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.
- package/LICENSE +21 -0
- package/README.md +256 -0
- package/dist/activities.d.ts +41 -0
- package/dist/activities.d.ts.map +1 -0
- package/dist/activities.js +823 -0
- package/dist/activities.js.map +1 -0
- package/dist/agent-smoke-main.d.ts +3 -0
- package/dist/agent-smoke-main.d.ts.map +1 -0
- package/dist/agent-smoke-main.js +54 -0
- package/dist/agent-smoke-main.js.map +1 -0
- package/dist/agent-smoke.d.ts +17 -0
- package/dist/agent-smoke.d.ts.map +1 -0
- package/dist/agent-smoke.js +98 -0
- package/dist/agent-smoke.js.map +1 -0
- package/dist/api-main.d.ts +2 -0
- package/dist/api-main.d.ts.map +1 -0
- package/dist/api-main.js +11 -0
- package/dist/api-main.js.map +1 -0
- package/dist/api.d.ts +3 -0
- package/dist/api.d.ts.map +1 -0
- package/dist/api.js +212 -0
- package/dist/api.js.map +1 -0
- package/dist/artifacts.d.ts +32 -0
- package/dist/artifacts.d.ts.map +1 -0
- package/dist/artifacts.js +311 -0
- package/dist/artifacts.js.map +1 -0
- package/dist/audit.d.ts +12 -0
- package/dist/audit.d.ts.map +1 -0
- package/dist/audit.js +69 -0
- package/dist/audit.js.map +1 -0
- package/dist/cli.d.ts +3 -0
- package/dist/cli.d.ts.map +1 -0
- package/dist/cli.js +297 -0
- package/dist/cli.js.map +1 -0
- package/dist/codex-review.d.ts +40 -0
- package/dist/codex-review.d.ts.map +1 -0
- package/dist/codex-review.js +189 -0
- package/dist/codex-review.js.map +1 -0
- package/dist/config.d.ts +35 -0
- package/dist/config.d.ts.map +1 -0
- package/dist/config.js +78 -0
- package/dist/config.js.map +1 -0
- package/dist/contracts.d.ts +223 -0
- package/dist/contracts.d.ts.map +1 -0
- package/dist/contracts.js +103 -0
- package/dist/contracts.js.map +1 -0
- package/dist/coupling.d.ts +45 -0
- package/dist/coupling.d.ts.map +1 -0
- package/dist/coupling.js +159 -0
- package/dist/coupling.js.map +1 -0
- package/dist/docker-executor.d.ts +11 -0
- package/dist/docker-executor.d.ts.map +1 -0
- package/dist/docker-executor.js +103 -0
- package/dist/docker-executor.js.map +1 -0
- package/dist/eval-main.d.ts +3 -0
- package/dist/eval-main.d.ts.map +1 -0
- package/dist/eval-main.js +80 -0
- package/dist/eval-main.js.map +1 -0
- package/dist/evaluate.d.ts +24 -0
- package/dist/evaluate.d.ts.map +1 -0
- package/dist/evaluate.js +208 -0
- package/dist/evaluate.js.map +1 -0
- package/dist/extensions/authoring.d.ts +3 -0
- package/dist/extensions/authoring.d.ts.map +1 -0
- package/dist/extensions/authoring.js +71 -0
- package/dist/extensions/authoring.js.map +1 -0
- package/dist/extensions/discovery.d.ts +3 -0
- package/dist/extensions/discovery.d.ts.map +1 -0
- package/dist/extensions/discovery.js +51 -0
- package/dist/extensions/discovery.js.map +1 -0
- package/dist/extensions/review.d.ts +3 -0
- package/dist/extensions/review.d.ts.map +1 -0
- package/dist/extensions/review.js +44 -0
- package/dist/extensions/review.js.map +1 -0
- package/dist/github.d.ts +3 -0
- package/dist/github.d.ts.map +1 -0
- package/dist/github.js +19 -0
- package/dist/github.js.map +1 -0
- package/dist/harbor-results.d.ts +14 -0
- package/dist/harbor-results.d.ts.map +1 -0
- package/dist/harbor-results.js +105 -0
- package/dist/harbor-results.js.map +1 -0
- package/dist/harbor-task.d.ts +12 -0
- package/dist/harbor-task.d.ts.map +1 -0
- package/dist/harbor-task.js +462 -0
- package/dist/harbor-task.js.map +1 -0
- package/dist/hash.d.ts +2 -0
- package/dist/hash.d.ts.map +1 -0
- package/dist/hash.js +5 -0
- package/dist/hash.js.map +1 -0
- package/dist/modal-auth.d.ts +2 -0
- package/dist/modal-auth.d.ts.map +1 -0
- package/dist/modal-auth.js +9 -0
- package/dist/modal-auth.js.map +1 -0
- package/dist/modal-executor.d.ts +12 -0
- package/dist/modal-executor.d.ts.map +1 -0
- package/dist/modal-executor.js +152 -0
- package/dist/modal-executor.js.map +1 -0
- package/dist/parallel.d.ts +2 -0
- package/dist/parallel.d.ts.map +1 -0
- package/dist/parallel.js +19 -0
- package/dist/parallel.js.map +1 -0
- package/dist/process.d.ts +27 -0
- package/dist/process.d.ts.map +1 -0
- package/dist/process.js +136 -0
- package/dist/process.js.map +1 -0
- package/dist/provenance.d.ts +22 -0
- package/dist/provenance.d.ts.map +1 -0
- package/dist/provenance.js +361 -0
- package/dist/provenance.js.map +1 -0
- package/dist/reaudit-main.d.ts +3 -0
- package/dist/reaudit-main.d.ts.map +1 -0
- package/dist/reaudit-main.js +155 -0
- package/dist/reaudit-main.js.map +1 -0
- package/dist/repair-main.d.ts +3 -0
- package/dist/repair-main.d.ts.map +1 -0
- package/dist/repair-main.js +168 -0
- package/dist/repair-main.js.map +1 -0
- package/dist/repair.d.ts +9 -0
- package/dist/repair.d.ts.map +1 -0
- package/dist/repair.js +46 -0
- package/dist/repair.js.map +1 -0
- package/dist/review/assets/abap-BdImnpbu.js +1 -0
- package/dist/review/assets/actionscript-3-B3316cI-.js +1 -0
- package/dist/review/assets/ada-bCR0ucgS.js +1 -0
- package/dist/review/assets/ahk-CsyLZFj1.js +1 -0
- package/dist/review/assets/ahk2-8Zs4aa1G.js +1 -0
- package/dist/review/assets/andromeeda-C4gqWexZ.js +1 -0
- package/dist/review/assets/angular-html-DA-rfuFy.js +1 -0
- package/dist/review/assets/angular-ts-BrjP3tb8.js +1 -0
- package/dist/review/assets/apache-Pmp26Uib.js +1 -0
- package/dist/review/assets/apex-DhZFqWV2.js +1 -0
- package/dist/review/assets/apl-CORt7UWP.js +1 -0
- package/dist/review/assets/applescript-Co6uUVPk.js +1 -0
- package/dist/review/assets/ara-BRHolxvo.js +1 -0
- package/dist/review/assets/asciidoc-CSVQ5wI8.js +1 -0
- package/dist/review/assets/asm-D_Q5rh1f.js +1 -0
- package/dist/review/assets/astro-HNnZUWAn.js +1 -0
- package/dist/review/assets/aurora-x-D-2ljcwZ.js +1 -0
- package/dist/review/assets/awk-DMzUqQB5.js +1 -0
- package/dist/review/assets/ayu-dark-DYE7WIF3.js +1 -0
- package/dist/review/assets/ayu-light-BA47KaF1.js +1 -0
- package/dist/review/assets/ayu-mirage-32ctXXKs.js +1 -0
- package/dist/review/assets/ballerina-BFfxhgS-.js +1 -0
- package/dist/review/assets/bat-CickPsom.js +1 -0
- package/dist/review/assets/beancount-k_qm7-4y.js +1 -0
- package/dist/review/assets/berry-uYugtg8r.js +1 -0
- package/dist/review/assets/bibtex-CHM0blh-.js +1 -0
- package/dist/review/assets/bicep-Bmn6On1c.js +1 -0
- package/dist/review/assets/bird2-Bx8U0n9b.js +1 -0
- package/dist/review/assets/blade-2xfisSek.js +1 -0
- package/dist/review/assets/bsl-DlhNcFeZ.js +1 -0
- package/dist/review/assets/c-BIGW1oBm.js +1 -0
- package/dist/review/assets/c3-Dp5svz6Z.js +1 -0
- package/dist/review/assets/cadence-Bv_4Rxtq.js +1 -0
- package/dist/review/assets/cairo-KRGpt6FW.js +1 -0
- package/dist/review/assets/catppuccin-frappe-CZL1YF0i.js +1 -0
- package/dist/review/assets/catppuccin-latte-DH-8KZSZ.js +1 -0
- package/dist/review/assets/catppuccin-macchiato-B7yYVSCf.js +1 -0
- package/dist/review/assets/catppuccin-mocha-Ct7hS0mc.js +1 -0
- package/dist/review/assets/chapel-DTp_pixX.js +1 -0
- package/dist/review/assets/clarity-Dn5IMItf.js +1 -0
- package/dist/review/assets/clojure-P80f7IUj.js +1 -0
- package/dist/review/assets/cmake-D1j8_8rp.js +1 -0
- package/dist/review/assets/cobol-nBiQ_Alo.js +1 -0
- package/dist/review/assets/codeowners-Bp6g37R7.js +1 -0
- package/dist/review/assets/codeql-DsOJ9woJ.js +1 -0
- package/dist/review/assets/coffee-Ch7k5sss.js +1 -0
- package/dist/review/assets/common-lisp-Cg-RD9OK.js +1 -0
- package/dist/review/assets/coq-C7JzOVbR.js +1 -0
- package/dist/review/assets/cpp-BMRokrvK.js +1 -0
- package/dist/review/assets/crystal-DGywbUpC.js +1 -0
- package/dist/review/assets/csharp-DSvCPggb.js +1 -0
- package/dist/review/assets/css-CLj8gQPS.js +1 -0
- package/dist/review/assets/csv-fuZLfV_i.js +1 -0
- package/dist/review/assets/cue-D82EKSYY.js +1 -0
- package/dist/review/assets/cypher-COkxafJQ.js +1 -0
- package/dist/review/assets/d-85-TOEBH.js +1 -0
- package/dist/review/assets/dark-plus-C3mMm8J8.js +1 -0
- package/dist/review/assets/dart-bE4Kk8sk.js +1 -0
- package/dist/review/assets/dax-CEL-wOlO.js +1 -0
- package/dist/review/assets/desktop-BmXAJ9_W.js +1 -0
- package/dist/review/assets/diff-D97Zzqfu.js +1 -0
- package/dist/review/assets/docker-BcOcwvcX.js +1 -0
