@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 +1,368 @@
1
+ """Verify and materialize runtime-signed workspaces inside a Pier container."""
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, PurePosixPath
10
+ from typing import Any, Protocol
11
+
12
+ from pier.environments.base import BaseEnvironment
13
+
14
+ from .candidate_contract import (
15
+ PreparedCandidateContract,
16
+ WorkspaceFile,
17
+ sha256_bytes,
18
+ )
19
+
20
+
21
+ def _workspace_check_script(
22
+ root: str,
23
+ files: tuple[WorkspaceFile, ...],
24
+ *,
25
+ allow_task_metadata: bool,
26
+ ) -> bytes:
27
+ quoted_root = shlex.quote(root)
28
+ prune = (
29
+ f"\\( -path {shlex.quote(root + '/.git')} -o "
30
+ f"-path {shlex.quote(root + '/.sidecar')} \\) -prune -o "
31
+ if allow_task_metadata
32
+ else ""
33
+ )
34
+ expected_args = " ".join(shlex.quote(file.path) for file in files)
35
+ expected = (
36
+ f"$(printf '%s\\n' {expected_args} | LC_ALL=C sort)" if files else "''"
37
+ )
38
+ checks = [
39
+ "#!/bin/sh",
40
+ "set -eu",
41
+ f"test -d {quoted_root}",
42
+ f"test ! -L {quoted_root}",
43
+ f'test "$(realpath -e -- {quoted_root})" = {quoted_root}',
44
+ f'test -z "$(find {quoted_root} {prune}-type l -print -quit)"',
45
+ f'test -z "$(find {quoted_root} {prune}! -type d ! -type f -print -quit)"',
46
+ f"observed=$(find {quoted_root} {prune}-type f -printf '%P\\n' | LC_ALL=C sort)",
47
+ f"expected={expected}",
48
+ 'test "$observed" = "$expected"',
49
+ ]
50
+ for file in files:
51
+ path = shlex.quote(f"{root}/{file.path}")
52
+ checks.extend(
53
+ [
54
+ f"test -f {path}",
55
+ f"test ! -L {path}",
56
+ f'test "$(stat -c %h -- {path})" = 1',
57
+ f'test "$(stat -c %a -- {path})" = {file.mode:o}',
58
+ f'test "$(wc -c < {path})" = {file.byte_length}',
59
+ f"test \"$(sha256sum -- {path} | cut -d' ' -f1)\" = "
60
+ f"{file.sha256.removeprefix('sha256:')}",
61
+ ]
62
+ )
63
+ return ("\n".join(checks) + "\n").encode("utf-8")
64
+
65
+
66
+ class WorkspaceBoundaryHost(Protocol):
67
+ async def _exec(
68
+ self, environment: BaseEnvironment, command: str, **kwargs: Any
69
+ ) -> Any: ...
70
+
71
+ async def _ensure_real_directory(
72
+ self,
73
+ environment: BaseEnvironment,
74
+ path: str,
75
+ *,
76
+ anchor: str,
77
+ mode: int = 0o755,
78
+ ) -> None: ...
79
+
80
+ async def _assert_real_directory(
81
+ self,
82
+ environment: BaseEnvironment,
83
+ path: str,
84
+ *,
85
+ label: str,
86
+ user: str | int | None = "root",
87
+ ) -> None: ...
88
+
89
+ async def _write_verified_file(
90
+ self,
91
+ environment: BaseEnvironment,
92
+ source: Path,
93
+ target: str,
94
+ *,
95
+ anchor: str,
96
+ mode: int,
97
+ digest: str,
98
+ byte_length: int,
99
+ ) -> None: ...
100
+
101
+
102
+ @dataclass(frozen=True)
103
+ class WorkspaceBoundaryConfig:
104
+ contract: PreparedCandidateContract
105
+ task_snapshot: Path
106
+ control_root: str
107
+ protected_roots: tuple[str, ...]
