self-bench 0.3.0

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 (500) hide show
  1. package/LICENSE +21 -0
  2. package/README.md +256 -0
  3. package/dist/activities.d.ts +41 -0
  4. package/dist/activities.d.ts.map +1 -0
  5. package/dist/activities.js +823 -0
  6. package/dist/activities.js.map +1 -0
  7. package/dist/agent-smoke-main.d.ts +3 -0
  8. package/dist/agent-smoke-main.d.ts.map +1 -0
  9. package/dist/agent-smoke-main.js +54 -0
  10. package/dist/agent-smoke-main.js.map +1 -0
  11. package/dist/agent-smoke.d.ts +17 -0
  12. package/dist/agent-smoke.d.ts.map +1 -0
  13. package/dist/agent-smoke.js +98 -0
  14. package/dist/agent-smoke.js.map +1 -0
  15. package/dist/api-main.d.ts +2 -0
  16. package/dist/api-main.d.ts.map +1 -0
  17. package/dist/api-main.js +11 -0
  18. package/dist/api-main.js.map +1 -0
  19. package/dist/api.d.ts +3 -0
  20. package/dist/api.d.ts.map +1 -0
  21. package/dist/api.js +212 -0
  22. package/dist/api.js.map +1 -0
  23. package/dist/artifacts.d.ts +32 -0
  24. package/dist/artifacts.d.ts.map +1 -0
  25. package/dist/artifacts.js +311 -0
  26. package/dist/artifacts.js.map +1 -0
  27. package/dist/audit.d.ts +12 -0
  28. package/dist/audit.d.ts.map +1 -0
  29. package/dist/audit.js +69 -0
  30. package/dist/audit.js.map +1 -0
  31. package/dist/cli.d.ts +3 -0
  32. package/dist/cli.d.ts.map +1 -0
  33. package/dist/cli.js +297 -0
  34. package/dist/cli.js.map +1 -0
  35. package/dist/codex-review.d.ts +40 -0
  36. package/dist/codex-review.d.ts.map +1 -0
  37. package/dist/codex-review.js +189 -0
  38. package/dist/codex-review.js.map +1 -0
  39. package/dist/config.d.ts +35 -0
  40. package/dist/config.d.ts.map +1 -0
  41. package/dist/config.js +78 -0
  42. package/dist/config.js.map +1 -0
  43. package/dist/contracts.d.ts +223 -0
  44. package/dist/contracts.d.ts.map +1 -0
  45. package/dist/contracts.js +103 -0
  46. package/dist/contracts.js.map +1 -0
  47. package/dist/coupling.d.ts +45 -0
  48. package/dist/coupling.d.ts.map +1 -0
  49. package/dist/coupling.js +159 -0
  50. package/dist/coupling.js.map +1 -0
  51. package/dist/docker-executor.d.ts +11 -0
  52. package/dist/docker-executor.d.ts.map +1 -0
  53. package/dist/docker-executor.js +103 -0
  54. package/dist/docker-executor.js.map +1 -0
  55. package/dist/eval-main.d.ts +3 -0
  56. package/dist/eval-main.d.ts.map +1 -0
  57. package/dist/eval-main.js +80 -0
  58. package/dist/eval-main.js.map +1 -0
  59. package/dist/evaluate.d.ts +24 -0
  60. package/dist/evaluate.d.ts.map +1 -0
  61. package/dist/evaluate.js +208 -0
  62. package/dist/evaluate.js.map +1 -0
  63. package/dist/extensions/authoring.d.ts +3 -0
  64. package/dist/extensions/authoring.d.ts.map +1 -0
  65. package/dist/extensions/authoring.js +71 -0
  66. package/dist/extensions/authoring.js.map +1 -0
  67. package/dist/extensions/discovery.d.ts +3 -0
  68. package/dist/extensions/discovery.d.ts.map +1 -0
  69. package/dist/extensions/discovery.js +51 -0
  70. package/dist/extensions/discovery.js.map +1 -0
  71. package/dist/extensions/review.d.ts +3 -0
  72. package/dist/extensions/review.d.ts.map +1 -0
  73. package/dist/extensions/review.js +44 -0
  74. package/dist/extensions/review.js.map +1 -0
  75. package/dist/github.d.ts +3 -0
  76. package/dist/github.d.ts.map +1 -0
  77. package/dist/github.js +19 -0
  78. package/dist/github.js.map +1 -0
  79. package/dist/harbor-results.d.ts +14 -0
  80. package/dist/harbor-results.d.ts.map +1 -0
  81. package/dist/harbor-results.js +105 -0
  82. package/dist/harbor-results.js.map +1 -0
  83. package/dist/harbor-task.d.ts +12 -0
  84. package/dist/harbor-task.d.ts.map +1 -0
  85. package/dist/harbor-task.js +462 -0
  86. package/dist/harbor-task.js.map +1 -0
  87. package/dist/hash.d.ts +2 -0
  88. package/dist/hash.d.ts.map +1 -0
  89. package/dist/hash.js +5 -0
  90. package/dist/hash.js.map +1 -0
  91. package/dist/modal-auth.d.ts +2 -0
  92. package/dist/modal-auth.d.ts.map +1 -0
  93. package/dist/modal-auth.js +9 -0
  94. package/dist/modal-auth.js.map +1 -0
  95. package/dist/modal-executor.d.ts +12 -0
  96. package/dist/modal-executor.d.ts.map +1 -0
  97. package/dist/modal-executor.js +152 -0
  98. package/dist/modal-executor.js.map +1 -0
  99. package/dist/parallel.d.ts +2 -0
  100. package/dist/parallel.d.ts.map +1 -0
  101. package/dist/parallel.js +19 -0
  102. package/dist/parallel.js.map +1 -0
  103. package/dist/process.d.ts +27 -0
  104. package/dist/process.d.ts.map +1 -0
  105. package/dist/process.js +136 -0
  106. package/dist/process.js.map +1 -0
  107. package/dist/provenance.d.ts +22 -0
  108. package/dist/provenance.d.ts.map +1 -0
  109. package/dist/provenance.js +361 -0
  110. package/dist/provenance.js.map +1 -0
  111. package/dist/reaudit-main.d.ts +3 -0
  112. package/dist/reaudit-main.d.ts.map +1 -0
  113. package/dist/reaudit-main.js +155 -0
  114. package/dist/reaudit-main.js.map +1 -0
  115. package/dist/repair-main.d.ts +3 -0
  116. package/dist/repair-main.d.ts.map +1 -0
  117. package/dist/repair-main.js +168 -0
  118. package/dist/repair-main.js.map +1 -0
  119. package/dist/repair.d.ts +9 -0
  120. package/dist/repair.d.ts.map +1 -0
  121. package/dist/repair.js +46 -0
  122. package/dist/repair.js.map +1 -0
  123. package/dist/review/assets/abap-BdImnpbu.js +1 -0
  124. package/dist/review/assets/actionscript-3-B3316cI-.js +1 -0
  125. package/dist/review/assets/ada-bCR0ucgS.js +1 -0
  126. package/dist/review/assets/ahk-CsyLZFj1.js +1 -0
  127. package/dist/review/assets/ahk2-8Zs4aa1G.js +1 -0
  128. package/dist/review/assets/andromeeda-C4gqWexZ.js +1 -0
  129. package/dist/review/assets/angular-html-DA-rfuFy.js +1 -0
  130. package/dist/review/assets/angular-ts-BrjP3tb8.js +1 -0
  131. package/dist/review/assets/apache-Pmp26Uib.js +1 -0
  132. package/dist/review/assets/apex-DhZFqWV2.js +1 -0
  133. package/dist/review/assets/apl-CORt7UWP.js +1 -0
  134. package/dist/review/assets/applescript-Co6uUVPk.js +1 -0
  135. package/dist/review/assets/ara-BRHolxvo.js +1 -0
  136. package/dist/review/assets/asciidoc-CSVQ5wI8.js +1 -0
  137. package/dist/review/assets/asm-D_Q5rh1f.js +1 -0
  138. package/dist/review/assets/astro-HNnZUWAn.js +1 -0
  139. package/dist/review/assets/aurora-x-D-2ljcwZ.js +1 -0
  140. package/dist/review/assets/awk-DMzUqQB5.js +1 -0
  141. package/dist/review/assets/ayu-dark-DYE7WIF3.js +1 -0
  142. package/dist/review/assets/ayu-light-BA47KaF1.js +1 -0
  143. package/dist/review/assets/ayu-mirage-32ctXXKs.js +1 -0
  144. package/dist/review/assets/ballerina-BFfxhgS-.js +1 -0
  145. package/dist/review/assets/bat-CickPsom.js +1 -0
  146. package/dist/review/assets/beancount-k_qm7-4y.js +1 -0
  147. package/dist/review/assets/berry-uYugtg8r.js +1 -0
  148. package/dist/review/assets/bibtex-CHM0blh-.js +1 -0
  149. package/dist/review/assets/bicep-Bmn6On1c.js +1 -0
  150. package/dist/review/assets/bird2-Bx8U0n9b.js +1 -0
  151. package/dist/review/assets/blade-2xfisSek.js +1 -0
  152. package/dist/review/assets/bsl-DlhNcFeZ.js +1 -0
  153. package/dist/review/assets/c-BIGW1oBm.js +1 -0
  154. package/dist/review/assets/c3-Dp5svz6Z.js +1 -0
  155. package/dist/review/assets/cadence-Bv_4Rxtq.js +1 -0
  156. package/dist/review/assets/cairo-KRGpt6FW.js +1 -0
  157. package/dist/review/assets/catppuccin-frappe-CZL1YF0i.js +1 -0
  158. package/dist/review/assets/catppuccin-latte-DH-8KZSZ.js +1 -0
  159. package/dist/review/assets/catppuccin-macchiato-B7yYVSCf.js +1 -0
  160. package/dist/review/assets/catppuccin-mocha-Ct7hS0mc.js +1 -0
  161. package/dist/review/assets/chapel-DTp_pixX.js +1 -0
  162. package/dist/review/assets/clarity-Dn5IMItf.js +1 -0
  163. package/dist/review/assets/clojure-P80f7IUj.js +1 -0
  164. package/dist/review/assets/cmake-D1j8_8rp.js +1 -0
  165. package/dist/review/assets/cobol-nBiQ_Alo.js +1 -0
  166. package/dist/review/assets/codeowners-Bp6g37R7.js +1 -0
  167. package/dist/review/assets/codeql-DsOJ9woJ.js +1 -0
  168. package/dist/review/assets/coffee-Ch7k5sss.js +1 -0
  169. package/dist/review/assets/common-lisp-Cg-RD9OK.js +1 -0
  170. package/dist/review/assets/coq-C7JzOVbR.js +1 -0
  171. package/dist/review/assets/cpp-BMRokrvK.js +1 -0
  172. package/dist/review/assets/crystal-DGywbUpC.js +1 -0
  173. package/dist/review/assets/csharp-DSvCPggb.js +1 -0
  174. package/dist/review/assets/css-CLj8gQPS.js +1 -0
  175. package/dist/review/assets/csv-fuZLfV_i.js +1 -0
  176. package/dist/review/assets/cue-D82EKSYY.js +1 -0
  177. package/dist/review/assets/cypher-COkxafJQ.js +1 -0
  178. package/dist/review/assets/d-85-TOEBH.js +1 -0
  179. package/dist/review/assets/dark-plus-C3mMm8J8.js +1 -0
  180. package/dist/review/assets/dart-bE4Kk8sk.js +1 -0
  181. package/dist/review/assets/dax-CEL-wOlO.js +1 -0
  182. package/dist/review/assets/desktop-BmXAJ9_W.js +1 -0
  183. package/dist/review/assets/diff-D97Zzqfu.js +1 -0
  184. package/dist/review/assets/docker-BcOcwvcX.js +1 -0
  185. package/dist/review/assets/dotenv-Da5cRb03.js +1 -0
  186. package/dist/review/assets/dracula-BzJJZx-M.js +1 -0
  187. package/dist/review/assets/dracula-soft-BXkSAIEj.js +1 -0
  188. package/dist/review/assets/dream-maker-BtqSS_iP.js +1 -0
  189. package/dist/review/assets/edge-FbVlp4U3.js +1 -0
  190. package/dist/review/assets/elixir-CkH2-t6x.js +1 -0
  191. package/dist/review/assets/elm-DbKCFpqz.js +1 -0
  192. package/dist/review/assets/emacs-lisp-C_m_b--Z.js +1 -0
  193. package/dist/review/assets/erb-DXfck5VN.js +1 -0