- package/dist/review/assets/dotenv-Da5cRb03.js +1 -0
- package/dist/review/assets/dracula-BzJJZx-M.js +1 -0
- package/dist/review/assets/dracula-soft-BXkSAIEj.js +1 -0
- package/dist/review/assets/dream-maker-BtqSS_iP.js +1 -0
- package/dist/review/assets/edge-FbVlp4U3.js +1 -0
- package/dist/review/assets/elixir-CkH2-t6x.js +1 -0
- package/dist/review/assets/elm-DbKCFpqz.js +1 -0
- package/dist/review/assets/emacs-lisp-C_m_b--Z.js +1 -0
- package/dist/review/assets/erb-DXfck5VN.js +1 -0
- package/dist/review/assets/erlang-DsQrWhSR.js +1 -0
- package/dist/review/assets/everforest-dark-BgDCqdQA.js +1 -0
- package/dist/review/assets/everforest-light-C8M2exoo.js +1 -0
- package/dist/review/assets/fennel-BYunw83y.js +1 -0
- package/dist/review/assets/fish-BvzEVeQv.js +1 -0
- package/dist/review/assets/fluent-C4IJs8-o.js +1 -0
- package/dist/review/assets/fortran-fixed-form-CkoXwp7k.js +1 -0
- package/dist/review/assets/fortran-free-form-BxgE0vQu.js +1 -0
- package/dist/review/assets/fsharp-CXgrBDvD.js +1 -0
- package/dist/review/assets/gdresource-TyuKm33G.js +1 -0
- package/dist/review/assets/gdscript-DqcFQ5yU.js +1 -0
- package/dist/review/assets/gdshader-DkwncUOv.js +1 -0
- package/dist/review/assets/genie-D0YGMca9.js +1 -0
- package/dist/review/assets/gherkin-DyxjwDmM.js +1 -0
- package/dist/review/assets/git-commit-F4YmCXRG.js +1 -0
- package/dist/review/assets/git-rebase-r7XF79zn.js +1 -0
- package/dist/review/assets/github-dark-DHJKELXO.js +1 -0
- package/dist/review/assets/github-dark-default-Cuk6v7N8.js +1 -0
- package/dist/review/assets/github-dark-dimmed-DH5Ifo-i.js +1 -0
- package/dist/review/assets/github-dark-high-contrast-E3gJ1_iC.js +1 -0
- package/dist/review/assets/github-light-DAi9KRSo.js +1 -0
- package/dist/review/assets/github-light-default-D7oLnXFd.js +1 -0
- package/dist/review/assets/github-light-high-contrast-BfjtVDDH.js +1 -0
- package/dist/review/assets/gleam-BspZqrRM.js +1 -0
- package/dist/review/assets/glimmer-js-ByusRIyA.js +1 -0
- package/dist/review/assets/glimmer-ts-BfAWNZQY.js +1 -0
- package/dist/review/assets/glsl-DplSGwfg.js +1 -0
- package/dist/review/assets/gn-n2N0HUVH.js +1 -0
- package/dist/review/assets/gnuplot-DdkO51Og.js +1 -0
- package/dist/review/assets/go-C27-OAKa.js +1 -0
- package/dist/review/assets/graphql-ChdNCCLP.js +1 -0
- package/dist/review/assets/groovy-gcz8RCvz.js +1 -0
- package/dist/review/assets/gruvbox-dark-hard-CFHQjOhq.js +1 -0
- package/dist/review/assets/gruvbox-dark-medium-GsRaNv29.js +1 -0
- package/dist/review/assets/gruvbox-dark-soft-CVdnzihN.js +1 -0
- package/dist/review/assets/gruvbox-light-hard-CH1njM8p.js +1 -0
- package/dist/review/assets/gruvbox-light-medium-DRw_LuNl.js +1 -0
- package/dist/review/assets/gruvbox-light-soft-hJgmCMqR.js +1 -0
- package/dist/review/assets/hack-BWmVpMyf.js +1 -0
- package/dist/review/assets/haml-D5jkg6IW.js +1 -0
- package/dist/review/assets/handlebars-BpdQsYii.js +1 -0
- package/dist/review/assets/haskell-Df6bDoY_.js +1 -0
- package/dist/review/assets/haxe-CfZj7gIn.js +1 -0
- package/dist/review/assets/hcl-BWvSN4gD.js +1 -0
- package/dist/review/assets/hjson-D5-asLiD.js +1 -0
- package/dist/review/assets/hlsl-D3lLCCz7.js +1 -0
- package/dist/review/assets/horizon-BUw7H-hv.js +1 -0
- package/dist/review/assets/horizon-bright-CUuTKBJd.js +1 -0
- package/dist/review/assets/houston-DnULxvSX.js +1 -0
- package/dist/review/assets/html-derivative-DlHx6ybY.js +1 -0
- package/dist/review/assets/html-pp8916En.js +1 -0
- package/dist/review/assets/http-jrhK8wxY.js +1 -0
- package/dist/review/assets/hurl-irOxFIW8.js +1 -0
- package/dist/review/assets/hxml-2-FPmUDs.js +1 -0
- package/dist/review/assets/hy-DFXneXwc.js +1 -0
- package/dist/review/assets/imba-DGztddWO.js +1 -0
- package/dist/review/assets/index-Dq-6DFzI.css +1 -0
- package/dist/review/assets/index-DurW_RD7.js +1567 -0
- package/dist/review/assets/ini-BEwlwnbL.js +1 -0
- package/dist/review/assets/java-CylS5w8V.js +1 -0
- package/dist/review/assets/javascript-wDzz0qaB.js +1 -0
- package/dist/review/assets/jinja-f2NsQr07.js +1 -0
- package/dist/review/assets/jison-wvAkD_A8.js +1 -0
- package/dist/review/assets/json-Cp-IABpG.js +1 -0
- package/dist/review/assets/json5-C9tS-k6U.js +1 -0
- package/dist/review/assets/jsonc-Des-eS-w.js +1 -0
- package/dist/review/assets/jsonl-DcaNXYhu.js +1 -0
- package/dist/review/assets/jsonnet-DFQXde-d.js +1 -0
- package/dist/review/assets/jssm-C2t-YnRu.js +1 -0
- package/dist/review/assets/jsx-g9-lgVsj.js +1 -0
- package/dist/review/assets/julia-5Bft2YPA.js +1 -0
- package/dist/review/assets/just-Cwhn7H3k.js +1 -0
- package/dist/review/assets/kanagawa-dragon-BuwD2xS4.js +1 -0
- package/dist/review/assets/kanagawa-lotus-C3DzagqV.js +1 -0
- package/dist/review/assets/kanagawa-wave-DFGoQZhC.js +1 -0
- package/dist/review/assets/kdl-DV7GczEv.js +1 -0
- package/dist/review/assets/kotlin-BdnUsdx6.js +1 -0
- package/dist/review/assets/kusto-wEQ09or8.js +1 -0
- package/dist/review/assets/laserwave-DUszq2jm.js +1 -0
- package/dist/review/assets/latex-D5pSuvFb.js +1 -0
- package/dist/review/assets/lean-BZvkOJ9d.js +1 -0
- package/dist/review/assets/less-B1dDrJ26.js +1 -0
- package/dist/review/assets/light-plus-B7mTdjB0.js +1 -0
- package/dist/review/assets/liquid-C0sCDyMI.js +1 -0
- package/dist/review/assets/llvm-BZoOZj88.js +1 -0
- package/dist/review/assets/log-2UxHyX5q.js +1 -0
- package/dist/review/assets/logo-BtOb2qkB.js +1 -0
- package/dist/review/assets/lua-BaeVxFsk.js +1 -0
- package/dist/review/assets/luau-BnpPk5vE.js +1 -0
- package/dist/review/assets/make-CHLpvVh8.js +1 -0
- package/dist/review/assets/markdown-Cvjx9yec.js +1 -0
- package/dist/review/assets/marko-DjSrsDqO.js +1 -0
- package/dist/review/assets/material-theme-D5KoaKCx.js +1 -0
- package/dist/review/assets/material-theme-darker-BfHTSMKl.js +1 -0
- package/dist/review/assets/material-theme-lighter-B0m2ddpp.js +1 -0
- package/dist/review/assets/material-theme-ocean-CyktbL80.js +1 -0
- package/dist/review/assets/material-theme-palenight-Csfq5Kiy.js +1 -0
- package/dist/review/assets/matlab-D7o27uSR.js +1 -0
- package/dist/review/assets/mdc-D1_yUvq7.js +1 -0
- package/dist/review/assets/mdx-Cmh6b_Ma.js +1 -0
- package/dist/review/assets/mermaid-CQcHuHx7.js +1 -0
- package/dist/review/assets/min-dark-CafNBF8u.js +1 -0
- package/dist/review/assets/min-light-CTRr51gU.js +1 -0
- package/dist/review/assets/mipsasm-CKIfxQSi.js +1 -0
- package/dist/review/assets/mojo-Blby8dQK.js +1 -0
- package/dist/review/assets/monokai-BRVnQi9A.js +1 -0
- package/dist/review/assets/moonbit-CHtswR0a.js +1 -0
- package/dist/review/assets/move-el3G9tDJ.js +1 -0
- package/dist/review/assets/narrat-DRg8JJMk.js +1 -0
- package/dist/review/assets/nextflow-C-mBbutL.js +1 -0
- package/dist/review/assets/nextflow-groovy-vE_lwT2v.js +1 -0
- package/dist/review/assets/nginx-BpAMiNFr.js +1 -0
- package/dist/review/assets/night-owl-C39BiMTA.js +1 -0
- package/dist/review/assets/night-owl-light-CMTm3GFP.js +1 -0
- package/dist/review/assets/nim-BIad80T-.js +1 -0
- package/dist/review/assets/nix-CwoSXNpI.js +1 -0
- package/dist/review/assets/nord-Ddv68eIx.js +1 -0
- package/dist/review/assets/nsis-BlV79W_Q.js +1 -0
- package/dist/review/assets/nushell-D3jzshHO.js +1 -0
- package/dist/review/assets/objective-c-DXmwc3jG.js +1 -0
- package/dist/review/assets/objective-cpp-CLxacb5B.js +1 -0
- package/dist/review/assets/ocaml-C0hk2d4L.js +1 -0
- package/dist/review/assets/odin-BBf5iR-q.js +1 -0
- package/dist/review/assets/one-dark-pro-DVMEJ2y_.js +1 -0
- package/dist/review/assets/one-light-C3Wv6jpd.js +1 -0
- package/dist/review/assets/openscad-C4EeE6gA.js +1 -0
- package/dist/review/assets/org-DM6o9KBp.js +1 -0
- package/dist/review/assets/pascal-D93ZcfNL.js +1 -0
- package/dist/review/assets/perl-B9cMNwum.js +1 -0
- package/dist/review/assets/php-Csjmro_R.js +1 -0
- package/dist/review/assets/pierre-dark-CyvmCCZW.js +1 -0
- package/dist/review/assets/pierre-dark-protanopia-deuteranopia-Rgc0TwpF.js +1 -0
- package/dist/review/assets/pierre-dark-soft-BHGpRqa4.js +1 -0
- package/dist/review/assets/pierre-dark-tritanopia-Beq2gCRQ.js +1 -0