108
+ error_type: type[Exception]
109
+
110
+
111
+ def candidate_git_env(task_root: str) -> dict[str, str]:
112
+ return {
113
+ "GIT_CONFIG_COUNT": "1",
114
+ "GIT_CONFIG_KEY_0": "safe.directory",
115
+ "GIT_CONFIG_VALUE_0": task_root,
116
+ "GIT_CONFIG_NOSYSTEM": "1",
117
+ "GIT_CONFIG_GLOBAL": "/dev/null",
118
+ "GIT_CONFIG_SYSTEM": "/dev/null",
119
+ "GIT_NO_REPLACE_OBJECTS": "1",
120
+ "GIT_TERMINAL_PROMPT": "0",
121
+ "LC_ALL": "C",
122
+ }
123
+
124
+
125
+ class CandidateWorkspaceBoundary:
126
+ def __init__(self, host: WorkspaceBoundaryHost, config: WorkspaceBoundaryConfig):
127
+ self._host = host
128
+ self._config = config
129
+
130
+ @property
131
+ def _contract(self) -> PreparedCandidateContract:
132
+ return self._config.contract
133
+
134
+ def _error(self, message: str) -> Exception:
135
+ return self._config.error_type(message)
136
+
137
+ async def verify_repository(self, environment: BaseEnvironment) -> None:
138
+ root = shlex.quote(self._contract.task_root)
139
+ git = f"git -c core.hooksPath=/dev/null -C {root}"
140
+ env = candidate_git_env(self._contract.task_root)
141
+ head = await self._host._exec(environment, f"{git} rev-parse HEAD", env=env)
142
+ tree = await self._host._exec(
143
+ environment, f"{git} rev-parse 'HEAD^{{tree}}'", env=env
144
+ )
145
+ replacements = await self._host._exec(
146
+ environment,
147
+ f"{git} for-each-ref --format='%(refname)' refs/replace",
148
+ env=env,
149
+ )
150
+ if (head.stdout or "").strip() != self._contract.repository_base_commit:
151
+ raise self._error("task checkout HEAD does not match the signed base commit")
152
+ if (tree.stdout or "").strip() != self._contract.repository_base_tree:
153
+ raise self._error("task checkout tree does not match the signed base tree")
154
+ if (replacements.stdout or "").strip():
155
+ raise self._error("task checkout contains forbidden Git replace refs")
156
+
157
+ async def materialize_task_workspace(self, environment: BaseEnvironment) -> None:
158
+ root = self._contract.task_root
159
+ quoted = shlex.quote(root)
160
+ exists = await self._host._exec(
161
+ environment,
162
+ f"test -e {quoted} || test -L {quoted}",
163
+ user="root",
164
+ check=False,
165
+ )
166
+ if exists.return_code == 0:
167
+ return
168
+ if exists.return_code != 1:
169
+ raise self._error("could not inspect the signed task root")
170
+
171
+ parent = str(PurePosixPath(root).parent)
172
+ await self._host._ensure_real_directory(environment, parent, anchor="/")
173
+ await self._host._exec(
174
+ environment,
175
+ "\n".join(
176
+ [
177
+ "set -eu",
178
+ f"test ! -e {quoted}",
179
+ f"test ! -L {quoted}",
180
+ f"mkdir -- {quoted}",
181
+ f'test "$(realpath -e -- {quoted})" = {quoted}',
182
+ ]
183
+ ),
184
+ user="root",
185
+ )
186
+ await environment.upload_dir(self._config.task_snapshot, root)
187
+ git = f"git -c core.hooksPath=/dev/null -C {quoted}"
188
+ await self._host._exec(
189
+ environment,
190
+ "\n".join(
191
+ [
192
+ "set -eu",
193
+ f"{git} init -b main",
194
+ f"{git} config user.email fixture@tangle.tools",
195
+ f'{git} config user.name "Tangle Fixture"',
196
+ f"{git} add -A",
197
+ f"GIT_AUTHOR_DATE=2000-01-01T00:00:00Z "
198
+ f"GIT_COMMITTER_DATE=2000-01-01T00:00:00Z "
199
+ f"{git} commit -m baseline",
200
+ f"{git} tag benchmark-base",
201
+ ]
202
+ ),
203
+ env=candidate_git_env(root),
204
+ user="root",
205
+ )
206
+
207
+ async def verify_pier_execution_identity(
208
+ self, environment: BaseEnvironment
209
+ ) -> None:
210
+ expected = self._contract.container
211
+ configured_image = getattr(environment.task_env_config, "docker_image", None)
212
+ pinned_image = f"{expected['image']}@{expected['indexDigest']}"
213
+ if configured_image != pinned_image:
214
+ raise self._error(
215
+ "Pier task image does not match the signed immutable image: "
216
+ f"expected={pinned_image} observed={configured_image}"
217
+ )
218
+ configured_os = getattr(environment.task_env_config, "os", None)
219
+ configured_os = getattr(configured_os, "value", configured_os)
220
+ if configured_os is not None and str(configured_os).lower() != "linux":
221
+ raise self._error(
222
+ f"Pier task OS does not match signed Linux platform: {configured_os}"
223
+ )
224
+ platform = expected["platform"]
225