  194. package/dist/review/assets/erlang-DsQrWhSR.js +1 -0
  195. package/dist/review/assets/everforest-dark-BgDCqdQA.js +1 -0
  196. package/dist/review/assets/everforest-light-C8M2exoo.js +1 -0
  197. package/dist/review/assets/fennel-BYunw83y.js +1 -0
  198. package/dist/review/assets/fish-BvzEVeQv.js +1 -0
  199. package/dist/review/assets/fluent-C4IJs8-o.js +1 -0
  200. package/dist/review/assets/fortran-fixed-form-CkoXwp7k.js +1 -0
  201. package/dist/review/assets/fortran-free-form-BxgE0vQu.js +1 -0
  202. package/dist/review/assets/fsharp-CXgrBDvD.js +1 -0
  203. package/dist/review/assets/gdresource-TyuKm33G.js +1 -0
  204. package/dist/review/assets/gdscript-DqcFQ5yU.js +1 -0
  205. package/dist/review/assets/gdshader-DkwncUOv.js +1 -0
  206. package/dist/review/assets/genie-D0YGMca9.js +1 -0
  207. package/dist/review/assets/gherkin-DyxjwDmM.js +1 -0
  208. package/dist/review/assets/git-commit-F4YmCXRG.js +1 -0
  209. package/dist/review/assets/git-rebase-r7XF79zn.js +1 -0
  210. package/dist/review/assets/github-dark-DHJKELXO.js +1 -0
  211. package/dist/review/assets/github-dark-default-Cuk6v7N8.js +1 -0
  212. package/dist/review/assets/github-dark-dimmed-DH5Ifo-i.js +1 -0
  213. package/dist/review/assets/github-dark-high-contrast-E3gJ1_iC.js +1 -0
  214. package/dist/review/assets/github-light-DAi9KRSo.js +1 -0
  215. package/dist/review/assets/github-light-default-D7oLnXFd.js +1 -0
  216. package/dist/review/assets/github-light-high-contrast-BfjtVDDH.js +1 -0
  217. package/dist/review/assets/gleam-BspZqrRM.js +1 -0
  218. package/dist/review/assets/glimmer-js-ByusRIyA.js +1 -0
  219. package/dist/review/assets/glimmer-ts-BfAWNZQY.js +1 -0
  220. package/dist/review/assets/glsl-DplSGwfg.js +1 -0
  221. package/dist/review/assets/gn-n2N0HUVH.js +1 -0
  222. package/dist/review/assets/gnuplot-DdkO51Og.js +1 -0
  223. package/dist/review/assets/go-C27-OAKa.js +1 -0
  224. package/dist/review/assets/graphql-ChdNCCLP.js +1 -0
  225. package/dist/review/assets/groovy-gcz8RCvz.js +1 -0
  226. package/dist/review/assets/gruvbox-dark-hard-CFHQjOhq.js +1 -0
  227. package/dist/review/assets/gruvbox-dark-medium-GsRaNv29.js +1 -0
  228. package/dist/review/assets/gruvbox-dark-soft-CVdnzihN.js +1 -0
  229. package/dist/review/assets/gruvbox-light-hard-CH1njM8p.js +1 -0
  230. package/dist/review/assets/gruvbox-light-medium-DRw_LuNl.js +1 -0
  231. package/dist/review/assets/gruvbox-light-soft-hJgmCMqR.js +1 -0
  232. package/dist/review/assets/hack-BWmVpMyf.js +1 -0
  233. package/dist/review/assets/haml-D5jkg6IW.js +1 -0
  234. package/dist/review/assets/handlebars-BpdQsYii.js +1 -0
  235. package/dist/review/assets/haskell-Df6bDoY_.js +1 -0
  236. package/dist/review/assets/haxe-CfZj7gIn.js +1 -0
  237. package/dist/review/assets/hcl-BWvSN4gD.js +1 -0
  238. package/dist/review/assets/hjson-D5-asLiD.js +1 -0
  239. package/dist/review/assets/hlsl-D3lLCCz7.js +1 -0
  240. package/dist/review/assets/horizon-BUw7H-hv.js +1 -0
  241. package/dist/review/assets/horizon-bright-CUuTKBJd.js +1 -0
  242. package/dist/review/assets/houston-DnULxvSX.js +1 -0
  243. package/dist/review/assets/html-derivative-DlHx6ybY.js +1 -0
  244. package/dist/review/assets/html-pp8916En.js +1 -0
  245. package/dist/review/assets/http-jrhK8wxY.js +1 -0
  246. package/dist/review/assets/hurl-irOxFIW8.js +1 -0
  247. package/dist/review/assets/hxml-2-FPmUDs.js +1 -0
  248. package/dist/review/assets/hy-DFXneXwc.js +1 -0
  249. package/dist/review/assets/imba-DGztddWO.js +1 -0
  250. package/dist/review/assets/index-Dq-6DFzI.css +1 -0
  251. package/dist/review/assets/index-DurW_RD7.js +1567 -0
  252. package/dist/review/assets/ini-BEwlwnbL.js +1 -0
  253. package/dist/review/assets/java-CylS5w8V.js +1 -0
  254. package/dist/review/assets/javascript-wDzz0qaB.js +1 -0
  255. package/dist/review/assets/jinja-f2NsQr07.js +1 -0
  256. package/dist/review/assets/jison-wvAkD_A8.js +1 -0
  257. package/dist/review/assets/json-Cp-IABpG.js +1 -0
  258. package/dist/review/assets/json5-C9tS-k6U.js +1 -0
  259. package/dist/review/assets/jsonc-Des-eS-w.js +1 -0
  260. package/dist/review/assets/jsonl-DcaNXYhu.js +1 -0
  261. package/dist/review/assets/jsonnet-DFQXde-d.js +1 -0
  262. package/dist/review/assets/jssm-C2t-YnRu.js +1 -0
  263. package/dist/review/assets/jsx-g9-lgVsj.js +1 -0
  264. package/dist/review/assets/julia-5Bft2YPA.js +1 -0
  265. package/dist/review/assets/just-Cwhn7H3k.js +1 -0
  266. package/dist/review/assets/kanagawa-dragon-BuwD2xS4.js +1 -0
  267. package/dist/review/assets/kanagawa-lotus-C3DzagqV.js +1 -0
  268. package/dist/review/assets/kanagawa-wave-DFGoQZhC.js +1 -0
  269. package/dist/review/assets/kdl-DV7GczEv.js +1 -0
  270. package/dist/review/assets/kotlin-BdnUsdx6.js +1 -0
  271. package/dist/review/assets/kusto-wEQ09or8.js +1 -0
  272. package/dist/review/assets/laserwave-DUszq2jm.js +1 -0
  273. package/dist/review/assets/latex-D5pSuvFb.js +1 -0
  274. package/dist/review/assets/lean-BZvkOJ9d.js +1 -0
  275. package/dist/review/assets/less-B1dDrJ26.js +1 -0
  276. package/dist/review/assets/light-plus-B7mTdjB0.js +1 -0
  277. package/dist/review/assets/liquid-C0sCDyMI.js +1 -0
  278. package/dist/review/assets/llvm-BZoOZj88.js +1 -0
  279. package/dist/review/assets/log-2UxHyX5q.js +1 -0
  280. package/dist/review/assets/logo-BtOb2qkB.js +1 -0
  281. package/dist/review/assets/lua-BaeVxFsk.js +1 -0
  282. package/dist/review/assets/luau-BnpPk5vE.js +1 -0
  283. package/dist/review/assets/make-CHLpvVh8.js +1 -0
  284. package/dist/review/assets/markdown-Cvjx9yec.js +1 -0
  285. package/dist/review/assets/marko-DjSrsDqO.js +1 -0
  286. package/dist/review/assets/material-theme-D5KoaKCx.js +1 -0
  287. package/dist/review/assets/material-theme-darker-BfHTSMKl.js +1 -0
  288. package/dist/review/assets/material-theme-lighter-B0m2ddpp.js +1 -0
  289. package/dist/review/assets/material-theme-ocean-CyktbL80.js +1 -0
  290. package/dist/review/assets/material-theme-palenight-Csfq5Kiy.js +1 -0
  291. package/dist/review/assets/matlab-D7o27uSR.js +1 -0
  292. package/dist/review/assets/mdc-D1_yUvq7.js +1 -0
  293. package/dist/review/assets/mdx-Cmh6b_Ma.js +1 -0
  294. package/dist/review/assets/mermaid-CQcHuHx7.js +1 -0
  295. package/dist/review/assets/min-dark-CafNBF8u.js +1 -0
  296. package/dist/review/assets/min-light-CTRr51gU.js +1 -0
  297. package/dist/review/assets/mipsasm-CKIfxQSi.js +1 -0
  298. package/dist/review/assets/mojo-Blby8dQK.js +1 -0
  299. package/dist/review/assets/monokai-BRVnQi9A.js +1 -0
  300. package/dist/review/assets/moonbit-CHtswR0a.js +1 -0
  301. package/dist/review/assets/move-el3G9tDJ.js +1 -0
  302. package/dist/review/assets/narrat-DRg8JJMk.js +1 -0
  303. package/dist/review/assets/nextflow-C-mBbutL.js +1 -0
  304. package/dist/review/assets/nextflow-groovy-vE_lwT2v.js +1 -0
  305. package/dist/review/assets/nginx-BpAMiNFr.js +1 -0
  306. package/dist/review/assets/night-owl-C39BiMTA.js +1 -0
  307. package/dist/review/assets/night-owl-light-CMTm3GFP.js +1 -0
  308. package/dist/review/assets/nim-BIad80T-.js +1 -0
  309. package/dist/review/assets/nix-CwoSXNpI.js +1 -0
  310. package/dist/review/assets/nord-Ddv68eIx.js +1 -0
  311. package/dist/review/assets/nsis-BlV79W_Q.js +1 -0
  312. package/dist/review/assets/nushell-D3jzshHO.js +1 -0
  313. package/dist/review/assets/objective-c-DXmwc3jG.js +1 -0
  314. package/dist/review/assets/objective-cpp-CLxacb5B.js +1 -0
  315. package/dist/review/assets/ocaml-C0hk2d4L.js +1 -0
  316. package/dist/review/assets/odin-BBf5iR-q.js +1 -0
  317. package/dist/review/assets/one-dark-pro-DVMEJ2y_.js +1 -0
  318. package/dist/review/assets/one-light-C3Wv6jpd.js +1 -0
  319. package/dist/review/assets/openscad-C4EeE6gA.js +1 -0
  320. package/dist/review/assets/org-DM6o9KBp.js +1 -0
  321. package/dist/review/assets/pascal-D93ZcfNL.js +1 -0
  322. package/dist/review/assets/perl-B9cMNwum.js +1 -0
  323. package/dist/review/assets/php-Csjmro_R.js +1 -0
  324. package/dist/review/assets/pierre-dark-CyvmCCZW.js +1 -0
  325. package/dist/review/assets/pierre-dark-protanopia-deuteranopia-Rgc0TwpF.js +1 -0
  326. package/dist/review/assets/pierre-dark-soft-BHGpRqa4.js +1 -0
  327. package/dist/review/assets/pierre-dark-tritanopia-Beq2gCRQ.js +1 -0
  328. package/dist/review/assets/pierre-dark-vibrant-BWBVywrn.js +1 -0
  329. package/dist/review/assets/pierre-light-480U9XYS.js +1 -0
  330. package/dist/review/assets/pierre-light-protanopia-deuteranopia-CaVOBURG.js +1 -0
  331. package/dist/review/assets/pierre-light-soft-CVdyfjmI.js +1 -0
  332. package/dist/review/assets/pierre-light-tritanopia-B4_gpKOM.js +1 -0
  333. package/dist/review/assets/pierre-light-vibrant-DdTDNdfJ.js +1 -0
  334. package/dist/review/assets/pkl-u5AG7uiY.js +1 -0
  335. package/dist/review/assets/plastic-3e1v2bzS.js +1 -0
  336. package/dist/review/assets/plsql-ChMvpjG-.js +1 -0
  337. package/dist/review/assets/po-BTJTHyun.js +1 -0
  338. package/dist/review/assets/poimandres-CS3Unz2-.js +1 -0
  339. package/dist/review/assets/polar-C0HS_06l.js +1 -0
  340. package/dist/review/assets/postcss-CXtECtnM.js +1 -0
  341. package/dist/review/assets/powerquery-CEu0bR-o.js +1 -0
  342. package/dist/review/assets/powershell-BmBUJMz7.js +1 -0
  343. package/dist/review/assets/prisma-Vru482bI.js +1 -0
  344. package/dist/review/assets/prolog-CbFg5uaA.js +1 -0
  345. package/dist/review/assets/proto-C7zT0LnQ.js +1 -0
  346. package/dist/review/assets/pug-DKIMFp6K.js +1 -0
