@leostera/evalkit 0.9.0 → 0.11.0
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/CHANGELOG.md +114 -1
- package/README.md +3 -2
- package/dist/cli.mjs +311 -92
- package/dist/dashboard/assets/index-CFJzcw3o.css +2 -0
- package/dist/dashboard/assets/{index-By1SJ1on.js → index-DgoX2x_M.js} +34 -31
- package/dist/dashboard/index.html +2 -2
- package/dist/{index-db-BbVblaHz.mjs → index-db-DI7xKMSX.mjs} +237 -47
- package/dist/index.d.mts +1141 -220
- package/dist/index.d.mts.map +1 -1
- package/dist/index.mjs +241 -52
- package/dist/runner.d.mts +782 -225
- package/dist/runner.d.mts.map +1 -1
- package/dist/runner.mjs +2 -2
- package/dist/share-worker.d.mts +1 -1
- package/dist/share-worker.d.mts.map +1 -1
- package/dist/share-worker.mjs +1 -1
- package/dist/share.d.mts +9 -6
- package/dist/share.d.mts.map +1 -1
- package/dist/share.mjs +2 -2
- package/dist/skills/evalkit/SKILL.md +3 -3
- package/dist/skills/evalkit-agents/SKILL.md +3 -3
- package/dist/skills/evalkit-datasets/SKILL.md +1 -1
- package/dist/skills/evalkit-evals/SKILL.md +1 -1
- package/dist/skills/evalkit-judges/SKILL.md +4 -4
- package/dist/skills/evalkit-results/SKILL.md +2 -2
- package/dist/{src-Bn6fnZnF.mjs → src-CXzbHAKN.mjs} +1 -1
- package/dist/{src-BrfIXfO2.mjs → src-D6Ax2t9b.mjs} +636 -220
- package/dist/src-Dc3GXacB.mjs +3637 -0
- package/dist/upgrade-rMsGzUEJ.mjs +230 -0
- package/dist/upgrade-transcript-C8DUykLd.mjs +402 -0
- package/package.json +2 -1
- package/dist/dashboard/assets/index-geXMXCx9.css +0 -2
- package/dist/src-CcdLZqFk.mjs +0 -1714
- package/dist/upgrade-60SIRius.mjs +0 -110
package/dist/index.d.mts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.mts","names":[],"sources":["../../core/src/identity.ts","../../core/src/dataset.ts","../../core/src/aut-events.ts","../../core/src/catalog.ts","../../core/src/canonical-json.ts","../../core/src/matrix.ts","../../core/src/config.ts","../../core/src/usage.ts","../../core/src/json.ts","../../core/src/definition-site.ts","../../core/src/schema.ts","../../core/src/trajectory.ts","../../core/src/index.ts","../../agents/src/index.ts"],"mappings":";;qBAEa,oBAAkB,OAAA;YACnB,sBAAsB,mBAAmB;;wBAGrC,SAAS;;qBAOZ,YAAU,OAAA,cAAA,OAAA;YAKX,cAAc,WAAW;YAEzB,YAAY,cAAc,eAAe,2BACxC;qBAEA,mBAAiB,OAAA,cAAA,OAAA;wBAMd,kBAAkB,cAAc,cAAc,MAAM,QAAK,OAAA,cAAA,OAAA;wBAWzD,YAAY,cAAc,cACxC,MAAM,OACN,MAAM,OACL,YAAY;wBAIC,iBAAiB,cAAc,cAC7C,eACA,eAAe;EACZ,MAAM;EAAO,MAAM;EAAM,KAAK,YAAY;;wBAY/B,aAAa,eAAe,eAAe,eAAe;;;;YCvD9D;;YAGA,6BAA6B;EAAe;;;;;;YAM5C,wBAAwB,SAAS,eAAe;;YAGhD;EACV,QAAQ;;EAER;;YAGU;;EAEV;EACA,QAAQ;;YAGE;EACV;EACA;;EAEA,QAAQ;;EAER;;EAEA,YAAY;EACZ,QAAQ;;KAGL;EACH;EACA;;EAEA,aAAa;;EAEb;;YAGU,0BAA0B;EACpC,QAAQ;;EAER;;EAEA;EACA;EACA;;YAGU,0BAA0B,gBAAgB,kBACpD;;EAEE,SAAS,MAAM,uBAAuB,UAAU,QAAQ;;EAExD,SACE,QAAQ,SACR,KAAK,sBACF,iBAAiB,QAAQ;EAC9B;EACA;EACA;;YAGQ;EAEN;EACA,QAAQ;EACR;EACA;;EAGA;EACA,SACE,MAAM,uBACH,iBAAiB,QAAQ;EAC9B,SACE,QAAQ,gBACR,KAAK,sBACF,iBAAiB,QAAQ;;;YAIxB;WACD;WACA;WACA;WACA,YAAY;;WAEZ;WACA,QAAQ;;;wBAIH,oBAAoB,eAAe;;;;;;;wBAuBnC,cAAc,gBAAgB,gBAC5C,YAAY,0BAA0B,0BAA0B,WAC/D;wBAmFa,UAAU,iBAAiB,SAAS;;;;;;;YC/MxC;EACN;EAAiB;;EAEjB;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;EACA;;EAEA;EAAsB;EAAc;;EAEpC;EACA;EACA;EACA;EACA,QAAQ;;EAER;EAAmB;EAAkB;;EACrC;EAAe,OAAO;EAAe;;YAE/B;EACV;EACA;;EAEA;EACA;;;YAIU;EAGN;EACA,qBAAqB;EACrB,QAAQ;;;wBAIE,kBACd,OAAO,eACP,WAAW,OACV;;;;;wBA0Da,iBACd,QAAQ,YACR,cACA,WAAW,OACV;;;;;;;;;YCtGS;;EAEV;;YAGU;EACV;EACA;EACA;EACA;EACA,UAAU;IAAQ,MAAM;IAAkB;;;EAE1C;;;YAIU;EACN;EAAe;;EACf;EAAgB;;EAChB;;YAEM;EACV;EACA;EACA,YAAY;EACZ;EACA;;EAEA;IAAU;;IAAwB;;;EAElC;IAAU,OAAO;;IAAiB;;;EAElC;IAAU;IAAc;;EACxB;;YAGU;EACV;EACA,MAAM;EACN;EACA;;EAEA;EACA;;YAGU;;;EAEN;EAAc;;EACd;EAAe,aAAa;;EAC5B;EAA0B;EAAc,YAAY;;;;EAEpD;EAAoB,QAAQ;;YAEtB;EACV;EACA;EACA;EACA;EACA,OAAO;;EAEP,aAAa;EACb,UAAU;;EAEV;EACA;IAAU;IAAoB;;EAC9B,YAAY;EACZ,SAAS;EACT,WAAW;;EAEX;;wBAGc,aACd,YAAY,gBACZ,OAAO,uBACP,UAAS,iBACR;;;;wBC7Fa,oBAAoB,OAAO;;;KCK/B;;EAEV;;EAEA,aAAa,SAAS,wBAAwB;;EAE9C,YAAY;;KAEF;EACV;;EAEA;EACA,MAAM;EACN,YAAY;;KAEF;EACV;EACA;EACA,gBAAgB;EAChB,YAAY,SAAS,wBAAwB;EAC7C,WAAW;;EAEX,iBAAiB;;EAEjB,mBAAmB;;KAET,aAAa;EACvB;EACA,MAAM,YAAY;EAClB,MAAM,YAAY,kBAAkB,iBAAiB;;wBAWvC,uBAAuB,UAAU,sBAC/C,YAAY,IACX,aAAa;;;KCjDJ;;EAEV;EACA;EACA;EACA;;IAEE;;;EAGF,iBAAiB;EACjB,WAAW;;EAEX,SAAS,KAAK;IAA0C;;EACxD;IAAc