- package/dist/review/assets/pierre-dark-vibrant-BWBVywrn.js +1 -0
- package/dist/review/assets/pierre-light-480U9XYS.js +1 -0
- package/dist/review/assets/pierre-light-protanopia-deuteranopia-CaVOBURG.js +1 -0
- package/dist/review/assets/pierre-light-soft-CVdyfjmI.js +1 -0
- package/dist/review/assets/pierre-light-tritanopia-B4_gpKOM.js +1 -0
- package/dist/review/assets/pierre-light-vibrant-DdTDNdfJ.js +1 -0
- package/dist/review/assets/pkl-u5AG7uiY.js +1 -0
- package/dist/review/assets/plastic-3e1v2bzS.js +1 -0
- package/dist/review/assets/plsql-ChMvpjG-.js +1 -0
- package/dist/review/assets/po-BTJTHyun.js +1 -0
- package/dist/review/assets/poimandres-CS3Unz2-.js +1 -0
- package/dist/review/assets/polar-C0HS_06l.js +1 -0
- package/dist/review/assets/postcss-CXtECtnM.js +1 -0
- package/dist/review/assets/powerquery-CEu0bR-o.js +1 -0
- package/dist/review/assets/powershell-BmBUJMz7.js +1 -0
- package/dist/review/assets/prisma-Vru482bI.js +1 -0
- package/dist/review/assets/prolog-CbFg5uaA.js +1 -0
- package/dist/review/assets/proto-C7zT0LnQ.js +1 -0
- package/dist/review/assets/pug-DKIMFp6K.js +1 -0
- package/dist/review/assets/puppet-BMWR74SV.js +1 -0
- package/dist/review/assets/purescript-CklMAg4u.js +1 -0
- package/dist/review/assets/python-B6aJPvgy.js +1 -0
- package/dist/review/assets/qml-3beO22l8.js +1 -0
- package/dist/review/assets/qmldir-C8lEn-DE.js +1 -0
- package/dist/review/assets/qss-IeuSbFQv.js +1 -0
- package/dist/review/assets/r-Cf5RLm7j.js +1 -0
- package/dist/review/assets/racket-BqYA7rlc.js +1 -0
- package/dist/review/assets/raku-DXvB9xmW.js +1 -0
- package/dist/review/assets/razor-BjBPvh-w.js +1 -0
- package/dist/review/assets/rbs-CpoqiR4B.js +1 -0
- package/dist/review/assets/red-hvxz__6c.js +1 -0
- package/dist/review/assets/reg-C-SQnVFl.js +1 -0
- package/dist/review/assets/regexp-CDVJQ6XC.js +1 -0
- package/dist/review/assets/rel-C3B-1QV4.js +1 -0
- package/dist/review/assets/riscv-BM1_JUlF.js +1 -0
- package/dist/review/assets/ron-D8l8udqQ.js +1 -0
- package/dist/review/assets/rose-pine-5qJOZa0Y.js +1 -0
- package/dist/review/assets/rose-pine-dawn-zx0QlTCp.js +1 -0
- package/dist/review/assets/rose-pine-moon-C9pEdX9L.js +1 -0
- package/dist/review/assets/rosmsg-BJDFO7_C.js +1 -0
- package/dist/review/assets/rst-bs7f0vWN.js +1 -0
- package/dist/review/assets/ruby-C0TQ7zu5.js +1 -0
- package/dist/review/assets/rust-B1yitclQ.js +1 -0
- package/dist/review/assets/sas-DEy46yEz.js +1 -0
- package/dist/review/assets/sass-Cj5Yp3dK.js +1 -0
- package/dist/review/assets/scala-CqE71os6.js +1 -0
- package/dist/review/assets/scheme-C98Dy4si.js +1 -0
- package/dist/review/assets/scss-D5BDwBP9.js +1 -0
- package/dist/review/assets/sdbl-DVxCFoDh.js +1 -0
- package/dist/review/assets/shaderlab-Dg9Lc6iA.js +1 -0
- package/dist/review/assets/shellscript-Yzrsuije.js +1 -0
- package/dist/review/assets/shellsession-BADoaaVG.js +1 -0
- package/dist/review/assets/slack-dark-BthQWCQV.js +1 -0
- package/dist/review/assets/slack-ochin-DqwNpetd.js +1 -0
- package/dist/review/assets/smalltalk-BOQMe2GC.js +1 -0
- package/dist/review/assets/smithy-cds9vsN8.js +1 -0
- package/dist/review/assets/snazzy-light-Bw305WKR.js +1 -0
- package/dist/review/assets/solarized-dark-CpvCGNkr.js +1 -0
- package/dist/review/assets/solarized-light-Dlz6yCKv.js +1 -0
- package/dist/review/assets/solidity-DijEV5ha.js +1 -0
- package/dist/review/assets/soy-8wufbnw4.js +1 -0
- package/dist/review/assets/sparql-rVzFXLq3.js +1 -0
- package/dist/review/assets/splunk-BtCnVYZw.js +1 -0
- package/dist/review/assets/sql-CRqJ_cUM.js +1 -0
- package/dist/review/assets/ssh-config-_ykCGR6B.js +1 -0
- package/dist/review/assets/stata-DI20mbqo.js +1 -0
- package/dist/review/assets/stylus-BEDo0Tqx.js +1 -0
- package/dist/review/assets/surrealql-Cjom0U5J.js +1 -0
- package/dist/review/assets/svelte-Cy7k_4gC.js +1 -0
- package/dist/review/assets/swift-C2oV4EkX.js +1 -0
- package/dist/review/assets/synthwave-84-CbfX1IO0.js +1 -0
- package/dist/review/assets/system-verilog-0hqHdDBg.js +1 -0
- package/dist/review/assets/systemd-4A_iFExJ.js +1 -0
- package/dist/review/assets/talonscript-CkByrt1z.js +1 -0
- package/dist/review/assets/tasl-QIJgUcNo.js +1 -0
- package/dist/review/assets/tcl-dwOrl1Do.js +1 -0
- package/dist/review/assets/templ-DhtptRzy.js +1 -0
- package/dist/review/assets/terraform-BETggiCN.js +1 -0
- package/dist/review/assets/tex-D96PA37w.js +1 -0
- package/dist/review/assets/tokyo-night-hegEt444.js +1 -0
- package/dist/review/assets/toml-vGWfd6FD.js +1 -0
- package/dist/review/assets/ts-tags-D351s5mN.js +1 -0
- package/dist/review/assets/tsv-B_m7g4N7.js +1 -0
- package/dist/review/assets/tsx-COt5Ahok.js +1 -0
- package/dist/review/assets/turtle-BsS91CYL.js +1 -0
- package/dist/review/assets/twig-27uCiNez.js +1 -0
- package/dist/review/assets/typescript-BPQ3VLAy.js +1 -0
- package/dist/review/assets/typespec-CAFt9gP4.js +1 -0
- package/dist/review/assets/typst-DHCkPAjA.js +1 -0
- package/dist/review/assets/v-BGw2Nkan.js +1 -0
- package/dist/review/assets/vala-CsfeWuGM.js +1 -0
- package/dist/review/assets/vb-Cu-pLBUe.js +1 -0
- package/dist/review/assets/verilog-nZwndyjY.js +1 -0
- package/dist/review/assets/vesper-DRje8inN.js +1 -0
- package/dist/review/assets/vhdl-CeAyd5Ju.js +1 -0
- package/dist/review/assets/viml-CJc9bBzg.js +1 -0
- package/dist/review/assets/vitesse-black-Bkuqu6BP.js +1 -0
- package/dist/review/assets/vitesse-dark-D0r3Knsf.js +1 -0
- package/dist/review/assets/vitesse-light-CVO1_9PV.js +1 -0
- package/dist/review/assets/vue-BqiEGhQt.js +1 -0
- package/dist/review/assets/vue-html-AaS7Mt5G.js +1 -0
- package/dist/review/assets/vue-vine-BoDAl6tE.js +1 -0
- package/dist/review/assets/vyper-CDx5xZoG.js +1 -0
- package/dist/review/assets/wasm-CG6Dc4jp.js +1 -0
- package/dist/review/assets/wasm-MzD3tlZU.js +1 -0
- package/dist/review/assets/wenyan-BV7otONQ.js +1 -0
- package/dist/review/assets/wgsl-Dx-B1_4e.js +1 -0
- package/dist/review/assets/wikitext-BhOHFoWU.js +1 -0
- package/dist/review/assets/wit-5i3qLPDT.js +1 -0
- package/dist/review/assets/wolfram-lXgVvXCa.js +1 -0
- package/dist/review/assets/xml-sdJ4AIDG.js +1 -0
- package/dist/review/assets/xsl-CtQFsRM5.js +1 -0
- package/dist/review/assets/yaml-Buea-lGh.js +1 -0
- package/dist/review/assets/zenscript-DVFEvuxE.js +1 -0
- package/dist/review/assets/zig-VOosw3JB.js +1 -0
- package/dist/review/index.html +15 -0
- package/dist/run-wait.d.ts +13 -0
- package/dist/run-wait.d.ts.map +1 -0
- package/dist/run-wait.js +25 -0
- package/dist/run-wait.js.map +1 -0
- package/dist/sandbox-author.bundle.js +14979 -0
- package/dist/sandbox-author.d.ts +3 -0
- package/dist/sandbox-author.d.ts.map +1 -0
- package/dist/sandbox-author.js +17 -0
- package/dist/sandbox-author.js.map +1 -0
- package/dist/sandbox-repair.bundle.js +314 -0
- package/dist/sandbox-repair.d.ts +3 -0
- package/dist/sandbox-repair.d.ts.map +1 -0
- package/dist/sandbox-repair.js +121 -0
- package/dist/sandbox-repair.js.map +1 -0
- package/dist/sandbox-review.bundle.js +14458 -0
- package/dist/sandbox-review.d.ts +2 -0
- package/dist/sandbox-review.d.ts.map +1 -0
- package/dist/sandbox-review.js +16 -0
- package/dist/sandbox-review.js.map +1 -0
- package/dist/sandbox-validation-repair.bundle.js +14744 -0
- package/dist/sandbox-validation-repair.d.ts +3 -0
- package/dist/sandbox-validation-repair.d.ts.map +1 -0
- package/dist/sandbox-validation-repair.js +134 -0
- package/dist/sandbox-validation-repair.js.map +1 -0
- package/dist/sandbox.d.ts +39 -0
- package/dist/sandbox.d.ts.map +1 -0
- package/dist/sandbox.js +8 -0
- package/dist/sandbox.js.map +1 -0
- package/dist/subscription-auth.d.ts +6 -0
- package/dist/subscription-auth.d.ts.map +1 -0
- package/dist/subscription-auth.js +59 -0
- package/dist/subscription-auth.js.map +1 -0
- package/dist/temporal.d.ts +6 -0
- package/dist/temporal.d.ts.map +1 -0
- package/dist/temporal.js +17 -0
- package/dist/temporal.js.map +1 -0
- package/dist/validate-main.d.ts +3 -0
- package/dist/validate-main.d.ts.map +1 -0
- package/dist/validate-main.js +137 -0
- package/dist/validate-main.js.map +1 -0
- package/dist/validation-repair.d.ts +10 -0