  347. package/dist/review/assets/puppet-BMWR74SV.js +1 -0
  348. package/dist/review/assets/purescript-CklMAg4u.js +1 -0
  349. package/dist/review/assets/python-B6aJPvgy.js +1 -0
  350. package/dist/review/assets/qml-3beO22l8.js +1 -0
  351. package/dist/review/assets/qmldir-C8lEn-DE.js +1 -0
  352. package/dist/review/assets/qss-IeuSbFQv.js +1 -0
  353. package/dist/review/assets/r-Cf5RLm7j.js +1 -0
  354. package/dist/review/assets/racket-BqYA7rlc.js +1 -0
  355. package/dist/review/assets/raku-DXvB9xmW.js +1 -0
  356. package/dist/review/assets/razor-BjBPvh-w.js +1 -0
  357. package/dist/review/assets/rbs-CpoqiR4B.js +1 -0
  358. package/dist/review/assets/red-hvxz__6c.js +1 -0
  359. package/dist/review/assets/reg-C-SQnVFl.js +1 -0
  360. package/dist/review/assets/regexp-CDVJQ6XC.js +1 -0
  361. package/dist/review/assets/rel-C3B-1QV4.js +1 -0
  362. package/dist/review/assets/riscv-BM1_JUlF.js +1 -0
  363. package/dist/review/assets/ron-D8l8udqQ.js +1 -0
  364. package/dist/review/assets/rose-pine-5qJOZa0Y.js +1 -0
  365. package/dist/review/assets/rose-pine-dawn-zx0QlTCp.js +1 -0
  366. package/dist/review/assets/rose-pine-moon-C9pEdX9L.js +1 -0
  367. package/dist/review/assets/rosmsg-BJDFO7_C.js +1 -0
  368. package/dist/review/assets/rst-bs7f0vWN.js +1 -0
  369. package/dist/review/assets/ruby-C0TQ7zu5.js +1 -0
  370. package/dist/review/assets/rust-B1yitclQ.js +1 -0
  371. package/dist/review/assets/sas-DEy46yEz.js +1 -0
  372. package/dist/review/assets/sass-Cj5Yp3dK.js +1 -0
  373. package/dist/review/assets/scala-CqE71os6.js +1 -0
  374. package/dist/review/assets/scheme-C98Dy4si.js +1 -0
  375. package/dist/review/assets/scss-D5BDwBP9.js +1 -0
  376. package/dist/review/assets/sdbl-DVxCFoDh.js +1 -0
  377. package/dist/review/assets/shaderlab-Dg9Lc6iA.js +1 -0
  378. package/dist/review/assets/shellscript-Yzrsuije.js +1 -0
  379. package/dist/review/assets/shellsession-BADoaaVG.js +1 -0
  380. package/dist/review/assets/slack-dark-BthQWCQV.js +1 -0
  381. package/dist/review/assets/slack-ochin-DqwNpetd.js +1 -0
  382. package/dist/review/assets/smalltalk-BOQMe2GC.js +1 -0
  383. package/dist/review/assets/smithy-cds9vsN8.js +1 -0
  384. package/dist/review/assets/snazzy-light-Bw305WKR.js +1 -0
  385. package/dist/review/assets/solarized-dark-CpvCGNkr.js +1 -0
  386. package/dist/review/assets/solarized-light-Dlz6yCKv.js +1 -0
  387. package/dist/review/assets/solidity-DijEV5ha.js +1 -0
  388. package/dist/review/assets/soy-8wufbnw4.js +1 -0
  389. package/dist/review/assets/sparql-rVzFXLq3.js +1 -0
  390. package/dist/review/assets/splunk-BtCnVYZw.js +1 -0
  391. package/dist/review/assets/sql-CRqJ_cUM.js +1 -0
  392. package/dist/review/assets/ssh-config-_ykCGR6B.js +1 -0
  393. package/dist/review/assets/stata-DI20mbqo.js +1 -0
  394. package/dist/review/assets/stylus-BEDo0Tqx.js +1 -0
  395. package/dist/review/assets/surrealql-Cjom0U5J.js +1 -0
  396. package/dist/review/assets/svelte-Cy7k_4gC.js +1 -0
  397. package/dist/review/assets/swift-C2oV4EkX.js +1 -0
  398. package/dist/review/assets/synthwave-84-CbfX1IO0.js +1 -0
  399. package/dist/review/assets/system-verilog-0hqHdDBg.js +1 -0
  400. package/dist/review/assets/systemd-4A_iFExJ.js +1 -0
  401. package/dist/review/assets/talonscript-CkByrt1z.js +1 -0
  402. package/dist/review/assets/tasl-QIJgUcNo.js +1 -0
  403. package/dist/review/assets/tcl-dwOrl1Do.js +1 -0
  404. package/dist/review/assets/templ-DhtptRzy.js +1 -0
  405. package/dist/review/assets/terraform-BETggiCN.js +1 -0
  406. package/dist/review/assets/tex-D96PA37w.js +1 -0
  407. package/dist/review/assets/tokyo-night-hegEt444.js +1 -0
  408. package/dist/review/assets/toml-vGWfd6FD.js +1 -0
  409. package/dist/review/assets/ts-tags-D351s5mN.js +1 -0
  410. package/dist/review/assets/tsv-B_m7g4N7.js +1 -0
  411. package/dist/review/assets/tsx-COt5Ahok.js +1 -0
  412. package/dist/review/assets/turtle-BsS91CYL.js +1 -0
  413. package/dist/review/assets/twig-27uCiNez.js +1 -0
  414. package/dist/review/assets/typescript-BPQ3VLAy.js +1 -0
  415. package/dist/review/assets/typespec-CAFt9gP4.js +1 -0
  416. package/dist/review/assets/typst-DHCkPAjA.js +1 -0
  417. package/dist/review/assets/v-BGw2Nkan.js +1 -0
  418. package/dist/review/assets/vala-CsfeWuGM.js +1 -0
  419. package/dist/review/assets/vb-Cu-pLBUe.js +1 -0
  420. package/dist/review/assets/verilog-nZwndyjY.js +1 -0
  421. package/dist/review/assets/vesper-DRje8inN.js +1 -0
  422. package/dist/review/assets/vhdl-CeAyd5Ju.js +1 -0
  423. package/dist/review/assets/viml-CJc9bBzg.js +1 -0
  424. package/dist/review/assets/vitesse-black-Bkuqu6BP.js +1 -0
  425. package/dist/review/assets/vitesse-dark-D0r3Knsf.js +1 -0
  426. package/dist/review/assets/vitesse-light-CVO1_9PV.js +1 -0
  427. package/dist/review/assets/vue-BqiEGhQt.js +1 -0
  428. package/dist/review/assets/vue-html-AaS7Mt5G.js +1 -0
  429. package/dist/review/assets/vue-vine-BoDAl6tE.js +1 -0
  430. package/dist/review/assets/vyper-CDx5xZoG.js +1 -0
  431. package/dist/review/assets/wasm-CG6Dc4jp.js +1 -0
  432. package/dist/review/assets/wasm-MzD3tlZU.js +1 -0
  433. package/dist/review/assets/wenyan-BV7otONQ.js +1 -0
  434. package/dist/review/assets/wgsl-Dx-B1_4e.js +1 -0
  435. package/dist/review/assets/wikitext-BhOHFoWU.js +1 -0
  436. package/dist/review/assets/wit-5i3qLPDT.js +1 -0
  437. package/dist/review/assets/wolfram-lXgVvXCa.js +1 -0
  438. package/dist/review/assets/xml-sdJ4AIDG.js +1 -0
  439. package/dist/review/assets/xsl-CtQFsRM5.js +1 -0
  440. package/dist/review/assets/yaml-Buea-lGh.js +1 -0
  441. package/dist/review/assets/zenscript-DVFEvuxE.js +1 -0
  442. package/dist/review/assets/zig-VOosw3JB.js +1 -0
  443. package/dist/review/index.html +15 -0
  444. package/dist/run-wait.d.ts +13 -0
  445. package/dist/run-wait.d.ts.map +1 -0
  446. package/dist/run-wait.js +25 -0
  447. package/dist/run-wait.js.map +1 -0
  448. package/dist/sandbox-author.bundle.js +14979 -0
  449. package/dist/sandbox-author.d.ts +3 -0
  450. package/dist/sandbox-author.d.ts.map +1 -0
  451. package/dist/sandbox-author.js +17 -0
  452. package/dist/sandbox-author.js.map +1 -0
  453. package/dist/sandbox-repair.bundle.js +314 -0
  454. package/dist/sandbox-repair.d.ts +3 -0
  455. package/dist/sandbox-repair.d.ts.map +1 -0
  456. package/dist/sandbox-repair.js +121 -0
  457. package/dist/sandbox-repair.js.map +1 -0
  458. package/dist/sandbox-review.bundle.js +14458 -0
  459. package/dist/sandbox-review.d.ts +2 -0
  460. package/dist/sandbox-review.d.ts.map +1 -0
  461. package/dist/sandbox-review.js +16 -0
  462. package/dist/sandbox-review.js.map +1 -0
  463. package/dist/sandbox-validation-repair.bundle.js +14744 -0
  464. package/dist/sandbox-validation-repair.d.ts +3 -0
  465. package/dist/sandbox-validation-repair.d.ts.map +1 -0
  466. package/dist/sandbox-validation-repair.js +134 -0
  467. package/dist/sandbox-validation-repair.js.map +1 -0
  468. package/dist/sandbox.d.ts +39 -0
  469. package/dist/sandbox.d.ts.map +1 -0
  470. package/dist/sandbox.js +8 -0
  471. package/dist/sandbox.js.map +1 -0
  472. package/dist/subscription-auth.d.ts +6 -0
  473. package/dist/subscription-auth.d.ts.map +1 -0
  474. package/dist/subscription-auth.js +59 -0
  475. package/dist/subscription-auth.js.map +1 -0
  476. package/dist/temporal.d.ts +6 -0
  477. package/dist/temporal.d.ts.map +1 -0
  478. package/dist/temporal.js +17 -0
  479. package/dist/temporal.js.map +1 -0
  480. package/dist/validate-main.d.ts +3 -0
  481. package/dist/validate-main.d.ts.map +1 -0
  482. package/dist/validate-main.js +137 -0
  483. package/dist/validate-main.js.map +1 -0
  484. package/dist/validation-repair.d.ts +10 -0
  485. package/dist/validation-repair.d.ts.map +1 -0
  486. package/dist/validation-repair.js +72 -0
  487. package/dist/validation-repair.js.map +1 -0
  488. package/dist/worker-main.d.ts +2 -0
  489. package/dist/worker-main.d.ts.map +1 -0
  490. package/dist/worker-main.js +30 -0
  491. package/dist/worker-main.js.map +1 -0
  492. package/dist/workflow.d.ts +6 -0
  493. package/dist/workflow.d.ts.map +1 -0
  494. package/dist/workflow.js +428 -0
  495. package/dist/workflow.js.map +1 -0
  496. package/package.json +74 -0
  497. package/src/extensions/authoring.ts +86 -0
  498. package/src/extensions/discovery.ts +61 -0
  499. package/src/extensions/review.ts +54 -0
  500. package/src/skills/selfbench/SKILL.md +119 -0
@@ -0,0 +1,11 @@
1
+ import type { SelfBenchConfig } from "./config.js";
2
+ import type { SandboxExecutor, SandboxRequest, SandboxResult, SandboxRunOptions } from "./sandbox.js";
3
+ export declare class DockerSandboxExecutor implements SandboxExecutor {
4
+ #private;
5
+ constructor(config: Extract<SelfBenchConfig["execution"], {
6
+ kind: "docker";
7
+ }>);
8
+ run(request: SandboxRequest, options?: SandboxRunOptions): Promise<SandboxResult>;
9
+ close(): void;
10
+ }
11
+ //# sourceMappingURL=docker-executor.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"docker-executor.d.ts","sourceRoot":"","sources":["../src/docker-executor.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAEnD,OAAO,KAAK,EACV,eAAe,EACf,cAAc,EACd,aAAa,EACb,iBAAiB,EAClB,MAAM,cAAc,CAAC;AAEtB,qBAAa,qBAAsB,YAAW,eAAe;;gBAG/C,MAAM,EAAE,OAAO,CAAC,eAAe,CAAC,WAAW,CAAC,EAAE;QAAE,IAAI,EAAE,QAAQ,CAAA;KAAE,CAAC;IAIvE,GAAG,CAAC,OAAO,EAAE,cAAc,EAAE,OAAO,GAAE,iBAAsB,GAAG,OAAO,CAAC,aAAa,CAAC;IA4E3F,KAAK,IAAI,IAAI;CACd"}
@@ -0,0 +1,103 @@
1
