\n' "$pid"
146
+ fi
147
+ break
148
+ fi
149
+ done < "$status"
150
+ done
151
+ }
152
+
153
+ for attempt in $(seq 1 100); do
154
+ pids=$(candidate_pids)
155
+ test -n "$pids" || exit 0
156
+ for pid in $pids; do
157
+ kill -KILL "$pid" 2>/dev/null || true
158
+ done
159
+ sleep 0.02
160
+ done
161
+ remaining=$(candidate_pids)
162
+ test -z "$remaining" || {
163
+ printf 'candidate processes survived SIGKILL: %s\n' "$remaining" >&2
164
+ exit 1
165
+ }
166
+ """
167
+ await self._host._exec(
168
+ environment,
169
+ command,
170
+ env={"TARGET_UID": str(self.config.uid)},
171
+ user="root",
172
+ timeout_sec=5,
173
+ )
174
+
175
+ async def remove_control_files(
176
+ self,
177
+ environment: BaseEnvironment,
178
+ wrapper: str | None,
179
+ marker: str | None,
180
+ ) -> None:
181
+ paths = [path for path in (wrapper, marker) if path is not None]
182
+ if not paths:
183
+ return
184
+ await self._host._assert_real_directory(
185
+ environment,
186
+ self.config.control_root,
187
+ label="candidate control directory",
188
+ )
189
+ await self._host._exec(
190
+ environment,
191
+ "rm -f -- " + " ".join(shlex.quote(path) for path in paths),
192
+ user="root",
193
+ )
194
+
195
+ async def timeout_marker_exists(
196
+ self, environment: BaseEnvironment, marker: str
197
+ ) -> bool:
198
+ quoted = shlex.quote(marker)
199
+ result = await self._host._exec(
200
+ environment,
201
+ f'test -f {quoted} && test ! -L {quoted} && test "$(cat -- {quoted})" = timeout',
202
+ user="root",
203
+ check=False,
204
+ )
205
+ return result.return_code == 0
206
+
207
+ def _runner_script(self) -> bytes:
208
+ if not self.config.isolate_user:
209
+ return b"""#!/bin/sh
210
+ set -eu
211
+ marker=$1
212
+ uid=$2
213
+ gid=$3
214
+ deadline=$4
215
+ grace=$5
216
+ stdin_path=$6
217
+ shift 6
218
+ rm -f -- "$marker" "$marker.tmp"
219
+ set +e
220
+ if test "$stdin_path" = -; then
221
+ timeout --signal=TERM --kill-after="${grace}s" "${deadline}s" env "$@"
222
+ else
223
+ timeout --signal=TERM --kill-after="${grace}s" "${deadline}s" env "$@" < "$stdin_path"
224
+ fi
225
+ status=$?
226
+ set -e
227
+ if test "$status" = 124; then
228
+ printf timeout > "$marker"
229
+ fi
230
+ exit "$status"
231
+ """
232
+ return b"""#!/bin/sh
233
+ set -eu
234
+ marker=$1
235
+ uid=$2
236
+ gid=$3
237
+ deadline=$4
238
+ grace=$5
239
+ stdin_path=$6
240
+ shift 6
241
+ rm -f -- "$marker" "$marker.tmp"
242
+
243
+ kill_candidate_uid() {
244
+ signal=$1
245
+ for status in /proc/[0-9]*/status; do
246
+ test -r "$status" || continue
247
+ pid=${status#/proc/}
248
+ pid=${pid%/status}
249
+ while IFS=: read -r key value; do
250
+ if test "$key" = Uid; then
251
+ set -- $value
252
+ if test "${1:-}" = "$uid"; then
253
+ kill "-$signal" "$pid" 2>/dev/null || true
254
+ fi
255
+ break
256
+ fi
257
+ done < "$status"
258
+ done
259
+ }
260
+
261
+ candidate_uid_is_dead() {
262
+ for status in /proc/[0-9]*/status; do
263
+ test -r "$status" || continue
264
+ while IFS=: read -r key value; do
265
+ if test "$key" = Uid; then
266
+ set -- $value
267
+ test "${1:-}" = "$uid" && return 1
268
+ break
269
+ fi
270
+ done < "$status"
271
+ done
272
+ return 0
273
+ }
274
+
275
+ kill_and_wait_candidate_uid() {
276
+ kill_candidate_uid KILL
277
+ attempt=0
278
+ while ! candidate_uid_is_dead; do
279
+ attempt=$((attempt + 1))
280
+ test "$attempt" -lt 100 || {
281
+ printf 'candidate processes survived SIGKILL\n' >&2
282
+ return 1
283
+ }
284
+ kill_candidate_uid KILL
285
+ sleep 0.02
286
+ done
287
+ }
288
+
289
+ if test "$stdin_path" = -; then
290
+ setsid setpriv --reuid="$uid" --regid="$gid" --clear-groups --no-new-privs \
291
+ --bounding-set=-all --inh-caps=-all --ambient-caps=-all --pdeathsig=KILL -- env "$@" &
292
+ else
293
+ setsid setpriv --reuid="$uid" --regid="$gid" --clear-groups --no-new-privs \
294
+ --bounding-set=-all --inh-caps=-all --ambient-caps=-all --pdeathsig=KILL -- env "$@" < "$stdin_path" &
295
+ fi
296
+ candidate_pid=$!
297
+ (
298
+ sleep "$deadline"
299
+ if kill -0 "$candidate_pid" 2>/dev/null; then
300
+ umask 077
301
+ printf timeout > "$marker.tmp"
302
+ mv -fT -- "$marker.tmp" "$marker"
303
+ kill_candidate_uid TERM
304
+ sleep "$grace"
305
+ kill_and_wait_candidate_uid
306
+ fi
307
+ ) &
308
+ watchdog_pid=$!
309
+
310
+ set +e
311
+ wait "$candidate_pid"
312
+ candidate_status=$?
313
+ set -e
314
+ kill "$watchdog_pid" 2>/dev/null || true
315
+ wait "$watchdog_pid" 2>/dev/null || true
316
+ kill_and_wait_candidate_uid
317
+ if test -f "$marker"; then
318
+ exit 124
319
+ fi
320
+ exit "$candidate_status"
321
+ """