- package/dist/validation-repair.d.ts.map +1 -0
- package/dist/validation-repair.js +72 -0
- package/dist/validation-repair.js.map +1 -0
- package/dist/worker-main.d.ts +2 -0
- package/dist/worker-main.d.ts.map +1 -0
- package/dist/worker-main.js +30 -0
- package/dist/worker-main.js.map +1 -0
- package/dist/workflow.d.ts +6 -0
- package/dist/workflow.d.ts.map +1 -0
- package/dist/workflow.js +428 -0
- package/dist/workflow.js.map +1 -0
- package/package.json +74 -0
- package/src/extensions/authoring.ts +86 -0
- package/src/extensions/discovery.ts +61 -0
- package/src/extensions/review.ts +54 -0
- package/src/skills/selfbench/SKILL.md +119 -0
|
@@ -0,0 +1,6 @@
|
|
|
1
|
+
import type { SelfBenchActivities } from "./activities.js";
|
|
2
|
+
import type { RunRequest, RunResult, RunStatus } from "./contracts.js";
|
|
3
|
+
export declare const statusQuery: import("@temporalio/workflow").QueryDefinition<RunStatus, [], string>;
|
|
4
|
+
export declare function selfBenchRunWorkflow(input: RunRequest): Promise<RunResult>;
|
|
5
|
+
export declare function executeRun(input: RunRequest, activitySet: SelfBenchActivities, installStatusQuery?: (status: () => RunStatus) => void): Promise<RunResult>;
|
|
6
|
+
//# sourceMappingURL=workflow.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"workflow.d.ts","sourceRoot":"","sources":["../src/workflow.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAKV,mBAAmB,EAGpB,MAAM,iBAAiB,CAAC;AACzB,OAAO,KAAK,EAMV,UAAU,EACV,SAAS,EACT,SAAS,EAEV,MAAM,gBAAgB,CAAC;AAExB,eAAO,MAAM,WAAW,uEAAmC,CAAC;AAqD5D,wBAAsB,oBAAoB,CAAC,KAAK,EAAE,UAAU,GAAG,OAAO,CAAC,SAAS,CAAC,CAEhF;AAED,wBAAsB,UAAU,CAC9B,KAAK,EAAE,UAAU,EACjB,WAAW,EAAE,mBAAmB,EAChC,kBAAkB,GAAE,CAAC,MAAM,EAAE,MAAM,SAAS,KAAK,IAAsB,GACtE,OAAO,CAAC,SAAS,CAAC,CAqQpB"}
|
package/dist/workflow.js
ADDED
|
@@ -0,0 +1,428 @@
|
|
|
1
|
+
import { ActivityFailure, ApplicationFailure, defineQuery, isCancellation, proxyActivities, setHandler, } from "@temporalio/workflow";
|
|
2
|
+
export const statusQuery = defineQuery("status");
|
|
3
|
+
const taskActivities = proxyActivities({
|
|
4
|
+
startToCloseTimeout: "7 hours",
|
|
5
|
+
heartbeatTimeout: "10 minutes",
|
|
6
|
+
cancellationType: "WAIT_CANCELLATION_COMPLETED",
|
|
7
|
+
retry: {
|
|
8
|
+
initialInterval: "5 seconds",
|
|
9
|
+
backoffCoefficient: 2,
|
|
10
|
+
maximumInterval: "2 minutes",
|
|
11
|
+
maximumAttempts: 4,
|
|
12
|
+
},
|
|
13
|
+
});
|
|
14
|
+
const validationRepairActivity = proxyActivities({
|
|
15
|
+
startToCloseTimeout: "2 hours",
|
|
16
|
+
heartbeatTimeout: "10 minutes",
|
|
17
|
+
cancellationType: "WAIT_CANCELLATION_COMPLETED",
|
|
18
|
+
retry: { maximumAttempts: 1 },
|
|
19
|
+
});
|
|
20
|
+
const discoveryActivities = proxyActivities({
|
|
21
|
+
startToCloseTimeout: "1 hour",
|
|
22
|
+
heartbeatTimeout: "10 minutes",
|
|
23
|
+
cancellationType: "WAIT_CANCELLATION_COMPLETED",
|
|
24
|
+
retry: {
|
|
25
|
+
initialInterval: "5 seconds",
|
|
26
|
+
backoffCoefficient: 2,
|
|
27
|
+
maximumInterval: "1 minute",
|
|
28
|
+
maximumAttempts: 3,
|
|
29
|
+
},
|
|
30
|
+
});
|
|
31
|
+
const activities = {
|
|
32
|
+
discoverCandidateShard: discoveryActivities.discoverCandidateShard,
|
|
33
|
+
authorCandidate: taskActivities.authorCandidate,
|
|
34
|
+
validateTask: taskActivities.validateTask,
|
|
35
|
+
repairValidationTask: validationRepairActivity.repairValidationTask,
|
|
36
|
+
reviewTask: taskActivities.reviewTask,
|
|
37
|
+
repairTask: taskActivities.repairTask,
|
|
38
|
+
auditTask: taskActivities.auditTask,
|
|
39
|
+
buildExport: taskActivities.buildExport,
|
|
40
|
+
};
|
|
41
|
+
const DISCOVERY_SHARD_COUNT = 8;
|
|
42
|
+
const DISCOVERY_SHARD_OVERFETCH = 3;
|
|
43
|
+
const MAX_CANDIDATES_PER_TIER_PER_SHARD = 8;
|
|
44
|
+
const MAX_DISCOVERED_CANDIDATES = 300;
|
|
45
|
+
export async function selfBenchRunWorkflow(input) {
|
|
46
|
+
return await executeRun(input, activities, (status) => setHandler(statusQuery, () => status()));
|
|
47
|
+
}
|
|
48
|
+
export async function executeRun(input, activitySet, installStatusQuery = () => undefined) {
|
|
49
|
+
const requested = Object.values(input.candidateCounts).reduce((sum, count) => sum + count, 0);
|
|
50
|
+
let status = {
|
|
51
|
+
runId: input.runId,
|
|
52
|
+
phase: "queued",
|
|
53
|
+
requested,
|
|
54
|
+
requestedByDifficulty: input.candidateCounts,
|
|
55
|
+
discovered: 0,
|
|
56
|
+
accepted: 0,
|
|
57
|
+
rejected: 0,
|
|
58
|
+
tasks: [],
|
|
59
|
+
};
|
|
60
|
+
installStatusQuery(() => status);
|
|
61
|
+
const setPhase = (phase) => {
|
|
62
|
+
status = { ...status, phase };
|
|
63
|
+
};
|
|
64
|
+
const setTasks = (tasks) => {
|
|
65
|
+
status = {
|
|
66
|
+
...status,
|
|
67
|
+
tasks,
|
|
68
|
+
accepted: tasks.filter((task) => task.status === "accepted").length,
|
|
69
|
+
rejected: tasks.filter((task) => task.status === "rejected").length,
|
|
70
|
+
};
|
|
71
|
+
};
|
|
72
|
+
const setDiscovered = (discovered) => {
|
|
73
|
+
status = {
|
|
74
|
+
...status,
|
|
75
|
+
discovered,
|
|
76
|
+
...(status.discovery ? { discovery: { ...status.discovery, candidates: discovered } } : {}),
|
|
77
|
+
};
|
|
78
|
+
};
|
|
79
|
+
try {
|
|
80
|
+
setPhase("discovering");
|
|
81
|
+
const updateDiscovery = (progress) => {
|
|
82
|
+
status = { ...status, discovery: progress };
|
|
83
|
+
};
|
|
84
|
+
const candidates = [];
|
|
85
|
+
let discoveryWave = 0;
|
|
86
|
+
while (true) {
|
|
87
|
+
const selected = selectCandidates(candidates, input.candidateCounts);
|
|
88
|
+
const missing = missingCandidateCounts(selected, input.candidateCounts);
|
|
89
|
+
if (Object.values(missing).every((count) => count === 0)) {
|
|
90
|
+
break;
|
|
91
|
+
}
|
|
92
|
+
const capacity = MAX_DISCOVERED_CANDIDATES - candidates.length;
|
|
93
|
+
if (capacity <= 0) {
|
|
94
|
+
setPhase("blocked");
|
|
95
|
+
throw candidatePoolExhausted(selected, input.candidateCounts);
|
|
96
|
+
}
|
|
97
|
+
const additions = await discoverWave(activitySet, input, {
|
|
98
|
+
wave: discoveryWave,
|
|
99
|
+
targetCounts: missing,
|
|
100
|
+
excludedSourcePrs: candidates.map((candidate) => candidate.sourcePr),
|
|
101
|
+
onProgress: updateDiscovery,
|
|
102
|
+
});
|
|
103
|
+
const knownPrs = new Set(candidates.map((candidate) => candidate.sourcePr));
|
|
104
|
+
const uniqueAdditions = additions.filter((candidate) => !knownPrs.has(candidate.sourcePr));
|
|
105
|
+
if (uniqueAdditions.length === 0) {
|
|
106
|
+
setPhase("blocked");
|
|
107
|
+
throw candidatePoolExhausted(selected, input.candidateCounts);
|
|
108
|
+
}
|
|
109
|
+
candidates.push(...uniqueAdditions.slice(0, capacity));
|
|
110
|
+
setDiscovered(candidates.length);
|
|
111
|
+
discoveryWave += 1;
|
|
112
|
+
}
|
|
113
|
+
const selectedCandidates = selectCandidates(candidates, input.candidateCounts);
|
|
114
|
+
const taskProgress = [];
|
|
115
|
+
const acceptedTasks = [];
|
|
116
|
+
const taskIds = new Set();
|
|
117
|
+
const rejectProgress = (progress, reason) => {
|
|
118
|
+
progress.status = "rejected";
|
|
119
|
+
progress.reason = reason;
|
|
120
|
+
setTasks(taskProgress);
|
|
121
|
+
};
|
|
122
|
+
const processCandidate = async (candidate) => {
|
|
123
|
+
const progress = {
|
|
124
|
+
candidateId: candidate.candidateId,
|
|
125
|
+
taskId: candidate.candidateId,
|
|
126
|
+
difficulty: candidate.difficulty,
|
|
127
|
+
status: "authoring",
|
|
128
|
+
};
|
|
129
|
+
taskProgress.push(progress);
|
|
130
|
+
setTasks(taskProgress);
|
|
131
|
+
try {
|
|
132
|
+
const authored = await activitySet.authorCandidate({
|
|
133
|
+
run: input,
|
|
134
|
+
candidate,
|
|
135
|
+
});
|
|
136
|
+
if (authored.kind === "rejected") {
|
|
137
|
+
rejectProgress(progress, authored.reason);
|
|
138
|
+
return;
|
|
139
|
+
}
|
|
140
|
+
let task = authored.task;
|
|
141
|
+
if (taskIds.has(task.taskId)) {
|
|
142
|
+
rejectProgress(progress, `authoring repeated task ID ${task.taskId}`);
|
|
143
|
+
return;
|
|
144
|
+
}
|
|
145
|
+
taskIds.add(task.taskId);
|
|
146
|
+
progress.taskId = task.taskId;
|
|
147
|
+
progress.status = "auditing";
|
|
148
|
+
setTasks(taskProgress);
|
|
149
|
+