+ import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
2
+ import { tmpdir } from "node:os";
3
+ import { dirname, join, relative, resolve } from "node:path";
4
+ import { runCommand } from "./process.js";
5
+ export class DockerSandboxExecutor {
6
+ #config;
7
+ constructor(config) {
8
+ this.#config = config;
9
+ }
10
+ async run(request, options = {}) {
11
+ const root = await mkdtemp(join(tmpdir(), "selfbench-docker-"));
12
+ const sandboxId = sandboxName(request.runId, request.stage);
13
+ try {
14
+ for (const file of request.files ?? []) {
15
+ const destination = hostPath(root, file.path);
16
+ await mkdir(dirname(destination), { recursive: true });
17
+ await writeFile(destination, file.contents);
18
+ }
19
+ const environment = { ...request.environment, ...request.secrets };
20
+ const args = [
21
+ "create",
22
+ "--name",
23
+ sandboxId,
24
+ "--cpus",
25
+ String(request.cpu ?? 4),
26
+ "--memory",
27
+ `${request.memoryMiB ?? 8192}m`,
28
+ "--volume",
29
+ `${sandboxId}:/work`,
30
+ "--workdir",
31
+ "/work",
32
+ ];
33
+ for (const key of Object.keys(environment)) {
34
+ args.push("--env", key);
35
+ }
36
+ args.push(this.#config.image, ...request.command);
37
+ await runCommand("docker", ["volume", "create", sandboxId]);
38
+ await runCommand("docker", args, { env: { ...process.env, ...environment } });
39
+ await runCommand("docker", ["cp", `${root}/.`, `${sandboxId}:/work/`]);
40
+ const abort = () => {
41
+ void runCommand("docker", ["rm", "--force", sandboxId], { allowFailure: true });
42
+ };
43
+ options.signal?.addEventListener("abort", abort, { once: true });
44
+ try {
45
+ const result = await runCommand("docker", ["start", "--attach", sandboxId], {
46
+ allowFailure: true,
47
+ timeoutMs: request.timeoutMs,
48
+ ...(request.inactivityTimeoutMs
49
+ ? { inactivityTimeoutMs: request.inactivityTimeoutMs }
50
+ : {}),
51
+ onOutput: (stream, chunk) => options.onProgress?.({ stream, bytes: chunk.byteLength }),
52
+ });
53
+ const outputs = {};
54
+ for (const path of request.outputPaths ?? []) {
55
+ const destination = hostPath(root, path);
56
+ await mkdir(dirname(destination), { recursive: true });
57
+ const copied = await runCommand("docker", ["cp", `${sandboxId}:${path}`, destination], {
58
+ allowFailure: true,
59
+ });
60
+ if (copied.exitCode === 0) {
61
+ outputs[path] = await readFile(destination);
62
+ }
63
+ else if (result.exitCode === 0) {
64
+ throw new Error(`sandbox ${sandboxId} exited successfully without output ${path}`);
65
+ }
66
+ }
67
+ return { sandboxId, ...result, outputs };
68
+ }
69
+ finally {
70
+ options.signal?.removeEventListener("abort", abort);
71
+ await runCommand("docker", ["rm", "--force", sandboxId], { allowFailure: true });
72
+ await runCommand("docker", ["volume", "rm", "--force", sandboxId], {
73
+ allowFailure: true,
74
+ });
75
+ }
76
+ }
77
+ finally {
78
+ await runCommand("docker", ["rm", "--force", sandboxId], { allowFailure: true });
79
+ await runCommand("docker", ["volume", "rm", "--force", sandboxId], {
80
+ allowFailure: true,
81
+ });
82
+ await rm(root, { recursive: true, force: true });
83
+ }
84
+ }
85
+ close() { }
86
+ }
87
+ function hostPath(root, containerPath) {
88
+ if (!containerPath.startsWith("/work/")) {
89
+ throw new Error(`sandbox path must be beneath /work: ${containerPath}`);
90
+ }
91
+ const path = resolve(root, relative("/work", containerPath));
92
+ if (path !== root && !path.startsWith(`${root}/`)) {
93
+ throw new Error(`sandbox path escapes /work: ${containerPath}`);
94
+ }
95
+ return path;
96
+ }
97
+ function sandboxName(runId, stage) {
98
+ const suffix = crypto.randomUUID().slice(0, 8);
99
+ return `selfbench-${runId.slice(0, 20)}-${stage.slice(0, 16)}-${suffix}`
100
+ .toLowerCase()
101
+ .replace(/[^a-z0-9_.-]/g, "-");
102
+ }
103
+ //# sourceMappingURL=docker-executor.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"docker-executor.js","sourceRoot":"","sources":["../src/docker-executor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAC3E,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAE7D,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAQ1C,MAAM,OAAO,qBAAqB;IACvB,OAAO,CAA4D;IAE5E,YAAY,MAAiE;QAC3E,IAAI,CAAC,OAAO,GAAG,MAAM,CAAC;IACxB,CAAC;IAED,KAAK,CAAC,GAAG,CAAC,OAAuB,EAAE,UAA6B,EAAE;QAChE,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,mBAAmB,CAAC,CAAC,CAAC;QAChE,MAAM,SAAS,GAAG,WAAW,CAAC,OAAO,CAAC,KAAK,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;QAC5D,IAAI,CAAC;YACH,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,KAAK,IAAI,EAAE,EAAE,CAAC;gBACvC,MAAM,WAAW,GAAG,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;gBAC9C,MAAM,KAAK,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;gBACvD,MAAM,SAAS,CAAC,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC9C,CAAC;YAED,MAAM,WAAW,GAAG,EAAE,GAAG,OAAO,CAAC,WAAW,EAAE,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;YACnE,MAAM,IAAI,GAAG;gBACX,QAAQ;gBACR,QAAQ;gBACR,SAAS;gBACT,QAAQ;gBACR,MAAM,CAAC,OAAO,CAAC,GAAG,IAAI,CAAC,CAAC;gBACxB,UAAU;gBACV,GAAG,OAAO,CAAC,SAAS,IAAI,IAAI,GAAG;gBAC/B,UAAU;gBACV,GAAG,SAAS,QAAQ;gBACpB,WAAW;gBACX,OAAO;aACR,CAAC;YACF,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,EAAE,CAAC;gBAC3C,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC;YAC1B,CAAC;YACD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;YAElD,MAAM,UAAU,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,QAAQ,EAAE,SAAS,CAAC,CAAC,CAAC;YAC5D,MAAM,UAAU,CAAC,QAAQ,EAAE,IAAI,EAAE,EAAE,GAAG,EAAE,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,GAAG,WAAW,EAAE,EAAE,CAAC,CAAC;YAC9E,MAAM,UAAU,CAAC,QAAQ,EAAE,CAAC,IAAI,EAAE,GAAG,IAAI,IAAI,EAAE,GAAG,SAAS,SAAS,CAAC,CAAC,CAAC;YAEvE,MAAM,KAAK,GAAG,GAAG,EAAE;gBACjB,KAAK,UAAU,CAAC,QAAQ,EAAE,CAAC,IAAI,EAAE,SAAS,EAAE,SAAS,CAAC,EAAE,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC;YAClF,CAAC,CAAC;YACF,OAAO,CAAC,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;YACjE,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,CAAC,OAAO,EAAE,UAAU,EAAE,SAAS,CAAC,EAAE;oBAC1E,YAAY,EAAE,IAAI;oBAClB,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,GAAG,CAAC,OAAO,CAAC,mBAAmB;wBAC7B,CAAC,CAAC,EAAE,mBAAmB,EAAE,OAAO,CAAC,mBAAmB,EAAE;wBACtD,CAAC,CAAC,EAAE,CAAC;oBACP,QAAQ,EAAE,CAAC,MAAM,EAAE,KAAK,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,CAAC,UAAU,EAAE,CAAC;iBACvF,CAAC,CAAC;gBACH,MAAM,OAAO,GAA+B,EAAE,CAAC;gBAC/C,KAAK,MAAM,IAAI,IAAI,OAAO,CAAC,WAAW,IAAI,EAAE,EAAE,CAAC;oBAC7C,MAAM,WAAW,GAAG,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;oBACzC,MAAM,KAAK,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;oBACvD,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,QAAQ,EAAE,CAAC,IAAI,EAAE,GAAG,SAAS,IAAI,IAAI,EAAE,EAAE,WAAW,CAAC,EAAE;wBACrF,YAAY,EAAE,IAAI;qBACnB,CAAC,CAAC;oBACH,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;wBAC1B,OAAO,CAAC,IAAI,CAAC,GAAG,MAAM,QAAQ,CAAC,WAAW,CAAC,CAAC;oBAC9C,CAAC;yBAAM,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;wBACjC,MAAM,IAAI,KAAK,CAAC,WAAW,SAAS,uCAAuC,IAAI,EAAE,CAAC,CAAC;oBACrF,CAAC;gBACH,CAAC;gBACD,OAAO,EAAE,SAAS,EAAE,GAAG,MAAM,EAAE,OAAO,EAAE,CAAC;YAC3C,CAAC;oBAAS,CAAC;gBACT,OAAO,CAAC,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;gBACpD,MAAM,UAAU,CAAC,QAAQ,EAAE,CAAC,IAAI,EAAE,SAAS,EAAE,SAAS,CAAC,EAAE,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC;gBACjF,MAAM,UAAU,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,SAAS,CAAC,EAAE;oBACjE,YAAY,EAAE,IAAI;iBACnB,CAAC,CAAC;YACL,CAAC;QACH,CAAC;gBAAS,CAAC;YACT,MAAM,UAAU,CAAC,QAAQ,EAAE,CAAC,IAAI,EAAE,SAAS,EAAE,SAAS,CAAC,EAAE,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC;YACjF,MAAM,UAAU,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,SAAS,CAAC,EAAE;gBACjE,YAAY,EAAE,IAAI;aACnB,CAAC,CAAC;YACH,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACnD,CAAC;IACH,CAAC;IAED,KAAK,KAAU,CAAC;CACjB;AAED,SAAS,QAAQ,CAAC,IAAY,EAAE,aAAqB;IACnD,IAAI,CAAC,aAAa,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;QACxC,MAAM,IAAI,KAAK,CAAC,uCAAuC,aAAa,EAAE,CAAC,CAAC;IAC1E,CAAC;IACD,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,EAAE,aAAa,CAAC,CAAC,CAAC;IAC7D,IAAI,IAAI,KAAK,IAAI,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,IAAI,GAAG,CAAC,EAAE,CAAC;QAClD,MAAM,IAAI,KAAK,CAAC,+BAA+B,aAAa,EAAE,CAAC,CAAC;IAClE,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,WAAW,CAAC,KAAa,EAAE,KAAa;IAC/C,MAAM,MAAM,GAAG,MAAM,CAAC,UAAU,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC/C,OAAO,aAAa,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,MAAM,EAAE;SACrE,WAAW,EAAE;SACb,OAAO,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;AACnC,CAAC"}
@@ -0,0 +1,3 @@
1
+ #!/usr/bin/env node
2
+ export {};
3
+ //# sourceMappingURL=eval-main.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"eval-main.d.ts","sourceRoot":"","sources":["../src/eval-main.ts"],"names":[],"mappings":""}
@@ -0,0 +1,80 @@
1
+ #!/usr/bin/env node
2
+ import { parseArgs } from "node:util";
3
+ import { MATRIX_MODELS, runMatrix } from "./evaluate.js";
4
+ const parsed = parseArgs({
5
+ options: {
6
+ export: { type: "string" },
7
+ tasks: { type: "string" },
8
+ jobs: { type: "string" },
9
+ harbor: { type: "string" },
10
+ environment: { type: "string", default: "modal" },
11
+ concurrency: { type: "string", default: "3" },
12
+ auth: { type: "string" },
13
+ model: { type: "string", multiple: true },
14
+ help: { type: "boolean", short: "h" },
15
+ },
16
+ strict: true,
17
+ });
18
+ if (parsed.values.help) {
19
+ console.log(`Run Harbor tasks through the fixed Codex subscription model matrix.