;IAAsB;IAAiB;;EACrD;;EAEA;;;;;;;IAOE;;IAEA,cAAc;;;EAGhB;;EAEA,gBAAgB;EAChB,QAAQ;;KAGE;KAEA;;EAEV;;EAEA;;qBAGW,2BAA2B,SAAS;;qBAMpC;;KAWD,eAAe;KAEf;;EAEV;;;;;;EAMA,UAAU,sBAAsB,eAAe,QAAQ;;;wBAIzC,mBAAmB,UAAU,eAC3C,eAAe,IACd;;;;;;;KC5ES;;;;;;EAMV;;EAEA;;EAEA;;EAEA;;;;;;;wBAQc,WAAW,OAAO;;;;qBCrBrB,qBAAqB;WAGrB;WACA;WAHO;EAClB,YACW,cACA;;;;;;;;wBAYG,OAAO,gBAAgB,gBAAiB;;wBA4CxC,aACd,gBACA;GACI,cAAc;;;;;;;;;;;KC3DR;EAAmB;EAAc;;wBAqB7B,eAAe,iBAAiB;;;;qBCvBnC,iBAAe,OAAA;;qBASf;qBAEA,qBAAmB,OAAA;;;;;qBAMnB,mBAAiB,OAAA;;;;;;qBAQjB,iBAAiB,OAAO,OAAO;qBAoB/B,iBAAe,OAAA;;;;;;;;;;;;;;qBAiBf,aAAW,OAAA;;;;;qBAMX,kBAAgB,OAAA;;;;;;qBAOhB,mBAAiB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAoBjB,mBAAiB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAKjB,aAAW,OAAA;;;;;;qBAOX,oBAAkB,OAAA;;;;;;;;;;;;;;;qBAalB,mBAAiB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAYjB,wBAAsB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAgBtB,oBAAkB,OAAA;;;;;;;;;;;;qBAQlB,kBAAgB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAUhB,qBAAmB,OAAA;;;;;qBAKnB,uBAAqB,OAAA;;;;;qBAKrB,kBAAgB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAoJhB,uBAAuB,OAAO,OAAO;qBAGrC,gBAAc,OAAA;;;;;;;wBCvUX,qBAAqB,mBAAmB;;;YCmB5C;YACA,YACV,gBAAgB;GAAiB,cAAc;;YACrC;GAAgB,cAAc;;YAE9B;;EAEV;EACA;EACA;EACA;;YAGU;;;;;YAMA;EACV;EACA,gBAAgB;;YAGN,gBAAgB,QAAQ,OAAO,kBAAkB;YAKjD;EACV;EACA;EACA;;YAGU;EACN;EAAiB;;EAEjB;EACA;EACA,SAAS;EACT;;EAGA;EACA;EACA;EACA,WAAW;EACX;;EAGA;EACA;EACA;EACA,QAAQ;EACR;;EAEA;EAAsB;EAAc;;EAEpC;EACA;EACA;EACA;EACA,QAAQ;;EAER;EAAmB,SAAS;EAAW;;EACvC;EAAe,OAAO;EAAe;;YAE/B;EACN;EAAqB;;EAErB;EACA;;EAEA;EACA;;EAEA;EAAmC;EAAc;;EACjD;EAAiC;EAAc;;EAE/C;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;;EAGA;EACA;EACA,OAAO;EACP;;EAEA;EAAwB;EAAgB;;EAExC;EACA;;EAEA;EACA;EACA;;EAGA;EACA;EACA,OAAO;EACP;;EAEA;EAAuB;;EACvB;EAAe,OAAO;EAAe;;YAE/B,mBACT;EAAa;MAAoB;EAAgB;;YAExC;;EAEV;;YAGU,qBAAqB;YACrB,qBAAqB;YAErB;EACV;EACA;;EAEA,QAAQ;EACR,UAAU;;EAEV,aAAa;EACb,UAAU;;EAEV,MAAM;;EAEN,WAAW;;EAEX,oBAAoB;;YAGV,WAAW;WACZ,WAAW;;WAEX,WAAW;EACpB,MAAM;IACJ,SAAS;IACT,UAAU;;IAEV,UAAU,OAAO,yBAAyB;;;;;;;;IAQ1C,SAAS,cAAc,kBAAkB,eAAe;MACtD,QAAQ,WAAW;;YAGb,WAAW;;