const audit = await activitySet.auditTask({ run: input, task });
|
|
150
|
+
if (!audit.accepted) {
|
|
151
|
+
rejectProgress(progress, audit.reason ?? "audit rejected task");
|
|
152
|
+
return;
|
|
153
|
+
}
|
|
154
|
+
progress.status = "validating";
|
|
155
|
+
setTasks(taskProgress);
|
|
156
|
+
let validation = await activitySet.validateTask({
|
|
157
|
+
run: input,
|
|
158
|
+
task,
|
|
159
|
+
});
|
|
160
|
+
if (!validation.accepted) {
|
|
161
|
+
progress.status = "repairing";
|
|
162
|
+
setTasks(taskProgress);
|
|
163
|
+
let repaired;
|
|
164
|
+
try {
|
|
165
|
+
repaired = await activitySet.repairValidationTask({
|
|
166
|
+
run: input,
|
|
167
|
+
task,
|
|
168
|
+
validation,
|
|
169
|
+
});
|
|
170
|
+
}
|
|
171
|
+
catch (error) {
|
|
172
|
+
if (isCancellation(error)) {
|
|
173
|
+
throw error;
|
|
174
|
+
}
|
|
175
|
+
rejectProgress(progress, "validation repair failed after its single activity attempt");
|
|
176
|
+
return;
|
|
177
|
+
}
|
|
178
|
+
if (repaired.kind === "rejected") {
|
|
179
|
+
rejectProgress(progress, repaired.reason);
|
|
180
|
+
return;
|
|
181
|
+
}
|
|
182
|
+
task = repaired.task;
|
|
183
|
+
progress.status = "auditing";
|
|
184
|
+
setTasks(taskProgress);
|
|
185
|
+
const repairAudit = await activitySet.auditTask({
|
|
186
|
+
run: input,
|
|
187
|
+
task,
|
|
188
|
+
});
|
|
189
|
+
if (!repairAudit.accepted) {
|
|
190
|
+
rejectProgress(progress, repairAudit.reason ?? "audit rejected validation repair");
|
|
191
|
+
return;
|
|
192
|
+
}
|
|
193
|
+
progress.status = "validating";
|
|
194
|
+
setTasks(taskProgress);
|
|
195
|
+
validation = await activitySet.validateTask({
|
|
196
|
+
run: input,
|
|
197
|
+
task,
|
|
198
|
+
});
|
|
199
|
+
if (!validation.accepted) {
|
|
200
|
+
rejectProgress(progress, validation.reason ?? "validation rejected repaired harness");
|
|
201
|
+
return;
|
|
202
|
+
}
|
|
203
|
+
}
|
|
204
|
+
progress.status = "reviewing";
|
|
205
|
+
setTasks(taskProgress);
|
|
206
|
+
let review = await activitySet.reviewTask({ run: input, task });
|
|
207
|
+
if (!review.accepted) {
|
|
208
|
+
progress.status = "repairing";
|
|
209
|
+
setTasks(taskProgress);
|
|
210
|
+
let repaired;
|
|
211
|
+
try {
|
|
212
|
+
repaired = await activitySet.repairTask({
|
|
213
|
+
run: input,
|
|
214
|
+
task,
|
|
215
|
+
review: review.report,
|
|
216
|
+
});
|
|
217
|
+
}
|
|
218
|
+
catch (error) {
|
|
219
|
+
if (isCancellation(error)) {
|
|
220
|
+
throw error;
|
|
221
|
+
}
|
|
222
|
+
rejectProgress(progress, "test repair failed after activity retries");
|
|
223
|
+
return;
|
|
224
|
+
}
|
|
225
|
+
if (repaired.kind === "rejected") {
|
|
226
|
+
rejectProgress(progress, repaired.reason);
|
|
227
|
+
return;
|
|
228
|
+
}
|
|
229
|
+
task = repaired.task;
|
|
230
|
+
progress.status = "auditing";
|
|
231
|
+
setTasks(taskProgress);
|
|
232
|
+
const repairAudit = await activitySet.auditTask({
|
|
233
|
+
run: input,
|
|
234
|
+
task,
|
|
235
|
+
});
|
|
236
|
+
if (!repairAudit.accepted) {
|
|
237
|
+
rejectProgress(progress, repairAudit.reason ?? "audit rejected repaired task");
|
|
238
|
+
return;
|
|
239
|
+
}
|
|
240
|
+
progress.status = "validating";
|
|
241
|
+
setTasks(taskProgress);
|
|
242
|
+
const repairValidation = await activitySet.validateTask({
|
|
243
|
+
run: input,
|
|
244
|
+
task,
|
|
245
|
+
});
|
|
246
|
+
if (!repairValidation.accepted) {
|
|
247
|
+
rejectProgress(progress, repairValidation.reason ?? "validation rejected repaired task");
|
|
248
|
+
return;
|
|
249
|
+
}
|
|
250
|
+
progress.status = "reviewing";
|
|
251
|
+
setTasks(taskProgress);
|
|
252
|
+
review = await activitySet.reviewTask({ run: input, task });
|
|
253
|
+
if (!review.accepted) {
|
|
254
|
+
rejectProgress(progress, review.reason ?? "review rejected repaired task");
|
|
255
|
+
return;
|
|
256
|
+
}
|
|
257
|
+
}
|
|
258
|
+
progress.status = "accepted";
|
|
259
|
+
acceptedTasks.push(task);
|
|
260
|
+
setTasks(taskProgress);
|
|
261
|
+
}
|
|
262
|
+
catch (error) {
|
|
263
|
+
if (!isHarborInfrastructureFailure(error)) {
|
|
264
|
+
throw error;
|
|
265
|
+
}
|
|
266
|
+
progress.status = "infrastructure_failed";
|
|
267
|
+
progress.reason = infrastructureFailureMessage(error);
|
|
268
|
+
setTasks(taskProgress);
|
|
269
|
+
}
|
|
270
|
+
};
|
|
271
|
+
setPhase("authoring");
|
|
272
|
+
await Promise.all(selectedCandidates.map(processCandidate));
|
|
273
|
+
setPhase("exporting");
|
|
274
|
+
const exportRef = await activitySet.buildExport({
|
|
275
|
+
run: input,
|
|
276
|
+
tasks: acceptedTasks,
|
|
277
|
+
});
|
|
278
|
+
status = { ...status, phase: "complete", export: exportRef };
|
|
279
|
+
return {
|
|
280
|
+
runId: input.runId,
|
|
281
|
+
export: exportRef,
|
|
282
|
+
acceptedTaskIds: acceptedTasks.map((task) => task.taskId),
|
|
283
|
+
};
|
|
284
|
+
}
|
|
285
|
+
catch (error) {
|
|
286
|
+
if (isCancellation(error)) {
|
|
287
|
+
status = { ...status, phase: "cancelled" };
|
|
288
|
+
}
|
|
289
|
+
else if (status.phase !== "blocked") {
|
|
290
|
+
status = {
|
|
291
|
+
...status,
|
|
292
|
+
phase: "failed",
|
|
293
|
+
error: error instanceof Error ? error.message : String(error),
|
|
294
|
+
};
|
|
295
|
+
}
|
|
296
|
+
throw error;
|
|
297
|
+
}
|
|
298
|
+
}
|
|
299
|
+
async function discoverWave(activitySet, run, options) {
|
|
300
|
+
const targetCounts = Object.fromEntries(["easy", "medium", "hard"].map((difficulty) => [
|
|
301
|
+
difficulty,
|
|
302
|
+
options.targetCounts[difficulty] === 0
|
|
303
|
+
? 0
|
|
304
|
+
: Math.min(MAX_CANDIDATES_PER_TIER_PER_SHARD, Math.ceil(options.targetCounts[difficulty] / DISCOVERY_SHARD_COUNT) +
|
|
305
|
+
DISCOVERY_SHARD_OVERFETCH),
|
|
306
|
+
]));
|
|
307
|
+
let completedShards = 0;
|
|
308
|
+
let failedShards = 0;
|
|
309
|
+
let candidateCount = 0;
|
|
310
|
+
const reportProgress = () => options.onProgress({
|
|
311
|
+
wave: options.wave,
|
|
312
|
+
totalShards: DISCOVERY_SHARD_COUNT,
|
|
313
|
+
completedShards,
|
|
314
|
+
failedShards,
|
|
315
|
+
candidates: candidateCount,
|
|
316
|
+
});
|
|
317
|
+
reportProgress();
|
|
318
|
+
const shards = await Promise.all(Array.from({ length: DISCOVERY_SHARD_COUNT }, async (_unused, shardIndex) => {
|
|
319
|
+
try {
|
|
320
|
+
const result = await activitySet.discoverCandidateShard({
|
|
321
|
+
run,
|
|
322
|
+
wave: options.wave,
|
|
323
|
+
shardIndex,
|
|
324
|
+
shardCount: DISCOVERY_SHARD_COUNT,
|
|
325
|
+
targetCounts,
|
|
326
|
+
excludedSourcePrs: options.excludedSourcePrs,
|
|
327
|
+
});
|
|
328
|
+
completedShards += 1;
|
|
329
|
+
candidateCount += result.candidates.length;
|
|
330
|
+
reportProgress();
|
|
331
|
+
return result;
|
|
332
|
+
}
|
|
333
|
+
catch (error) {
|
|
334
|
+
if (isCancellation(error)) {
|
|
335
|
+
throw error;
|
|
336
|
+
}
|
|
337
|
+
if (isNonRetryableActivityFailure(error)) {
|
|
338
|
+
throw error;
|
|
339
|
+
}
|
|
340
|
+
failedShards += 1;
|
|
341
|
+
reportProgress();
|
|
342
|
+
return { candidates: [], report: undefined };
|
|
343
|
+
}
|
|
344
|
+
}));
|
|
345
|
+
const ranked = interleave(shards.map((shard) => shard.candidates));
|
|
346
|
+
const seenPrs = new Set();
|
|
347
|
+
return ranked.filter((candidate) => {
|
|
348
|
+
if (seenPrs.has(candidate.sourcePr)) {
|
|
349
|
+
return false;
|
|
350
|
+
}
|
|
351
|
+
seenPrs.add(candidate.sourcePr);
|
|
352
|
+
return true;
|
|
353
|
+
});
|
|
354
|
+
}
|
|
355
|
+
function selectCandidates(candidates, counts) {
|
|
356
|
+
const selectedCounts = { easy: 0, medium: 0, hard: 0 };
|
|
357
|
+
return candidates.filter((candidate) => {
|
|
358
|
+
if (selectedCounts[candidate.difficulty] >= counts[candidate.difficulty]) {
|
|
359
|
+
return false;
|
|
360
|
+
}
|
|
361
|
+
selectedCounts[candidate.difficulty] += 1;
|
|
362
|
+
return true;
|
|
363
|
+
});
|
|
364
|
+
}
|
|
365
|
+
function missingCandidateCounts(selected, requested) {
|
|
366
|
+
const actual = { easy: 0, medium: 0, hard: 0 };
|
|
367
|
+
for (const candidate of selected) {
|
|
368
|
+
actual[candidate.difficulty] += 1;
|
|
369
|
+
}
|
|
370
|
+
return {
|
|
371
|
+
easy: requested.easy - actual.easy,
|
|
372
|
+