20
+
21
+ Usage:
22
+ selfbench-eval (--export FILE.tar.gz | --tasks DIRECTORY) --jobs DIRECTORY [options]
23
+
24
+ Options:
25
+ --tasks DIRECTORY Expanded Harbor tasks (one or more)
26
+ --harbor PATH Harbor executable (default: harbor)
27
+ --environment docker|modal Execution environment (default: modal)
28
+ --concurrency N Concurrent trials (default: 3)
29
+ --model MODEL Run only this model; may be repeated
30
+ --auth FILE Codex ChatGPT auth.json path
31
+ -h, --help Show this help`);
32
+ process.exit(0);
33
+ }
34
+ const environment = parsed.values.environment;
35
+ if (environment !== "docker" && environment !== "modal") {
36
+ throw new Error("--environment must be docker or modal");
37
+ }
38
+ const concurrency = positiveInteger(parsed.values.concurrency, "--concurrency");
39
+ const models = evaluationModels(parsed.values.model);
40
+ const summaries = await runMatrix({
41
+ ...(parsed.values.export ? { exportPath: parsed.values.export } : {}),
42
+ ...(parsed.values.tasks ? { tasksPath: parsed.values.tasks } : {}),
43
+ jobsDirectory: parsed.values.jobs ?? fail("--jobs is required"),
44
+ environment,
45
+ concurrency,
46
+ ...(parsed.values.harbor ? { harborPath: parsed.values.harbor } : {}),
47
+ ...(parsed.values.auth ? { authPath: parsed.values.auth } : {}),
48
+ ...(models ? { models } : {}),
49
+ onTrialComplete: (summary, completed, total) => {
50
+ const status = summary.passed ? "passed" : summary.exception ? "error" : "failed";
51
+ console.error(`[${completed}/${total}] ${summary.model} ${summary.taskId}: ${status}`);
52
+ },
53
+ });
54
+ console.log(JSON.stringify({
55
+ trials: summaries.length,
56
+ passed: summaries.filter((summary) => summary.passed).length,
57
+ failed: summaries.filter((summary) => !summary.passed).length,
58
+ }, null, 2));
59
+ function fail(message) {
60
+ throw new Error(message);
61
+ }
62
+ function evaluationModels(values) {
63
+ if (!values) {
64
+ return undefined;
65
+ }
66
+ const allowed = new Set(MATRIX_MODELS);
67
+ const invalid = values.filter((value) => !allowed.has(value));
68
+ if (invalid.length > 0) {
69
+ throw new Error(`--model must be one of ${MATRIX_MODELS.join(", ")}; got ${invalid.join(", ")}`);
70
+ }
71
+ return [...new Set(values)];
72
+ }
73
+ function positiveInteger(value, label) {
74
+ const parsed = Number(value);
75
+ if (!Number.isInteger(parsed) || parsed < 1) {
76
+ throw new Error(`${label} must be a positive integer`);
77
+ }
78
+ return parsed;
79
+ }
80
+ //# sourceMappingURL=eval-main.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"eval-main.js","sourceRoot":"","sources":["../src/eval-main.ts"],"names":[],"mappings":";AAEA,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AACtC,OAAO,EAAE,aAAa,EAAoB,SAAS,EAAE,MAAM,eAAe,CAAC;AAE3E,MAAM,MAAM,GAAG,SAAS,CAAC;IACvB,OAAO,EAAE;QACP,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC1B,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACzB,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACxB,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QAC1B,WAAW,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE;QACjD,WAAW,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,EAAE;QAC7C,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;QACxB,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE;QACzC,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,GAAG,EAAE;KACtC;IACD,MAAM,EAAE,IAAI;CACb,CAAC,CAAC;AACH,IAAI,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;IACvB,OAAO,CAAC,GAAG,CAAC;;;;;;;;;;;;4CAY8B,CAAC,CAAC;IAC5C,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC;AACD,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC;AAC9C,IAAI,WAAW,KAAK,QAAQ,IAAI,WAAW,KAAK,OAAO,EAAE,CAAC;IACxD,MAAM,IAAI,KAAK,CAAC,uCAAuC,CAAC,CAAC;AAC3D,CAAC;AACD,MAAM,WAAW,GAAG,eAAe,CAAC,MAAM,CAAC,MAAM,CAAC,WAAW,EAAE,eAAe,CAAC,CAAC;AAChF,MAAM,MAAM,GAAG,gBAAgB,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AACrD,MAAM,SAAS,GAAG,MAAM,SAAS,CAAC;IAChC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACrE,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,MAAM,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IAClE,aAAa,EAAE,MAAM,CAAC,MAAM,CAAC,IAAI,IAAI,IAAI,CAAC,oBAAoB,CAAC;IAC/D,WAAW;IACX,WAAW;IACX,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACrE,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IAC/D,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IAC7B,eAAe,EAAE,CAAC,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,EAAE;QAC7C,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC;QAClF,OAAO,CAAC,KAAK,CAAC,IAAI,SAAS,IAAI,KAAK,KAAK,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,MAAM,KAAK,MAAM,EAAE,CAAC,CAAC;IACzF,CAAC;CACF,CAAC,CAAC;AACH,OAAO,CAAC,GAAG,CACT,IAAI,CAAC,SAAS,CACZ;IACE,MAAM,EAAE,SAAS,CAAC,MAAM;IACxB,MAAM,EAAE,SAAS,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM;IAC5D,MAAM,EAAE,SAAS,CAAC,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM;CAC9D,EACD,IAAI,EACJ,CAAC,CACF,CACF,CAAC;AAEF,SAAS,IAAI,CAAC,OAAe;IAC3B,MAAM,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;AAC3B,CAAC;AAED,SAAS,gBAAgB,CAAC,MAA4B;IACpD,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,MAAM,OAAO,GAAG,IAAI,GAAG,CAAS,aAAa,CAAC,CAAC;IAC/C,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC;IAC9D,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvB,MAAM,IAAI,KAAK,CACb,0BAA0B,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAChF,CAAC;IACJ,CAAC;IACD,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,CAAkB,CAAC;AAC/C,CAAC;AAED,SAAS,eAAe,CAAC,KAAyB,EAAE,KAAa;IAC/D,MAAM,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;IAC7B,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QAC5C,MAAM,IAAI,KAAK,CAAC,GAAG,KAAK,6BAA6B,CAAC,CAAC;IACzD,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}
@@ -0,0 +1,24 @@
1
+ export declare const MATRIX_MODELS: readonly ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"];
2
+ export type MatrixModel = (typeof MATRIX_MODELS)[number];
3
+ export interface MatrixOptions {
4
+ readonly exportPath?: string;
5
+ readonly tasksPath?: string;
6
+ readonly jobsDirectory: string;
7
+ readonly harborPath?: string;
8
+ readonly environment?: "docker" | "modal";
9
+ readonly concurrency?: number;
10
+ readonly authPath?: string;
11
+ readonly models?: readonly MatrixModel[];
12
+ readonly onTrialComplete?: (summary: MatrixTrialSummary, completed: number, total: number) => void;
13
+ }
14
+ export interface MatrixTrialSummary {
15
+ readonly taskId: string;
16
+ readonly model: MatrixModel;
17
+ readonly jobName: string;
18
+ readonly passed: boolean;
19
+ readonly rewards: Readonly<Record<string, number>>;
20
+ readonly exception?: string;
21
+ }
22
+ export declare function runMatrix(options: MatrixOptions): Promise<readonly MatrixTrialSummary[]>;
23
+ export declare function summarizeResult(taskId: string, model: MatrixModel, jobName: string, result: unknown): MatrixTrialSummary;
24
+ //# sourceMappingURL=evaluate.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"evaluate.d.ts","sourceRoot":"","sources":["../src/evaluate.ts"],"names":[],"mappings":"AAQA,eAAO,MAAM,aAAa,2DAA4D,CAAC;AACvF,MAAM,MAAM,WAAW,GAAG,CAAC,OAAO,aAAa,CAAC,CAAC,MAAM,CAAC,CAAC;AAGzD,MAAM,WAAW,aAAa;IAC5B,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,aAAa,EAAE,MAAM,CAAC;IAC/B,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,WAAW,CAAC,EAAE,QAAQ,GAAG,OAAO,CAAC;IAC1C,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;IAC9B,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,MAAM,CAAC,EAAE,SAAS,WAAW,EAAE,CAAC;IACzC,QAAQ,CAAC,eAAe,CAAC,EAAE,CACzB,OAAO,EAAE,kBAAkB,EAC3B,SAAS,EAAE,MAAM,EACjB,KAAK,EAAE,MAAM,KACV,IAAI,CAAC;CACX;AAED,MAAM,WAAW,kBAAkB;IACjC,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,KAAK,EAAE,WAAW,CAAC;IAC5B,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;IACzB,QAAQ,CAAC,OAAO,EAAE,QAAQ,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC;IACnD,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED,wBAAsB,SAAS,CAAC,OAAO,EAAE,aAAa,GAAG,OAAO,CAAC,SAAS,kBAAkB,EAAE,CAAC,CA6C9F;AA8FD,wBAAgB,eAAe,CAC7B,MAAM,EAAE,MAAM,EACd,KAAK,EAAE,WAAW,EAClB,OAAO,EAAE,MAAM,EACf,MAAM,EAAE,OAAO,GACd,kBAAkB,CA6BpB"}
@@ -0,0 +1,208 @@
1
+ import { cp, mkdir, mkdtemp, readdir, readFile, rm, writeFile } from "node:fs/promises";
2
+ import { homedir, tmpdir } from "node:os";
3
+ import { basename, join, resolve } from "node:path";
4
+ import { archiveIncompleteHarborJob, tryReadHarborJobResult } from "./harbor-results.js";
5
+ import { parallelMap } from "./parallel.js";
6
+ import { runCommand } from "./process.js";
7
+ import { assertCodexSubscriptionAuth } from "./subscription-auth.js";
8
+ export const MATRIX_MODELS = ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"];
9
+ const CODEX_VERSION = "0.146.1";
10
+ export async function runMatrix(options) {
11
+ const root = resolve(options.jobsDirectory);
12
+ const tasksDirectory = join(root, "tasks");
13
+ await mkdir(root, { recursive: true });
14
+ const taskDirectories = await resolveMatrixTasks(options, tasksDirectory);
15
+ const authPath = resolve(options.authPath ?? join(homedir(), ".codex/auth.json"));
16
+ await assertSubscriptionAuth(authPath);
17
+ const models = options.models ?? MATRIX_MODELS;
18
+ if (models.length < 1) {
19
+ throw new Error("provide at least one evaluation model");
20
+ }
21
+ const work = models.flatMap((model) => taskDirectories.map((taskDirectory) => ({ model, taskDirectory })));