+ observed = await self._host._exec(
226
+ environment,
227
+ 'printf \'%s %s\' "$(uname -s)" "$(uname -m)"',
228
+ user="root",
229
+ )
230
+ kernel, _, machine = (observed.stdout or "").strip().partition(" ")
231
+ architectures = {"x86_64": "amd64", "aarch64": "arm64"}
232
+ if kernel != "Linux" or architectures.get(machine) != platform["architecture"]:
233
+ raise self._error(
234
+ "running Pier platform does not match the signed container platform: "
235
+ f"observed={kernel}/{machine} expected=linux/{platform['architecture']}"
236
+ )
237
+
238
+ async def verify_resolved_root_isolation(
239
+ self, environment: BaseEnvironment
240
+ ) -> None:
241
+ roots = [self._contract.task_root]
242
+ if self._contract.candidate_root is not None:
243
+ roots.append(self._contract.candidate_root)
244
+ for root in roots:
245
+ await self._host._assert_real_directory(
246
+ environment, root, label="execution workspace"
247
+ )
248
+ if self._contract.candidate_root is not None:
249
+ task = shlex.quote(self._contract.task_root)
250
+ candidate = shlex.quote(self._contract.candidate_root)
251
+ await self._host._exec(
252
+ environment,
253
+ "\n".join(
254
+ [
255
+ "set -eu",
256
+ f"task=$(realpath -e -- {task})",
257
+ f"candidate=$(realpath -e -- {candidate})",
258
+ 'test "$task" != "$candidate"',
259
+ 'case "$task" in "$candidate"/*) exit 1;; esac',
260
+ 'case "$candidate" in "$task"/*) exit 1;; esac',
261
+ ]
262
+ ),
263
+ user="root",
264
+ )
265
+ for protected in self._config.protected_roots:
266
+ quoted_protected = shlex.quote(protected)
267
+ for root in roots:
268
+ quoted_root = shlex.quote(root)
269
+ await self._host._exec(
270
+ environment,
271
+ "\n".join(
272
+ [
273
+ "set -eu",
274
+ f"if test -e {quoted_protected}; then",
275
+ f" protected=$(realpath -e -- {quoted_protected})",
276
+ f" root=$(realpath -e -- {quoted_root})",
277
+ ' test "$root" != "$protected"',
278
+ ' case "$root" in "$protected"/*) exit 1;; esac',
279
+ ' case "$protected" in "$root"/*) exit 1;; esac',
280
+ "fi",
281
+ ]
282
+ ),
283
+ user="root",
284
+ )
285
+
286
+ async def verify_container_workspace(
287
+ self,
288
+ environment: BaseEnvironment,
289
+ root: str,
290
+ files: tuple[WorkspaceFile, ...],
291
+ *,
292
+ allow_task_metadata: bool,
293
+ ) -> None:
294
+ script = _workspace_check_script(
295
+ root,
296
+ files,
297
+ allow_task_metadata=allow_task_metadata,
298
+ )
299
+ control_root = self._config.control_root
300
+ target = f"{control_root}/workspace-check-{secrets.token_hex(24)}.sh"
301
+ with tempfile.NamedTemporaryFile(delete=False) as handle:
302
+ handle.write(script)
303
+ local_path = Path(handle.name)
304
+ control_ready = False
305
+
306
+ async def cleanup() -> None:
307
+ local_path.unlink(missing_ok=True)
308
+ if not control_ready:
309
+ return
310
+ await self._host._assert_real_directory(
311
+ environment,
312
+ control_root,
313
+ label="candidate control directory",
314
+ )
315
+ await self._host._exec(
316
+ environment,
317
+ f"rm -f -- {shlex.quote(target)}",
318
+ user="root",
319
+ )
320
+ await self._host._exec(
321
+ environment,
322
+ "\n".join(
323
+ [
324
+ "set -eu",
325
+ f"test ! -e {shlex.quote(target)}",
326
+ f"test ! -L {shlex.quote(target)}",
327
+ f'test -z "$(find {shlex.quote(control_root)} -maxdepth 1 '
328
+ "-name '.tangle-*' -print -quit)\"",
329
+ ]
330
+ ),
331
+ user="root",
332
+ )
333
+
334
+ try:
335
+ await self._host._ensure_real_directory(
336
+ environment,
337
+ control_root,
338
+ anchor="/",
339
+ mode=0o700,
340
+ )
341
+ control_ready = True
342
+ await self._host._exec(
343
+ environment,
344
+ f"chmod 0700 -- {shlex.quote(control_root)}",
345
+ user="root",
346
+ )
347
+ await self._host._write_verified_file(
348
+ environment,
349
+ local_path,
350
+ target,
351
+ anchor=control_root,
352
+ mode=0o600,
353
+ digest=sha256_bytes(script),
354
+ byte_length=len(script),
355
+ )
356
+ await self._host._exec(
357
+ environment,
358
+ f"/bin/sh {shlex.quote(target)}",
359
+ user="root",
360
+ )
361
+ except BaseException as primary_error:
362
+ try:
363
+ await cleanup()
364
+ except BaseException as cleanup_error:
365
+ primary_error.add_note(f"workspace check cleanup failed: {cleanup_error}")
366
+ raise
367
+ else:
368
+ await cleanup()