medium: requested.medium - actual.medium,
|
|
373
|
+
hard: requested.hard - actual.hard,
|
|
374
|
+
};
|
|
375
|
+
}
|
|
376
|
+
function candidatePoolExhausted(selected, requested) {
|
|
377
|
+
const missing = missingCandidateCounts(selected, requested);
|
|
378
|
+
return ApplicationFailure.nonRetryable(`candidate pool exhausted; missing easy=${missing.easy}, medium=${missing.medium}, hard=${missing.hard}`, "CandidatePoolExhausted");
|
|
379
|
+
}
|
|
380
|
+
function infrastructureFailureMessage(error) {
|
|
381
|
+
let cause = error;
|
|
382
|
+
let message = error instanceof Error ? error.message : String(error);
|
|
383
|
+
while (cause instanceof Error) {
|
|
384
|
+
if (cause instanceof ApplicationFailure && cause.type === "HarborInfrastructureFailure") {
|
|
385
|
+
return cause.message;
|
|
386
|
+
}
|
|
387
|
+
message = cause.message;
|
|
388
|
+
cause = cause.cause;
|
|
389
|
+
}
|
|
390
|
+
return message;
|
|
391
|
+
}
|
|
392
|
+
function isHarborInfrastructureFailure(error) {
|
|
393
|
+
let cause = error;
|
|
394
|
+
while (cause instanceof Error) {
|
|
395
|
+
if (cause instanceof ApplicationFailure && cause.type === "HarborInfrastructureFailure") {
|
|
396
|
+
return true;
|
|
397
|
+
}
|
|
398
|
+
cause = cause.cause;
|
|
399
|
+
}
|
|
400
|
+
return false;
|
|
401
|
+
}
|
|
402
|
+
function isNonRetryableActivityFailure(error) {
|
|
403
|
+
if (!(error instanceof ActivityFailure)) {
|
|
404
|
+
return false;
|
|
405
|
+
}
|
|
406
|
+
let cause = error.cause;
|
|
407
|
+
while (cause instanceof Error) {
|
|
408
|
+
if (cause instanceof ApplicationFailure) {
|
|
409
|
+
return cause.nonRetryable === true;
|
|
410
|
+
}
|
|
411
|
+
cause = cause.cause;
|
|
412
|
+
}
|
|
413
|
+
return false;
|
|
414
|
+
}
|
|
415
|
+
function interleave(groups) {
|
|
416
|
+
const output = [];
|
|
417
|
+
const maxLength = Math.max(0, ...groups.map((group) => group.length));
|
|
418
|
+
for (let index = 0; index < maxLength; index += 1) {
|
|
419
|
+
for (const group of groups) {
|
|
420
|
+
const value = group[index];
|
|
421
|
+
if (value !== undefined) {
|
|
422
|
+
output.push(value);
|
|
423
|
+
}
|
|
424
|
+
}
|
|
425
|
+
}
|
|
426
|
+
return output;
|
|
427
|
+
}
|
|
428
|
+
//# sourceMappingURL=workflow.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"workflow.js","sourceRoot":"","sources":["../src/workflow.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,eAAe,EACf,kBAAkB,EAClB,WAAW,EACX,cAAc,EACd,eAAe,EACf,UAAU,GACX,MAAM,sBAAsB,CAAC;AAsB9B,MAAM,CAAC,MAAM,WAAW,GAAG,WAAW,CAAY,QAAQ,CAAC,CAAC;AAE5D,MAAM,cAAc,GAAG,eAAe,CAEpC;IACA,mBAAmB,EAAE,SAAS;IAC9B,gBAAgB,EAAE,YAAY;IAC9B,gBAAgB,EAAE,6BAA6B;IAC/C,KAAK,EAAE;QACL,eAAe,EAAE,WAAW;QAC5B,kBAAkB,EAAE,CAAC;QACrB,eAAe,EAAE,WAAW;QAC5B,eAAe,EAAE,CAAC;KACnB;CACF,CAAC,CAAC;AAEH,MAAM,wBAAwB,GAAG,eAAe,CAC9C;IACE,mBAAmB,EAAE,SAAS;IAC9B,gBAAgB,EAAE,YAAY;IAC9B,gBAAgB,EAAE,6BAA6B;IAC/C,KAAK,EAAE,EAAE,eAAe,EAAE,CAAC,EAAE;CAC9B,CACF,CAAC;AAEF,MAAM,mBAAmB,GAAG,eAAe,CAAsD;IAC/F,mBAAmB,EAAE,QAAQ;IAC7B,gBAAgB,EAAE,YAAY;IAC9B,gBAAgB,EAAE,6BAA6B;IAC/C,KAAK,EAAE;QACL,eAAe,EAAE,WAAW;QAC5B,kBAAkB,EAAE,CAAC;QACrB,eAAe,EAAE,UAAU;QAC3B,eAAe,EAAE,CAAC;KACnB;CACF,CAAC,CAAC;AAEH,MAAM,UAAU,GAAwB;IACtC,sBAAsB,EAAE,mBAAmB,CAAC,sBAAsB;IAClE,eAAe,EAAE,cAAc,CAAC,eAAe;IAC/C,YAAY,EAAE,cAAc,CAAC,YAAY;IACzC,oBAAoB,EAAE,wBAAwB,CAAC,oBAAoB;IACnE,UAAU,EAAE,cAAc,CAAC,UAAU;IACrC,UAAU,EAAE,cAAc,CAAC,UAAU;IACrC,SAAS,EAAE,cAAc,CAAC,SAAS;IACnC,WAAW,EAAE,cAAc,CAAC,WAAW;CACxC,CAAC;AAEF,MAAM,qBAAqB,GAAG,CAAC,CAAC;AAChC,MAAM,yBAAyB,GAAG,CAAC,CAAC;AACpC,MAAM,iCAAiC,GAAG,CAAC,CAAC;AAC5C,MAAM,yBAAyB,GAAG,GAAG,CAAC;AAEtC,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,KAAiB;IAC1D,OAAO,MAAM,UAAU,CAAC,KAAK,EAAE,UAAU,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,UAAU,CAAC,WAAW,EAAE,GAAG,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;AAClG,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,UAAU,CAC9B,KAAiB,EACjB,WAAgC,EAChC,qBAAwD,GAAG,EAAE,CAAC,SAAS;IAEvE,MAAM,SAAS,GAAG,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,EAAE,CAAC,CAAC,CAAC;IAC9F,IAAI,MAAM,GAAc;QACtB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,KAAK,EAAE,QAAQ;QACf,SAAS;QACT,qBAAqB,EAAE,KAAK,CAAC,eAAe;QAC5C,UAAU,EAAE,CAAC;QACb,QAAQ,EAAE,CAAC;QACX,QAAQ,EAAE,CAAC;QACX,KAAK,EAAE,EAAE;KACV,CAAC;IACF,kBAAkB,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC,CAAC;IACjC,MAAM,QAAQ,GAAG,CAAC,KAAyB,EAAQ,EAAE;QACnD,MAAM,GAAG,EAAE,GAAG,MAAM,EAAE,KAAK,EAAE,CAAC;IAChC,CAAC,CAAC;IACF,MAAM,QAAQ,GAAG,CAAC,KAA8B,EAAQ,EAAE;QACxD,MAAM,GAAG;YACP,GAAG,MAAM;YACT,KAAK;YACL,QAAQ,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,UAAU,CAAC,CAAC,MAAM;YACnE,QAAQ,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,UAAU,CAAC,CAAC,MAAM;SACpE,CAAC;IACJ,CAAC,CAAC;IACF,MAAM,aAAa,GAAG,CAAC,UAAkB,EAAQ,EAAE;QACjD,MAAM,GAAG;YACP,GAAG,MAAM;YACT,UAAU;YACV,GAAG,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,EAAE,GAAG,MAAM,CAAC,SAAS,EAAE,UAAU,EAAE,UAAU,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC5F,CAAC;IACJ,CAAC,CAAC;IAEF,IAAI,CAAC;QACH,QAAQ,CAAC,aAAa,CAAC,CAAC;QACxB,MAAM,eAAe,GAAG,CAAC,QAA2B,EAAQ,EAAE;YAC5D,MAAM,GAAG,EAAE,GAAG,MAAM,EAAE,SAAS,EAAE,QAAQ,EAAE,CAAC;QAC9C,CAAC,CAAC;QACF,MAAM,UAAU,GAAgB,EAAE,CAAC;QACnC,IAAI,aAAa,GAAG,CAAC,CAAC;QACtB,OAAO,IAAI,EAAE,CAAC;YACZ,MAAM,QAAQ,GAAG,gBAAgB,CAAC,UAAU,EAAE,KAAK,CAAC,eAAe,CAAC,CAAC;YACrE,MAAM,OAAO,GAAG,sBAAsB,CAAC,QAAQ,EAAE,KAAK,CAAC,eAAe,CAAC,CAAC;YACxE,IAAI,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,CAAC,CAAC,EAAE,CAAC;gBACzD,MAAM;YACR,CAAC;YACD,MAAM,QAAQ,GAAG,yBAAyB,GAAG,UAAU,CAAC,MAAM,CAAC;YAC/D,IAAI,QAAQ,IAAI,CAAC,EAAE,CAAC;gBAClB,QAAQ,CAAC,SAAS,CAAC,CAAC;gBACpB,MAAM,sBAAsB,CAAC,QAAQ,EAAE,KAAK,CAAC,eAAe,CAAC,CAAC;YAChE,CAAC;YACD,MAAM,SAAS,GAAG,MAAM,YAAY,CAAC,WAAW,EAAE,KAAK,EAAE;gBACvD,IAAI,EAAE,aAAa;gBACnB,YAAY,EAAE,OAAO;gBACrB,iBAAiB,EAAE,UAAU,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC;gBACpE,UAAU,EAAE,eAAe;aAC5B,CAAC,CAAC;YACH,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC;YAC5E,MAAM,eAAe,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC;YAC3F,IAAI,eAAe,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACjC,QAAQ,CAAC,SAAS,CAAC,CAAC;gBACpB,MAAM,sBAAsB,CAAC,QAAQ,EAAE,KAAK,CAAC,eAAe,CAAC,CAAC;YAChE,CAAC;YACD,UAAU,CAAC,IAAI,CAAC,GAAG,eAAe,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC;YACvD,aAAa,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC;YACjC,aAAa,IAAI,CAAC,CAAC;QACrB,CAAC;QACD,MAAM,kBAAkB,GAAG,gBAAgB,CAAC,UAAU,EAAE,KAAK,CAAC,eAAe,CAAC,CAAC;QAC/E,MAAM,YAAY,GAAmB,EAAE,CAAC;QACxC,MAAM,aAAa,GAAmB,EAAE,CAAC;QACzC,MAAM,OAAO,GAAG,IAAI,GAAG,EAAU,CAAC;QAElC,MAAM,cAAc,GAAG,CAAC,QAAsB,EAAE,MAAc,EAAQ,EAAE;YACtE,QAAQ,CAAC,MAAM,GAAG,UAAU,CAAC;YAC7B,QAAQ,CAAC,MAAM,GAAG,MAAM,CAAC;YACzB,QAAQ,CAAC,YAAY,CAAC,CAAC;QACzB,CAAC,CAAC;QACF,MAAM,gBAAgB,GAAG,KAAK,EAAE,SAAoB,EAAiB,EAAE;YACrE,MAAM,QAAQ,GAAiB;gBAC7B,WAAW,EAAE,SAAS,CAAC,WAAW;gBAClC,MAAM,EAAE,SAAS,CAAC,WAAW;gBAC7B,UAAU,EAAE,SAAS,CAAC,UAAU;gBAChC,MAAM,EAAE,WAAW;aACpB,CAAC;YACF,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC5B,QAAQ,CAAC,YAAY,CAAC,CAAC;YACvB,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,MAAM,WAAW,CAAC,eAAe,CAAC;oBACjD,GAAG,EAAE,KAAK;oBACV,SAAS;iBACqB,CAAC,CAAC;gBAClC,IAAI,QAAQ,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;oBACjC,cAAc,CAAC,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;oBAC1C,OAAO;gBACT,CAAC;gBACD,IAAI,IAAI,GAAG,QAAQ,CAAC,IAAI,CAAC;gBACzB,IAAI,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;oBAC7B,cAAc,CAAC,QAAQ,EAAE,8BAA8B,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;oBACtE,OAAO;gBACT,CAAC;gBACD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACzB,QAAQ,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC;gBAE9B,QAAQ,CAAC,MAAM,GAAG,UAAU,CAAC;gBAC7B,QAAQ,CAAC,YAAY,CAAC,CAAC;gBACvB,MAAM,KAAK,GAAG,MAAM,WAAW,CAAC,SAAS,CAAC,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAA2B,CAAC,CAAC;gBACzF,IAAI,CAAC,KAAK,CAAC,QAAQ,EAAE,CAAC;oBACpB,cAAc,CAAC,QAAQ,EAAE,KAAK,CAAC,MAAM,IAAI,qBAAqB,CAAC,CAAC;oBAChE,OAAO;gBACT,CAAC;gBAED,QAAQ,CAAC,MAAM,GAAG,YAAY,CAAC;gBAC/B,QAAQ,CAAC,YAAY,CAAC,CAAC;gBACvB,IAAI,UAAU,GAAG,MAAM,WAAW,CAAC,YAAY,CAAC;oBAC9C,GAAG,EAAE,KAAK;oBACV,IAAI;iBACoB