22
+ let completed = 0;
23
+ const summaries = await parallelMap(work, options.concurrency ?? 3, async (item) => {
24
+ const summary = await runTrial({
25
+ jobsDirectory: root,
26
+ harborPath: options.harborPath ?? "harbor",
27
+ environment: options.environment ?? "modal",
28
+ authPath,
29
+ ...item,
30
+ });
31
+ completed += 1;
32
+ options.onTrialComplete?.(summary, completed, work.length);
33
+ return summary;
34
+ });
35
+ await writeFile(join(root, "summary.json"), `${JSON.stringify({
36
+ schemaVersion: 1,
37
+ agent: "codex",
38
+ agentVersion: CODEX_VERSION,
39
+ auth: "codex-subscription",
40
+ reasoningEffort: "high",
41
+ models,
42
+ taskCount: taskDirectories.length,
43
+ trials: summaries,
44
+ }, null, 2)}\n`);
45
+ return summaries;
46
+ }
47
+ async function resolveMatrixTasks(options, materializedTasksDirectory) {
48
+ if (options.exportPath && options.tasksPath) {
49
+ throw new Error("provide exactly one of exportPath or tasksPath");
50
+ }
51
+ if (options.tasksPath) {
52
+ const tasks = await taskDirectories(resolve(options.tasksPath));
53
+ if (tasks.length < 1) {
54
+ throw new Error(`found no Harbor tasks in ${resolve(options.tasksPath)}`);
55
+ }
56
+ return tasks;
57
+ }
58
+ if (!options.exportPath) {
59
+ throw new Error("provide exactly one of exportPath or tasksPath");
60
+ }
61
+ const tasks = await materializeExport(resolve(options.exportPath), materializedTasksDirectory);
62
+ if (tasks.length < 1) {
63
+ throw new Error(`found no Harbor tasks in ${resolve(options.exportPath)}`);
64
+ }
65
+ return tasks;
66
+ }
67
+ async function runTrial(input) {
68
+ const taskId = basename(input.taskDirectory);
69
+ const jobName = `${taskId}-${input.model}`.toLowerCase().replace(/[^a-z0-9_.-]/g, "-");
70
+ const existing = await tryReadHarborJobResult(input.jobsDirectory, jobName);
71
+ if (existing) {
72
+ return summarizeResult(taskId, input.model, jobName, existing.trial);
73
+ }
74
+ await archiveIncompleteHarborJob(input.jobsDirectory, jobName);
75
+ const environment = { ...process.env };
76
+ delete environment.OPENAI_API_KEY;
77
+ environment.CODEX_FORCE_AUTH_JSON = "1";
78
+ environment.CODEX_AUTH_JSON_PATH = input.authPath;
79
+ let result;
80
+ try {
81
+ result = await runCommand(input.harborPath, [
82
+ "run",
83
+ "--path",
84
+ input.taskDirectory,
85
+ "--agent",
86
+ "codex",
87
+ "--model",
88
+ input.model,
89
+ "--ak",
90
+ `version=${CODEX_VERSION}`,
91
+ "--ak",
92
+ "reasoning_effort=high",
93
+ "--env",
94
+ input.environment,
95
+ "--job-name",
96
+ jobName,
97
+ "--jobs-dir",
98
+ input.jobsDirectory,
99
+ "--n-concurrent",
100
+ "1",
101
+ "--max-retries",
102
+ "1",
103
+ "--delete",
104
+ "--yes",
105
+ "--quiet",
106
+ ], { allowFailure: true, env: environment, timeoutMs: 4 * 60 * 60 * 1000 });
107
+ }
108
+ catch (error) {
109
+ return failedTrial(taskId, input.model, jobName, errorMessage(error));
110
+ }
111
+ const completed = await tryReadHarborJobResult(input.jobsDirectory, jobName);
112
+ if (!completed) {
113
+ const detail = result.stderr.trim() || result.stdout.trim();
114
+ return failedTrial(taskId, input.model, jobName, `Harbor produced no result (exit ${result.exitCode})${detail ? `: ${detail.slice(-1000)}` : ""}`);
115
+ }
116
+ return summarizeResult(taskId, input.model, jobName, completed.trial);
117
+ }
118
+ export function summarizeResult(taskId, model, jobName, result) {
119
+ const trial = isRecord(result) && Array.isArray(result.trial_results) ? result.trial_results[0] : result;
120
+ if (!isRecord(trial)) {
121
+ throw new Error(`invalid Harbor result for ${taskId}/${model}`);
122
+ }
123
+ const rawRewards = isRecord(trial.verifier_result) && isRecord(trial.verifier_result.rewards)
124
+ ? trial.verifier_result.rewards
125
+ : {};
126
+ const rewards = Object.fromEntries(Object.entries(rawRewards).flatMap(([key, value]) => typeof value === "number" ? [[key, value]] : []));
127
+ const exception = exceptionText(trial.exception_info ?? trial.exception);
128
+ return {
129
+ taskId,
130
+ model,
131
+ jobName,
132
+ passed: !exception &&
133
+ Object.keys(rewards).length > 0 &&
134
+ (rewards.reward === undefined
135
+ ? Object.values(rewards).every((reward) => reward >= 1)
136
+ : rewards.reward >= 1),
137
+ rewards,
138
+ ...(exception ? { exception } : {}),
139
+ };
140
+ }
141
+ function failedTrial(taskId, model, jobName, exception) {
142
+ return { taskId, model, jobName, passed: false, rewards: {}, exception };
143
+ }
144
+ function errorMessage(error) {
145
+ return error instanceof Error ? error.message : String(error);
146
+ }
147
+ async function materializeExport(exportPath, tasksDirectory) {
148
+ const existing = await taskDirectories(tasksDirectory);
149
+ if (existing.length > 0) {
150
+ return existing;
151
+ }
152
+ return await withTemporaryDirectory("selfbench-matrix-", async (temporary) => {
153
+ await runCommand("tar", ["-xzf", exportPath, "-C", temporary]);
154
+ const archivesRoot = join(temporary, "tasks");
155
+ const archives = (await readdir(archivesRoot))
156
+ .filter((name) => name.endsWith(".tar.gz"))
157
+ .sort();
158
+ await mkdir(tasksDirectory, { recursive: true });
159
+ for (const archive of archives) {
160
+ const taskId = archive.slice(0, -".tar.gz".length);
161
+ const expanded = join(temporary, `expanded-${taskId}`);
162
+ await mkdir(expanded);
163
+ await runCommand("tar", ["-xzf", join(archivesRoot, archive), "-C", expanded]);
164
+ await cp(join(expanded, "harbor-task"), join(tasksDirectory, taskId), { recursive: true });
165
+ }
166
+ return await taskDirectories(tasksDirectory);
167
+ });
168
+ }
169
+ async function taskDirectories(root) {
170
+ const entries = await readdir(root, { withFileTypes: true }).catch(() => []);
171
+ const paths = [];
172
+ for (const entry of entries) {
173
+ if (!entry.isDirectory()) {
174
+ continue;
175
+ }
176
+ const path = join(root, entry.name);
177
+ if (await readFile(join(path, "task.toml")).catch(() => undefined)) {
178
+ paths.push(path);
179
+ }
180
+ }
181
+ return paths.sort();
182
+ }
183
+ async function assertSubscriptionAuth(path) {
184
+ const parsed = JSON.parse(await readFile(path, "utf8"));
185
+ assertCodexSubscriptionAuth(parsed, path);
186
+ }
187
+ async function withTemporaryDirectory(prefix, action) {
188
+ const root = await mkdtemp(join(tmpdir(), prefix));
189
+ try {
190
+ return await action(root);
191
+ }
192
+ finally {
193
+ await rm(root, { recursive: true, force: true });
194
+ }
195
+ }
196
+ function exceptionText(value) {
197
+ if (typeof value === "string" && value) {
198
+ return value;
199
+ }
200
+ if (isRecord(value)) {
201
+ return typeof value.message === "string" ? value.message : JSON.stringify(value);
202
+ }
203
+ return undefined;
204
+ }
205
+ function isRecord(value) {
206
+ return typeof value === "object" && value !== null && !Array.isArray(value);
207
+ }
208
+ //# sourceMappingURL=evaluate.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"evaluate.js","sourceRoot":"","sources":["../src/evaluate.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AACxF,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAC1C,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpD,OAAO,EAAE,0BAA0B,EAAE,sBAAsB,EAAE,MAAM,qBAAqB,CAAC;AACzF,OAAO,EAAE,WAAW,EAAE,MAAM,eAAe,CAAC;AAC5C,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAC1C,OAAO,EAAE,2BAA2B,EAAE,MAAM,wBAAwB,CAAC;AAErE,MAAM,CAAC,MAAM,aAAa,GAAG,CAAC,aAAa,EAAE,eAAe,EAAE,cAAc,CAAU,CAAC;AAEvF,MAAM,aAAa,GAAG,SAAS,CAAC;AA2BhC,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,OAAsB;IACpD,MAAM,IAAI,GAAG,OAAO,CAAC,OAAO,CAAC,aAAa,CAAC,CAAC;IAC5C,MAAM,cAAc,GAAG,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;IAC3C,MAAM,KAAK,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACvC,MAAM,eAAe,GAAG,MAAM,kBAAkB,CAAC,OAAO,EAAE,cAAc,CAAC,CAAC;IAC1E,MAAM,QAAQ,GAAG,OAAO,CAAC,OAAO,CAAC,QAAQ,IAAI,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,CAAC,CAAC,CAAC;IAClF,MAAM,sBAAsB,CAAC,QAAQ,CAAC,CAAC;IACvC,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,IAAI,aAAa,CAAC;IAC/C,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CAAC,uCAAuC,CAAC,CAAC;IAC3D,CAAC;IACD,MAAM,IAAI,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE,CACpC,eAAe,CAAC,GAAG,CAAC,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,aAAa,EAAE,CAAC,CAAC,CACnE,CAAC;IACF,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,MAAM,SAAS,GAAG,MAAM,WAAW,CAAC,IAAI,EAAE,OAAO,CAAC,WAAW,IAAI,CAAC,EAAE,KAAK,EAAE,IAAI,EAAE,EAAE;QACjF,MAAM,OAAO,GAAG,MAAM,QAAQ,CAAC;YAC7B,aAAa,EAAE,IAAI;YACnB,UAAU,EAAE,OAAO,CAAC,UAAU,IAAI,QAAQ;YAC1C,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,OAAO;YAC3C,QAAQ;YACR,GAAG,IAAI;SACR,CAAC,CAAC;QACH,SAAS,IAAI,CAAC,CAAC;QACf,OAAO,CAAC,eAAe,EAAE,CAAC,OAAO,EAAE,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC;QAC3D,OAAO,OAAO,CAAC;IACjB,CAAC,CAAC,CAAC;IACH,MAAM,SAAS,CACb,IAAI,CAAC,IAAI,EAAE,cAAc,CAAC,EAC1B,GAAG,IAAI,CAAC,SAAS,CACf;QACE,aAAa,EAAE,CAAC;QAChB,KAAK,EAAE,OAAO;QACd,YAAY,EAAE,aAAa;QAC3B,IAAI,EAAE,oBAAoB;QAC1B,eAAe,EAAE,MAAM;QACvB,MAAM;QACN,SAAS,EAAE,eAAe,CAAC,MAAM;QACjC,MAAM,EAAE,SAAS;KAClB,EACD,IAAI,EACJ,CAAC,CACF,IAAI,CACN,CAAC;IACF,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,KAAK,UAAU,kBAAkB,CAC/B,OAAsB,EACtB,0BAAkC;IAElC,IAAI,OAAO,CAAC,UAAU,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;QAC5C,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;IACpE,CAAC;IACD,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;QACtB,MAAM,KAAK,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC;QAChE,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACrB,MAAM,IAAI,KAAK,CAAC,4BAA4B,OAAO,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC;QAC5E,CAAC;QACD,OAAO,KAAK,CAAC;IACf,CAAC;IACD,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;QACxB,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;IACpE,CAAC;IACD,MAAM,KAAK,GAAG,MAAM,iBAAiB,CAAC,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,EAAE,0BAA0B,CAAC,CAAC;IAC/F,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACrB,MAAM,IAAI,KAAK,CAAC,4BAA4B,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,EAAE,CAAC,CAAC;IAC7E,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,KAAK,UAAU,QAAQ,CAAC,KAOvB;IACC,MAAM,MAAM,GAAG,QAAQ,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;IAC7C,MAAM,OAAO,GAAG,GAAG,MAAM,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,CAAC,OAAO,CAAC,eAAe,EAAE,GAAG,CAAC,CAAC;IACvF,MAAM,QAAQ,GAAG,MAAM,sBAAsB,CAAC,KAAK,CAAC,aAAa,EAAE,OAAO,CAAC,CAAC;IAC5E,IAAI,QAAQ,EAAE,CAAC;QACb,OAAO,eAAe,CAAC,MAAM,EAAE,KAAK,CAAC,KAAK,EAAE,OAAO,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAC;IACvE,CAAC;IACD,MAAM,0BAA0B,CAAC,KAAK,CAAC,aAAa,EAAE,OAAO,CAAC,CAAC;IAE/D,MAAM,WAAW,GAAG,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC;IACvC,OAAO,WAAW,CAAC,cAAc,CAAC;IAClC,WAAW,CAAC,qBAAqB,GAAG,GAAG,CAAC;IACxC,WAAW,CAAC,oBAAoB,GAAG,KAAK,CAAC,QAAQ,CAAC;IAClD,IAAI,MAA8C,CAAC;IACnD,IAAI,CAAC;QACH,MAAM,GAAG,MAAM,UAAU,CACvB,KAAK,CAAC,UAAU,EAChB;YACE,KAAK;YACL,QAAQ;YACR,KAAK,CAAC,aAAa;YACnB,SAAS;YACT,OAAO;YACP,SAAS;YACT,KAAK,CAAC,KAAK;YACX,MAAM;YACN,WAAW,aAAa,EAAE;YAC1B,MAAM;YACN,uBAAuB;YACvB,OAAO;YACP,KAAK,CAAC,WAAW;YACjB,YAAY;YACZ,OAAO;YACP,YAAY;YACZ,KAAK,CAAC,aAAa;YACnB,gBAAgB;YAChB,GAAG;YACH,eAAe;YACf,GAAG;YACH,UAAU;YACV,OAAO;YACP,SAAS;SACV,EACD,EAAE,YAAY,EAAE,IAAI,EAAE,GAAG,EAAE,WAAW,EAAE,SAAS,EAAE,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,EAAE,CACxE,CAAC;IACJ,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,OAAO,WAAW,CAAC,MAAM,EAAE,KAAK,CAAC,KAAK,EAAE,OAAO,EAAE,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC;IACxE,CAAC;IACD,MAAM,SAAS,GAAG,MAAM,sBAAsB,CAAC,KAAK,CAAC,aAAa,EAAE,OAAO,CAAC,CAAC;IAC7E,IAAI,CAAC,SAAS,EAAE,CAAC;QACf,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,IAAI,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;QAC5D,OAAO,WAAW,CAChB,MAAM,EACN,KAAK,CAAC,KAAK,EACX,OAAO,EACP,mCAAmC,MAAM,CAAC,QAAQ,IAAI,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CACjG,CAAC;IACJ,CAAC;IACD,OAAO,eAAe,CAAC,MAAM,EAAE,KAAK,CAAC,KAAK,EAAE,OAAO,EAAE,SAAS,CAAC,KAAK,CAAC,CAAC;AACxE,CAAC;AAED,MAAM,UAAU,eAAe,CAC7B,MAAc,EACd,KAAkB,EAClB,OAAe,EACf,MAAe;IAEf,MAAM,KAAK,GACT,QAAQ,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;IAC7F,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QACrB,MAAM,IAAI,KAAK,CAAC,6BAA6B,MAAM,IAAI,KAAK,EAAE,CAAC,CAAC;IAClE,CAAC;IACD,MAAM,UAAU,GACd,QAAQ,CAAC,KAAK,CAAC,eAAe,CAAC,IAAI,QAAQ,CAAC,KAAK,CAAC,eAAe,CAAC,OAAO,CAAC;QACxE,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC,OAAO;QAC/B,CAAC,CAAC,EAAE,CAAC;IACT,MAAM,OAAO,GAAG,MAAM,CAAC,WAAW,CAChC,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAClD,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAU,CAAC,CAAC,CAAC,CAAC,EAAE,CACzD,CACF,CAAC;IACF,MAAM,SAAS,GAAG,aAAa,CAAC,KAAK,CAAC,cAAc,IAAI,KAAK,CAAC,SAAS,CAAC,CAAC;IACzE,OAAO;QACL,MAAM;QACN,KAAK;QACL,OAAO;QACP,MAAM,EACJ,CAAC,SAAS;YACV,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC;YAC/B,CAAC,OAAO,CAAC,MAAM,KAAK,SAAS;gBAC3B,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,IAAI,CAAC,CAAC;gBACvD,CAAC,CAAC,OAAO,CAAC,MAAM,IAAI,CAAC,CAAC;QAC1B,OAAO;QACP,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACpC,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAClB,MAAc,EACd,KAAkB,EAClB,OAAe,EACf,SAAiB;IAEjB,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE,EAAE,SAAS,EAAE,CAAC;AAC3E,CAAC;AAED,SAAS,YAAY,CAAC,KAAc;IAClC,OAAO,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAChE,CAAC;AAED,KAAK,UAAU,iBAAiB,CAAC,UAAkB,EAAE,cAAsB;IACzE,MAAM,QAAQ,GAAG,MAAM,eAAe,CAAC,cAAc,CAAC,CAAC;IACvD,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,OAAO,QAAQ,CAAC;IAClB,CAAC;IACD,OAAO,MAAM,sBAAsB,CAAC,mBAAmB,EAAE,KAAK,EAAE,SAAS,EAAE,EAAE;QAC3E,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,MAAM,EAAE,UAAU,EAAE,IAAI,EAAE,SAAS,CAAC,CAAC,CAAC;QAC/D,MAAM,YAAY,GAAG,IAAI,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;QAC9C,MAAM,QAAQ,GAAG,CAAC,MAAM,OAAO,CAAC,YAAY,CAAC,CAAC;aAC3C,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;aAC1C,IAAI,EAAE,CAAC;QACV,MAAM,KAAK,CAAC,cAAc,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACjD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,MAAM,MAAM,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;YACnD,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,EAAE,YAAY,MAAM,EAAE,CAAC,CAAC;YACvD,MAAM,KAAK,CAAC,QAAQ,CAAC,CAAC;YACtB,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,MAAM,EAAE,IAAI,CAAC,YAAY,EAAE,OAAO,CAAC,EAAE,IAAI,EAAE,QAAQ,CAAC,CAAC,CAAC;YAC/E,MAAM,EAAE,CAAC,IAAI,CAAC,QAAQ,EAAE,aAAa,CAAC,EAAE,IAAI,CAAC,cAAc,EAAE,MAAM,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC7F,CAAC;QACD,OAAO,MAAM,eAAe,CAAC,cAAc,CAAC,CAAC;IAC/C,CAAC,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,eAAe,CAAC,IAAY;IACzC,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,IAAI,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,CAAC;IAC7E,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC5B,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE,EAAE,CAAC;YACzB,SAAS;QACX,CAAC;QACD,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;QACpC,IAAI,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,EAAE,WAAW,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,EAAE,CAAC;YACnE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnB,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,EAAE,CAAC;AACtB,CAAC;AAED,KAAK,UAAU,sBAAsB,CAAC,IAAY;IAChD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAY,CAAC;IACnE,2BAA2B,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;AAC5C,CAAC;AAED,KAAK,UAAU,sBAAsB,CACnC,MAAc,EACd,MAAoC;IAEpC,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,MAAM,CAAC,CAAC,CAAC;IACnD,IAAI,CAAC;QACH,OAAO,MAAM,MAAM,CAAC,IAAI,CAAC,CAAC;IAC5B,CAAC;YAAS,CAAC;QACT,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;IACnD,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,KAAc;IACnC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,EAAE,CAAC;QACvC,OAAO,KAAK,CAAC;IACf,CAAC;IACD,IAAI,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QACpB,OAAO,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IACnF,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,QAAQ,CAAC,KAAc;IAC9B,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;AAC9E,CAAC"}
@@ -0,0 +1,3 @@
1
+ import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
2
+ export default function authoringExtension(pi: ExtensionAPI): void;
3
+ //# sourceMappingURL=authoring.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"authoring.d.ts","sourceRoot":"","sources":["../../src/extensions/authoring.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,iCAAiC,CAAC;AA4CpE,MAAM,CAAC,OAAO,UAAU,kBAAkB,CAAC,EAAE,EAAE,YAAY,GAAG,IAAI,CAuCjE"}
@@ -0,0 +1,71 @@
1
+ import { mkdirSync, writeFileSync } from "node:fs";
2
+ import { join } from "node:path";
3
+ import { Type } from "@sinclair/typebox";
4
+ const taskDefinition = Type.Object({
5
+ schemaVersion: Type.Literal(1),
6
+ difficulty: Type.Union([Type.Literal("easy"), Type.Literal("medium"), Type.Literal("hard")]),
7
+ taskId: Type.String({ pattern: "^[A-Za-z0-9][A-Za-z0-9._-]*$" }),
8
+ repo: Type.String({ minLength: 1 }),
9
+ baseCommit: Type.String({ pattern: "^[0-9a-fA-F]{40}$" }),
10
+ workdir: Type.String({ minLength: 1 }),
11
+ setupCommand: Type.String({ minLength: 1 }),
12
+ testCommand: Type.String({ minLength: 1 }),
13
+ failToPass: Type.Array(Type.String({ minLength: 1 }), { minItems: 1 }),
14
+ passToPass: Type.Array(Type.String({ minLength: 1 })),
15
+ testPaths: Type.Array(Type.String({ minLength: 1 }), { minItems: 1 }),
16
+ toolchains: Type.Array(Type.Union([
17
+ Type.Literal("uv"),
18
+ Type.Literal("bun"),
19
+ Type.Literal("go"),
20
+ Type.Literal("node"),
21
+ Type.Literal("python"),
22
+ Type.Literal("rust"),
23
+ ]), { minItems: 1 }),
24
+ sourcePr: Type.Integer({ minimum: 1 }),
25
+ sourceUrl: Type.String({ minLength: 1 }),
26
+ prompt: Type.String({ minLength: 1 }),
27
+ timeouts: Type.Object({
28
+ setupSeconds: Type.Integer({ minimum: 1 }),
29
+ agentSeconds: Type.Integer({ minimum: 1 }),
30
+ testsSeconds: Type.Integer({ minimum: 1 }),
31
+ }),
32
+ resources: Type.Object({
33
+ cpus: Type.Number({ exclusiveMinimum: 0 }),
34
+ memoryMb: Type.Integer({ minimum: 1 }),
35
+ storageMb: Type.Integer({ minimum: 1 }),
36
+ }),
37
+ }, { additionalProperties: false });
38
+ export default function authoringExtension(pi) {
39
+ pi.registerTool({
40
+ name: "submit_task",
41
+ label: "Submit SelfBench task",
42
+ description: "Submit exactly one complete task at the assigned difficulty.",
43
+ parameters: Type.Object({
44
+ definition: taskDefinition,
45
+ testPatch: Type.String({ minLength: 1 }),
46
+ goldPatch: Type.String({ minLength: 1 }),
47
+ }, { additionalProperties: false }),
48
+ async execute(_toolCallId, input) {
49
+ const root = process.env.SELFBENCH_TASK_OUTPUT;
50
+ if (!root) {
51