,CAAC,CAAC;gBAC5B,IAAI,CAAC,UAAU,CAAC,QAAQ,EAAE,CAAC;oBACzB,QAAQ,CAAC,MAAM,GAAG,WAAW,CAAC;oBAC9B,QAAQ,CAAC,YAAY,CAAC,CAAC;oBACvB,IAAI,QAAuB,CAAC;oBAC5B,IAAI,CAAC;wBACH,QAAQ,GAAG,MAAM,WAAW,CAAC,oBAAoB,CAAC;4BAChD,GAAG,EAAE,KAAK;4BACV,IAAI;4BACJ,UAAU;yBACyB,CAAC,CAAC;oBACzC,CAAC;oBAAC,OAAO,KAAK,EAAE,CAAC;wBACf,IAAI,cAAc,CAAC,KAAK,CAAC,EAAE,CAAC;4BAC1B,MAAM,KAAK,CAAC;wBACd,CAAC;wBACD,cAAc,CAAC,QAAQ,EAAE,4DAA4D,CAAC,CAAC;wBACvF,OAAO;oBACT,CAAC;oBACD,IAAI,QAAQ,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;wBACjC,cAAc,CAAC,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;wBAC1C,OAAO;oBACT,CAAC;oBACD,IAAI,GAAG,QAAQ,CAAC,IAAI,CAAC;oBAErB,QAAQ,CAAC,MAAM,GAAG,UAAU,CAAC;oBAC7B,QAAQ,CAAC,YAAY,CAAC,CAAC;oBACvB,MAAM,WAAW,GAAG,MAAM,WAAW,CAAC,SAAS,CAAC;wBAC9C,GAAG,EAAE,KAAK;wBACV,IAAI;qBACoB,CAAC,CAAC;oBAC5B,IAAI,CAAC,WAAW,CAAC,QAAQ,EAAE,CAAC;wBAC1B,cAAc,CAAC,QAAQ,EAAE,WAAW,CAAC,MAAM,IAAI,kCAAkC,CAAC,CAAC;wBACnF,OAAO;oBACT,CAAC;oBAED,QAAQ,CAAC,MAAM,GAAG,YAAY,CAAC;oBAC/B,QAAQ,CAAC,YAAY,CAAC,CAAC;oBACvB,UAAU,GAAG,MAAM,WAAW,CAAC,YAAY,CAAC;wBAC1C,GAAG,EAAE,KAAK;wBACV,IAAI;qBACoB,CAAC,CAAC;oBAC5B,IAAI,CAAC,UAAU,CAAC,QAAQ,EAAE,CAAC;wBACzB,cAAc,CAAC,QAAQ,EAAE,UAAU,CAAC,MAAM,IAAI,sCAAsC,CAAC,CAAC;wBACtF,OAAO;oBACT,CAAC;gBACH,CAAC;gBAED,QAAQ,CAAC,MAAM,GAAG,WAAW,CAAC;gBAC9B,QAAQ,CAAC,YAAY,CAAC,CAAC;gBACvB,IAAI,MAAM,GAAG,MAAM,WAAW,CAAC,UAAU,CAAC,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAA2B,CAAC,CAAC;gBACzF,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,CAAC;oBACrB,QAAQ,CAAC,MAAM,GAAG,WAAW,CAAC;oBAC9B,QAAQ,CAAC,YAAY,CAAC,CAAC;oBACvB,IAAI,QAAuB,CAAC;oBAC5B,IAAI,CAAC;wBACH,QAAQ,GAAG,MAAM,WAAW,CAAC,UAAU,CAAC;4BACtC,GAAG,EAAE,KAAK;4BACV,IAAI;4BACJ,MAAM,EAAE,MAAM,CAAC,MAAM;yBACI,CAAC,CAAC;oBAC/B,CAAC;oBAAC,OAAO,KAAK,EAAE,CAAC;wBACf,IAAI,cAAc,CAAC,KAAK,CAAC,EAAE,CAAC;4BAC1B,MAAM,KAAK,CAAC;wBACd,CAAC;wBACD,cAAc,CAAC,QAAQ,EAAE,2CAA2C,CAAC,CAAC;wBACtE,OAAO;oBACT,CAAC;oBACD,IAAI,QAAQ,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;wBACjC,cAAc,CAAC,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;wBAC1C,OAAO;oBACT,CAAC;oBACD,IAAI,GAAG,QAAQ,CAAC,IAAI,CAAC;oBAErB,QAAQ,CAAC,MAAM,GAAG,UAAU,CAAC;oBAC7B,QAAQ,CAAC,YAAY,CAAC,CAAC;oBACvB,MAAM,WAAW,GAAG,MAAM,WAAW,CAAC,SAAS,CAAC;wBAC9C,GAAG,EAAE,KAAK;wBACV,IAAI;qBACoB,CAAC,CAAC;oBAC5B,IAAI,CAAC,WAAW,CAAC,QAAQ,EAAE,CAAC;wBAC1B,cAAc,CAAC,QAAQ,EAAE,WAAW,CAAC,MAAM,IAAI,8BAA8B,CAAC,CAAC;wBAC/E,OAAO;oBACT,CAAC;oBAED,QAAQ,CAAC,MAAM,GAAG,YAAY,CAAC;oBAC/B,QAAQ,CAAC,YAAY,CAAC,CAAC;oBACvB,MAAM,gBAAgB,GAAG,MAAM,WAAW,CAAC,YAAY,CAAC;wBACtD,GAAG,EAAE,KAAK;wBACV,IAAI;qBACoB,CAAC,CAAC;oBAC5B,IAAI,CAAC,gBAAgB,CAAC,QAAQ,EAAE,CAAC;wBAC/B,cAAc,CACZ,QAAQ,EACR,gBAAgB,CAAC,MAAM,IAAI,mCAAmC,CAC/D,CAAC;wBACF,OAAO;oBACT,CAAC;oBAED,QAAQ,CAAC,MAAM,GAAG,WAAW,CAAC;oBAC9B,QAAQ,CAAC,YAAY,CAAC,CAAC;oBACvB,MAAM,GAAG,MAAM,WAAW,CAAC,UAAU,CAAC,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAA2B,CAAC,CAAC;oBACrF,IAAI,CAAC,MAAM,CAAC,QAAQ,EAAE,CAAC;wBACrB,cAAc,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,IAAI,+BAA+B,CAAC,CAAC;wBAC3E,OAAO;oBACT,CAAC;gBACH,CAAC;gBAED,QAAQ,CAAC,MAAM,GAAG,UAAU,CAAC;gBAC7B,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;gBACzB,QAAQ,CAAC,YAAY,CAAC,CAAC;YACzB,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,IAAI,CAAC,6BAA6B,CAAC,KAAK,CAAC,EAAE,CAAC;oBAC1C,MAAM,KAAK,CAAC;gBACd,CAAC;gBACD,QAAQ,CAAC,MAAM,GAAG,uBAAuB,CAAC;gBAC1C,QAAQ,CAAC,MAAM,GAAG,4BAA4B,CAAC,KAAK,CAAC,CAAC;gBACtD,QAAQ,CAAC,YAAY,CAAC,CAAC;YACzB,CAAC;QACH,CAAC,CAAC;QAEF,QAAQ,CAAC,WAAW,CAAC,CAAC;QACtB,MAAM,OAAO,CAAC,GAAG,CAAC,kBAAkB,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC,CAAC;QAE5D,QAAQ,CAAC,WAAW,CAAC,CAAC;QACtB,MAAM,SAAS,GAAG,MAAM,WAAW,CAAC,WAAW,CAAC;YAC9C,GAAG,EAAE,KAAK;YACV,KAAK,EAAE,aAAa;SACC,CAAC,CAAC;QACzB,MAAM,GAAG,EAAE,GAAG,MAAM,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;QAC7D,OAAO;YACL,KAAK,EAAE,KAAK,CAAC,KAAK;YAClB,MAAM,EAAE,SAAS;YACjB,eAAe,EAAE,aAAa,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC;SAC1D,CAAC;IACJ,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,cAAc,CAAC,KAAK,CAAC,EAAE,CAAC;YAC1B,MAAM,GAAG,EAAE,GAAG,MAAM,EAAE,KAAK,EAAE,WAAW,EAAE,CAAC;QAC7C,CAAC;aAAM,IAAI,MAAM,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YACtC,MAAM,GAAG;gBACP,GAAG,MAAM;gBACT,KAAK,EAAE,QAAQ;gBACf,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;aAC9D,CAAC;QACJ,CAAC;QACD,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AASD,KAAK,UAAU,YAAY,CACzB,WAAgC,EAChC,GAAe,EACf,OAA6B;IAE7B,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CACpC,CAAC,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAW,CAAC,GAAG,CAAC,CAAC,UAAU,EAAE,EAAE,CAAC;QACxD,UAAU;QACV,OAAO,CAAC,YAAY,CAAC,UAAU,CAAC,KAAK,CAAC;YACpC,CAAC,CAAC,CAAC;YACH,CAAC,CAAC,IAAI,CAAC,GAAG,CACN,iCAAiC,EACjC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC,UAAU,CAAC,GAAG,qBAAqB,CAAC;gBACjE,yBAAyB,CAC5B;KACN,CAAC,CAC2B,CAAC;IAChC,IAAI,eAAe,GAAG,CAAC,CAAC;IACxB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,MAAM,cAAc,GAAG,GAAS,EAAE,CAChC,OAAO,CAAC,UAAU,CAAC;QACjB,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,WAAW,EAAE,qBAAqB;QAClC,eAAe;QACf,YAAY;QACZ,UAAU,EAAE,cAAc;KAC3B,CAAC,CAAC;IACL,cAAc,EAAE,CAAC;IACjB,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,CAC9B,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,qBAAqB,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,EAAE;QAC1E,IAAI,CAAC;YACH,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,sBAAsB,CAAC;gBACtD,GAAG;gBACH,IAAI,EAAE,OAAO,CAAC,IAAI;gBAClB,UAAU;gBACV,UAAU,EAAE,qBAAqB;gBACjC,YAAY;gBACZ,iBAAiB,EAAE,OAAO,CAAC,iBAAiB;aACf,CAAC,CAAC;YACjC,eAAe,IAAI,CAAC,CAAC;YACrB,cAAc,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC;YAC3C,cAAc,EAAE,CAAC;YACjB,OAAO,MAAM,CAAC;QAChB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,cAAc,CAAC,KAAK,CAAC,EAAE,CAAC;gBAC1B,MAAM,KAAK,CAAC;YACd,CAAC;YACD,IAAI,6BAA6B,CAAC,KAAK,CAAC,EAAE,CAAC;gBACzC,MAAM,KAAK,CAAC;YACd,CAAC;YACD,YAAY,IAAI,CAAC,CAAC;YAClB,cAAc,EAAE,CAAC;YACjB,OAAO,EAAE,UAAU,EAAE,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC;QAC/C,CAAC;IACH,CAAC,CAAC,CACH,CAAC;IACF,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC;IACnE,MAAM,OAAO,GAAG,IAAI,GAAG,EAAU,CAAC;IAClC,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,SAAS,EAAE,EAAE;QACjC,IAAI,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,QAAQ,CAAC,EAAE,CAAC;YACpC,OAAO,KAAK,CAAC;QACf,CAAC;QACD,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;QAChC,OAAO,IAAI,CAAC;IACd,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,gBAAgB,CACvB,UAAgC,EAChC,MAAqC;IAErC,MAAM,cAAc,GAA+B,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;IACnF,OAAO,UAAU,CAAC,MAAM,CAAC,CAAC,SAAS,EAAE,EAAE;QACrC,IAAI,cAAc,CAAC,SAAS,CAAC,UAAU,CAAC,IAAI,MAAM,CAAC,SAAS,CAAC,UAAU,CAAC,EAAE,CAAC;YACzE,OAAO,KAAK,CAAC;QACf,CAAC;QACD,cAAc,CAAC,SAAS,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;QAC1C,OAAO,IAAI,CAAC;IACd,CAAC,CAAC,CAAC;AACL,CAAC;AAED,SAAS,sBAAsB,CAC7B,QAA8B,EAC9B,SAAwC;IAExC,MAAM,MAAM,GAA+B,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;IAC3E,KAAK,MAAM,SAAS,IAAI,QAAQ,EAAE,CAAC;QACjC,MAAM,CAAC,SAAS,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IACpC,CAAC;IACD,OAAO;QACL,IAAI,EAAE,SAAS,CAAC,IAAI,GAAG,MAAM,CAAC,IAAI;QAClC,MAAM,EAAE,SAAS,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM;QACxC,IAAI,EAAE,SAAS,CAAC,IAAI,GAAG,MAAM,CAAC,IAAI;KACnC,CAAC;AACJ,CAAC;AAED,SAAS,sBAAsB,CAC7B,QAA8B,EAC9B,SAAwC;IAExC,MAAM,OAAO,GAAG,sBAAsB,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IAC5D,OAAO,kBAAkB,CAAC,YAAY,CACpC,0CAA0C,OAAO,CAAC,IAAI,YAAY,OAAO,CAAC,MAAM,UAAU,OAAO,CAAC,IAAI,EAAE,EACxG,wBAAwB,CACzB,CAAC;AACJ,CAAC;AAED,SAAS,4BAA4B,CAAC,KAAc;IAClD,IAAI,KAAK,GAAG,KAAK,CAAC;IAClB,IAAI,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACrE,OAAO,KAAK,YAAY,KAAK,EAAE,CAAC;QAC9B,IAAI,KAAK,YAAY,kBAAkB,IAAI,KAAK,CAAC,IAAI,KAAK,6BAA6B,EAAE,CAAC;YACxF,OAAO,KAAK,CAAC,OAAO,CAAC;QACvB,CAAC;QACD,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;QACxB,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;IACtB,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,6BAA6B,CAAC,KAAc;IACnD,IAAI,KAAK,GAAG,KAAK,CAAC;IAClB,OAAO,KAAK,YAAY,KAAK,EAAE,CAAC;QAC9B,IAAI,KAAK,YAAY,kBAAkB,IAAI,KAAK,CAAC,IAAI,KAAK,6BAA6B,EAAE,CAAC;YACxF,OAAO,IAAI,CAAC;QACd,CAAC;QACD,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;IACtB,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,6BAA6B,CAAC,KAAc;IACnD,IAAI,CAAC,CAAC,KAAK,YAAY,eAAe,CAAC,EAAE,CAAC;QACxC,OAAO,KAAK,CAAC;IACf,CAAC;IACD,IAAI,KAAK,GAAY,KAAK,CAAC,KAAK,CAAC;IACjC,OAAO,KAAK,YAAY,KAAK,EAAE,CAAC;QAC9B,IAAI,KAAK,YAAY,kBAAkB,EAAE,CAAC;YACxC,OAAO,KAAK,CAAC,YAAY,KAAK,IAAI,CAAC;QACrC,CAAC;QACD,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;IACtB,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,UAAU,CAAI,MAAiC;IACtD,MAAM,MAAM,GAAQ,EAAE,CAAC;IACvB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;IACtE,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,SAAS,EAAE,KAAK,IAAI,CAAC,EAAE,CAAC;QAClD,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;YAC3B,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;gBACxB,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YACrB,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC"}