+ throw new Error("SELFBENCH_TASK_OUTPUT is not configured");
52
+ }
53
+ const { definition, testPatch, goldPatch } = input;
54
+ if (!definition?.taskId || typeof testPatch !== "string" || typeof goldPatch !== "string") {
55
+ throw new Error("submit_task received an incomplete task");
56
+ }
57
+ const directory = join(root, definition.taskId);
58
+ mkdirSync(directory, { recursive: false });
59
+ writeFileSync(join(directory, "definition.json"), `${JSON.stringify(definition, null, 2)}\n`, {
60
+ flag: "wx",
61
+ });
62
+ writeFileSync(join(directory, "test.patch"), testPatch, { flag: "wx" });
63
+ writeFileSync(join(directory, "gold.patch"), goldPatch, { flag: "wx" });
64
+ return {
65
+ content: [{ type: "text", text: `Submitted ${definition.taskId}.` }],
66
+ details: { taskId: definition.taskId },
67
+ };
68
+ },
69
+ });
70
+ }
71
+ //# sourceMappingURL=authoring.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"authoring.js","sourceRoot":"","sources":["../../src/extensions/authoring.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,OAAO,EAAE,IAAI,EAAE,MAAM,mBAAmB,CAAC;AAEzC,MAAM,cAAc,GAAG,IAAI,CAAC,MAAM,CAChC;IACE,aAAa,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC;IAC9B,UAAU,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC;IAC5F,MAAM,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,8BAA8B,EAAE,CAAC;IAChE,IAAI,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;IACnC,UAAU,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,mBAAmB,EAAE,CAAC;IACzD,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;IACtC,YAAY,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;IAC3C,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;IAC1C,UAAU,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;IACtE,UAAU,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC,CAAC;IACrD,SAAS,EAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;IACrE,UAAU,EAAE,IAAI,CAAC,KAAK,CACpB,IAAI,CAAC,KAAK,CAAC;QACT,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC;QAClB,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC;QACnB,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC;QAClB,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC;QACpB,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC;QACtB,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC;KACrB,CAAC,EACF,EAAE,QAAQ,EAAE,CAAC,EAAE,CAChB;IACD,QAAQ,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;IACtC,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;IACxC,MAAM,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;IACrC,QAAQ,EAAE,IAAI,CAAC,MAAM,CAAC;QACpB,YAAY,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;QAC1C,YAAY,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;QAC1C,YAAY,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;KAC3C,CAAC;IACF,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC;QACrB,IAAI,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,gBAAgB,EAAE,CAAC,EAAE,CAAC;QAC1C,QAAQ,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;QACtC,SAAS,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;KACxC,CAAC;CACH,EACD,EAAE,oBAAoB,EAAE,KAAK,EAAE,CAChC,CAAC;AAEF,MAAM,CAAC,OAAO,UAAU,kBAAkB,CAAC,EAAgB;IACzD,EAAE,CAAC,YAAY,CAAC;QACd,IAAI,EAAE,aAAa;QACnB,KAAK,EAAE,uBAAuB;QAC9B,WAAW,EAAE,8DAA8D;QAC3E,UAAU,EAAE,IAAI,CAAC,MAAM,CACrB;YACE,UAAU,EAAE,cAAc;YAC1B,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;YACxC,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;SACzC,EACD,EAAE,oBAAoB,EAAE,KAAK,EAAE,CAChC;QACD,KAAK,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK;YAC9B,MAAM,IAAI,GAAG,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC;YAC/C,IAAI,CAAC,IAAI,EAAE,CAAC;gBACV,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;YAC7D,CAAC;YACD,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,SAAS,EAAE,GAAG,KAAK,CAAC;YACnD,IAAI,CAAC,UAAU,EAAE,MAAM,IAAI,OAAO,SAAS,KAAK,QAAQ,IAAI,OAAO,SAAS,KAAK,QAAQ,EAAE,CAAC;gBAC1F,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;YAC7D,CAAC;YACD,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,EAAE,UAAU,CAAC,MAAM,CAAC,CAAC;YAChD,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC,CAAC;YAC3C,aAAa,CACX,IAAI,CAAC,SAAS,EAAE,iBAAiB,CAAC,EAClC,GAAG,IAAI,CAAC,SAAS,CAAC,UAAU,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAC1C;gBACE,IAAI,EAAE,IAAI;aACX,CACF,CAAC;YACF,aAAa,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;YACxE,aAAa,CAAC,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;YACxE,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,aAAa,UAAU,CAAC,MAAM,GAAG,EAAE,CAAC;gBACpE,OAAO,EAAE,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,EAAE;aACvC,CAAC;QACJ,CAAC;KACF,CAAC,CAAC;AACL,CAAC"}
@@ -0,0 +1,3 @@
1
+ import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
2
+ export default function discoveryExtension(pi: ExtensionAPI): void;
3
+ //# sourceMappingURL=discovery.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"discovery.d.ts","sourceRoot":"","sources":["../../src/extensions/discovery.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,iCAAiC,CAAC;AAGpE,MAAM,CAAC,OAAO,UAAU,kBAAkB,CAAC,EAAE,EAAE,YAAY,GAAG,IAAI,CAwDjE"}
@@ -0,0 +1,51 @@
1
+ import { writeFileSync } from "node:fs";
2
+ import { Type } from "@sinclair/typebox";
3
+ export default function discoveryExtension(pi) {
4
+ pi.registerTool({
5
+ name: "submit_discovery",
6
+ label: "Submit SelfBench discovery",
7
+ description: "Submit the final ranked tiered candidate list exactly once.",
8
+ parameters: Type.Object({
9
+ candidates: Type.Array(Type.Object({
10
+ candidateId: Type.String({ minLength: 1 }),
11
+ difficulty: Type.Union([
12
+ Type.Literal("easy"),
13
+ Type.Literal("medium"),
14
+ Type.Literal("hard"),
15
+ ]),
16
+ sourcePr: Type.Integer({ minimum: 1 }),
17
+ sourceUrl: Type.String({ minLength: 1 }),
18
+ baseCommit: Type.String({ pattern: "^[0-9a-fA-F]{40}$" }),
19
+ completedCommit: Type.String({ pattern: "^[0-9a-fA-F]{40}$" }),
20
+ request: Type.String({ minLength: 1 }),
21
+ provenance: Type.Object({
22
+ sourceType: Type.Union([
23
+ Type.Literal("pi"),
24
+ Type.Literal("claude-code"),
25
+ Type.Literal("codex"),
26
+ Type.Literal("generic"),
27
+ Type.Literal("github-pull-request"),
28
+ ]),
29
+ sessionId: Type.String({ minLength: 1 }),
30
+ messageIndex: Type.Integer({ minimum: 0 }),
31
+ }),
32
+ }, { additionalProperties: false }), { maxItems: 100 }),
33
+ }, { additionalProperties: false }),
34
+ async execute(_toolCallId, input) {
35
+ const output = process.env.SELFBENCH_DISCOVERY_OUTPUT;
36
+ if (!output) {
37
+ throw new Error("SELFBENCH_DISCOVERY_OUTPUT is not configured");
38
+ }
39
+ const candidates = input.candidates;
40
+ if (!candidates) {
41
+ throw new Error("submit_discovery received no candidates");
42
+ }
43
+ writeFileSync(output, `${JSON.stringify({ candidates }, null, 2)}\n`, { flag: "wx" });
44
+ return {
45
+ content: [{ type: "text", text: `Submitted ${candidates.length} candidates.` }],
46
+ details: { candidates },
47
+ };
48
+ },
49
+ });
50
+ }
51
+ //# sourceMappingURL=discovery.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"discovery.js","sourceRoot":"","sources":["../../src/extensions/discovery.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAExC,OAAO,EAAE,IAAI,EAAE,MAAM,mBAAmB,CAAC;AAEzC,MAAM,CAAC,OAAO,UAAU,kBAAkB,CAAC,EAAgB;IACzD,EAAE,CAAC,YAAY,CAAC;QACd,IAAI,EAAE,kBAAkB;QACxB,KAAK,EAAE,4BAA4B;QACnC,WAAW,EAAE,6DAA6D;QAC1E,UAAU,EAAE,IAAI,CAAC,MAAM,CACrB;YACE,UAAU,EAAE,IAAI,CAAC,KAAK,CACpB,IAAI,CAAC,MAAM,CACT;gBACE,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;gBAC1C,UAAU,EAAE,IAAI,CAAC,KAAK,CAAC;oBACrB,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC;oBACpB,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC;oBACtB,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC;iBACrB,CAAC;gBACF,QAAQ,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;gBACtC,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;gBACxC,UAAU,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,mBAAmB,EAAE,CAAC;gBACzD,eAAe,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,OAAO,EAAE,mBAAmB,EAAE,CAAC;gBAC9D,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;gBACtC,UAAU,EAAE,IAAI,CAAC,MAAM,CAAC;oBACtB,UAAU,EAAE,IAAI,CAAC,KAAK,CAAC;wBACrB,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC;wBAClB,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC;wBAC3B,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC;wBACrB,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC;wBACvB,IAAI,CAAC,OAAO,CAAC,qBAAqB,CAAC;qBACpC,CAAC;oBACF,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;oBACxC,YAAY,EAAE,IAAI,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,CAAC;iBAC3C,CAAC;aACH,EACD,EAAE,oBAAoB,EAAE,KAAK,EAAE,CAChC,EACD,EAAE,QAAQ,EAAE,GAAG,EAAE,CAClB;SACF,EACD,EAAE,oBAAoB,EAAE,KAAK,EAAE,CAChC;QACD,KAAK,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK;YAC9B,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,0BAA0B,CAAC;YACtD,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;YAClE,CAAC;YACD,MAAM,UAAU,GAAG,KAAK,CAAC,UAAU,CAAC;YACpC,IAAI,CAAC,UAAU,EAAE,CAAC;gBAChB,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;YAC7D,CAAC;YACD,aAAa,CAAC,MAAM,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,EAAE,UAAU,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;YACtF,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,aAAa,UAAU,CAAC,MAAM,cAAc,EAAE,CAAC;gBAC/E,OAAO,EAAE,EAAE,UAAU,EAAE;aACxB,CAAC;QACJ,CAAC;KACF,CAAC,CAAC;AACL,CAAC"}