|
package/package.json
ADDED
|
@@ -0,0 +1,74 @@
|
|
|
1
|
+
{
|
|
2
|
+
"name": "self-bench",
|
|
3
|
+
"version": "0.3.0",
|
|
4
|
+
"description": "Turn completed repository changes into durable, private Harbor evaluations",
|
|
5
|
+
"license": "MIT",
|
|
6
|
+
"repository": {
|
|
7
|
+
"type": "git",
|
|
8
|
+
"url": "https://github.com/mupt-ai/self-bench.git"
|
|
9
|
+
},
|
|
10
|
+
"type": "module",
|
|
11
|
+
"bin": {
|
|
12
|
+
"selfbench": "dist/cli.js",
|
|
13
|
+
"selfbench-agent-smoke": "dist/agent-smoke-main.js",
|
|
14
|
+
"selfbench-eval": "dist/eval-main.js",
|
|
15
|
+
"selfbench-reaudit": "dist/reaudit-main.js",
|
|
16
|
+
"selfbench-repair": "dist/repair-main.js",
|
|
17
|
+
"selfbench-validate": "dist/validate-main.js"
|
|
18
|
+
},
|
|
19
|
+
"files": [
|
|
20
|
+
"dist",
|
|
21
|
+
"src/extensions",
|
|
22
|
+
"src/skills"
|
|
23
|
+
],
|
|
24
|
+
"engines": {
|
|
25
|
+
"node": ">=22"
|
|
26
|
+
},
|
|
27
|
+
"scripts": {
|
|
28
|
+
"build": "bun run clean && bun run build:server && bun run build:review",
|
|
29
|
+
"clean": "node -e \"require('node:fs').rmSync('dist', { recursive: true, force: true })\"",
|
|
30
|
+
"cli": "bun run src/cli.ts",
|
|
31
|
+
"build:review": "vite build --config review/vite.config.ts",
|
|
32
|
+
"build:server": "tsc -p tsconfig.build.json && bun build src/sandbox-author.ts --target=node --outfile=dist/sandbox-author.bundle.js && bun build src/sandbox-review.ts --target=node --outfile=dist/sandbox-review.bundle.js && bun build src/sandbox-repair.ts --target=node --outfile=dist/sandbox-repair.bundle.js && bun build src/sandbox-validation-repair.ts --target=node --outfile=dist/sandbox-validation-repair.bundle.js",
|
|
33
|
+
"check": "biome check . && tsc --noEmit -p tsconfig.json && bun run typecheck:review",
|
|
34
|
+
"dev:api": "tsx src/api-main.ts",
|
|
35
|
+
"dev:review": "vite --config review/vite.config.ts",
|
|
36
|
+
"dev:worker": "tsx src/worker-main.ts",
|
|
37
|
+
"format": "biome format --write .",
|
|
38
|
+
"prepack": "bun run build",
|
|
39
|
+
"start:api": "node dist/api-main.js",
|
|
40
|
+
"start:worker": "node dist/worker-main.js",
|
|
41
|
+
"test": "bun test tests review/src",
|
|
42
|
+
"typecheck:review": "tsc --noEmit -p review/tsconfig.json",
|
|
43
|
+
"validate": "bun run check && bun run test && bun run build"
|
|
44
|
+
},
|
|
45
|
+
"publishConfig": {
|
|
46
|
+
"access": "public"
|
|
47
|
+
},
|
|
48
|
+
"dependencies": {
|
|
49
|
+
"@google-cloud/storage": "^7.17.0",
|
|
50
|
+
"@pierre/diffs": "1.2.12",
|
|
51
|
+
"@temporalio/activity": "^1.13.2",
|
|
52
|
+
"@temporalio/client": "^1.13.2",
|
|
53
|
+
"@temporalio/common": "^1.13.2",
|
|
54
|
+
"@temporalio/worker": "^1.13.2",
|
|
55
|
+
"@temporalio/workflow": "^1.13.2",
|
|
56
|
+
"modal": "^0.9.0",
|
|
57
|
+
"react": "19.2.7",
|
|
58
|
+
"react-dom": "19.2.7",
|
|
59
|
+
"zod": "^4.1.12"
|
|
60
|
+
},
|
|
61
|
+
"devDependencies": {
|
|
62
|
+
"@biomejs/biome": "^2.2.6",
|
|
63
|
+
"@earendil-works/pi-coding-agent": "0.84.0",
|
|
64
|
+
"@sinclair/typebox": "0.34.41",
|
|
65
|
+
"@types/bun": "^1.3.14",
|
|
66
|
+
"@types/node": "24.10.13",
|
|
67
|
+
"@types/react": "19.2.14",
|
|
68
|
+
"@types/react-dom": "19.2.3",
|
|
69
|
+
"@vitejs/plugin-react": "4.7.0",
|
|
70
|
+
"tsx": "^4.20.6",
|
|
71
|
+
"typescript": "5.9.3",
|
|
72
|
+
"vite": "7.3.1"
|
|
73
|
+
}
|
|
74
|
+
}
|
|
@@ -0,0 +1,86 @@
|
|
|
1
|
+
import { mkdirSync, writeFileSync } from "node:fs";
|
|
2
|
+
import { join } from "node:path";
|
|
3
|
+
import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
|
|
4
|
+
import { Type } from "@sinclair/typebox";
|
|
5
|
+
|
|
6
|
+
const taskDefinition = Type.Object(
|
|
7
|
+
{
|
|
8
|
+
schemaVersion: Type.Literal(1),
|
|
9
|
+
difficulty: Type.Union([Type.Literal("easy"), Type.Literal("medium"), Type.Literal("hard")]),
|
|
10
|
+
taskId: Type.String({ pattern: "^[A-Za-z0-9][A-Za-z0-9._-]*$" }),
|
|
11
|
+
repo: Type.String({ minLength: 1 }),
|
|
12
|
+
baseCommit: Type.String({ pattern: "^[0-9a-fA-F]{40}$" }),
|
|
13
|
+
workdir: Type.String({ minLength: 1 }),
|
|
14
|
+
setupCommand: Type.String({ minLength: 1 }),
|
|
15
|
+
testCommand: Type.String({ minLength: 1 }),
|
|
16
|
+
failToPass: Type.Array(Type.String({ minLength: 1 }), { minItems: 1 }),
|
|
17
|
+
passToPass: Type.Array(Type.String({ minLength: 1 })),
|
|
18
|
+
testPaths: Type.Array(Type.String({ minLength: 1 }), { minItems: 1 }),
|
|
19
|
+
toolchains: Type.Array(
|
|
20
|
+
Type.Union([
|
|
21
|
+
Type.Literal("uv"),
|
|
22
|
+
Type.Literal("bun"),
|
|
23
|
+
Type.Literal("go"),
|
|
24
|
+
Type.Literal("node"),
|
|
25
|
+
Type.Literal("python"),
|
|
26
|
+
Type.Literal("rust"),
|
|
27
|
+
]),
|
|
28
|
+
{ minItems: 1 },
|
|
29
|
+
),
|
|
30
|
+
sourcePr: Type.Integer({ minimum: 1 }),
|
|
31
|
+
sourceUrl: Type.String({ minLength: 1 }),
|
|
32
|
+
prompt: Type.String({ minLength: 1 }),
|
|
33
|
+
timeouts: Type.Object({
|
|
34
|
+
setupSeconds: Type.Integer({ minimum: 1 }),
|
|
35
|
+
agentSeconds: Type.Integer({ minimum: 1 }),
|
|
36
|
+
testsSeconds: Type.Integer({ minimum: 1 }),
|
|
37
|
+
}),
|
|
38
|
+
resources: Type.Object({
|
|
39
|
+
cpus: Type.Number({ exclusiveMinimum: 0 }),
|
|
40
|
+
memoryMb: Type.Integer({ minimum: 1 }),
|
|
41
|
+
storageMb: Type.Integer({ minimum: 1 }),
|
|
42
|
+
}),
|
|
43
|
+
},
|
|
44
|
+
{ additionalProperties: false },
|
|
45
|
+
);
|
|
46
|
+
|
|
47
|
+
export default function authoringExtension(pi: ExtensionAPI): void {
|
|
48
|
+
pi.registerTool({
|
|
49
|
+
name: "submit_task",
|
|
50
|
+
label: "Submit SelfBench task",
|
|
51
|
+
description: "Submit exactly one complete task at the assigned difficulty.",
|
|
52
|
+
parameters: Type.Object(
|
|
53
|
+
{
|
|
54
|
+
definition: taskDefinition,
|
|
55
|
+
testPatch: Type.String({ minLength: 1 }),
|
|
56
|
+
goldPatch: Type.String({ minLength: 1 }),
|
|
57
|
+
},
|
|
58
|
+
{ additionalProperties: false },
|
|
59
|
+
),
|
|
60
|
+
async execute(_toolCallId, input) {
|
|
61
|
+
const root = process.env.SELFBENCH_TASK_OUTPUT;
|
|
62
|
+
if (!root) {
|
|
63
|
+
throw new Error("SELFBENCH_TASK_OUTPUT is not configured");
|
|
64
|
+
}
|
|
65
|
+
const { definition, testPatch, goldPatch } = input;
|
|
66
|
+
if (!definition?.taskId || typeof testPatch !== "string" || typeof goldPatch !== "string") {
|
|
67
|
+
throw new Error("submit_task received an incomplete task");
|
|
68
|
+
}
|
|
69
|
+
const directory = join(root, definition.taskId);
|
|
70
|
+
mkdirSync(directory, { recursive: false });
|
|
71
|
+
writeFileSync(
|
|
72
|
+
join(directory, "definition.json"),
|
|
73
|
+
`${JSON.stringify(definition, null, 2)}\n`,
|
|
74
|
+
{
|
|
75
|
+
flag: "wx",
|
|
76
|
+
},
|
|
77
|
+
);
|
|
78
|
+
writeFileSync(join(directory, "test.patch"), testPatch, { flag: "wx" });
|
|
79
|
+
writeFileSync(join(directory, "gold.patch"), goldPatch, { flag: "wx" });
|
|
80
|
+
return {
|
|
81
|
+
content: [{ type: "text", text: `Submitted ${definition.taskId}.` }],
|
|
82
|
+
details: { taskId: definition.taskId },
|
|
83
|
+
};
|
|
84
|
+
},
|
|
85
|
+
});
|
|
86
|
+
}
|