;;;EAKrB,KAAK,SAAS,WAAW,eAAe;;EAExC,UAAU;;;qBAIC,wBAAwB;;wBAKrB,cAAc,OAAO,kBAAkB;;;;;wBAUvC,kBAAkB,UAAU,YAAY,OAAO,IAAI;;YAMvD;;EAEV,YAAY;;EAEZ,MAAM;;EAEN;;EAEA,QAAQ;;YAGE;EACV;;EAEA,oBAAoB,OAAO;;YAGjB;EACV;EACA,aAAa;;YAGH;EACV;EACA,kBAAkB;IAAa;;EAC/B,6BAA6B,QAAQ;IAAY;;IAC/C;IACA;;EAEF;IACE;IACA;IACA;IACA,WAAW;IACX;IACA,SAAS;;EAEX;;YAGU;EACV;EACA;EACA;IAAY;IAAc,YAAY;;;YAE5B,iBACV,WAAW,YAAY,cAAc;;wBAGvB,eACd,UAAU,iCACT;;;;;wBAUa,KAAK,SAAS,sBAAsB;wBAIpC,MAAM,aAAa,aAAa;YAIpC;EACV,QAAQ;EACR,QAAQ;;EAER,SAAS;;;YAIC;EAAiB;EAAe;;YAEhC,WAAW,+BAEnB;EAEE;EACA;EACA;;;;;EAKA,WAAW;;EAEX,QAAQ;;;;;;wBAOE,KAAK,iBAAiB;YAK1B;EACV;EACA;EACA;EACA;EACA;EACA,WAAW;EACX,QAAQ;EACR;EACA,QAAQ;;EAER;;YAGU;EACV,WAAW;EACX,WAAW;;YAGD;EACV,SAAS;EACT;IAAc,iBAAiB;;EAC/B,WAAW;;EAEX,OAAO;;EAEP,SAAS;;EAET,MAAM;;;;;;EAMN,sBAAsB;;;;;;;EAOtB,WAAW,UACT,MAAM,gBAAgB,YACrB,kBAAkB;EACrB,WACE,MAAM,YAAY,qBACjB,kBAAkB;;;;;;EAMrB,OAAO,cAAc,kBAAkB,aAAa;;;;;;;YAQ1C,kBAAkB,YAAY,KAAK;EAC7C,WAAW;;YAGD,gBAAgB;EAC1B;EACA;EACA;;EAEA;EACA,IACE,SAAS,2BACE,WAAW,YAAY,kBAAkB,WAAW;;YAGvD;EACV;EACA;EACA;;EAEA;;YAGU,cAAc,kBAAkB;;wBAG5B,UAAU,oBACxB,cACA,KAAK,gBAAgB,kBACrB,UAAS,KAAK,iDACb,gBAAgB;;;;;;YAqBP;EACV;EACA;;EAEA;;EAEA;EACA,IACE,SAAS,sCACa;;;wBAIV,aAAa,qBAC3B,MAAM,aAAa,4BAA4B,MAC/C,KAAK,eACL,UAAS,KAAK,6BACb;;YAMS;EACV,MAAM;;EAEN;EACA;EACA;;EAEA;EACA,QAAQ;;wBAGM,MACd,cACA,SAAS,KAAK,2CACb;YAMS;EACV;;EAEA;;YAGU,eAAe,eAAe,aAAa;EACrD;EACA;EACA,OAAO;;EAEP,QAAQ;;EAER,oBAAoB;EACpB,YAAY;EACZ,SAAS;;EAET,mBAAmB;EACnB,SAAS;;;;;EAKT;;EAEA,WAAW;EACX,YAAY;;;YAIF;;;;;;;;;EASV;;wBAGc,iBAAiB,UAAU,gBAAgB,YAAY,IAAI;YAuD/D,UACV,wBAAwB,4BAA4B;EAEpD;EACA;EACA,OAAO;;wBAGO,kBAAkB,eAAe,WAC/C,OAAO,SACN;YAKS,mBAAmB,iBAAiB,YAAY;wBAE5C,YAAY;EAAS;;YAkBzB;;EAEV,gBAAgB;EAChB,iBAAiB;EACjB,mBAAmB;EACnB,IAAI,cAAc;EAClB,SAAS,cAAc;EACvB,YAAY;EACZ,iBAAiB;;EAEjB,QAAQ,UAAU,iBAAiB;;YAGzB;EACV;EACA;EACA;;YAGU;EACV;EACA;EACA;;;;;;wBAOc,oBACR,gCAAgC,oBACtC,eAAe,iBAAiB;;;;;YA2EtB;;YAIA;EACV;;EAEA;EACA;;;;;;YAOU;;EAEV;;EAEA;;EAEA;EACA;EACA;;EAEA;EACA;IAAO;IAAkB;IAAmB;IAAc;;;;YAIhD;EACV,SAAS;EACT,OAAO;EACP,QAAQ;;;YAIE;EAAe;EAAa;;YAE5B;EACV;EACA;EACA,QAAQ;EACR;EACA;EACA,aAAa;EACb;IAAW;IAAY;;EACvB,MAAM;EACN;;EAEA,SAAS;EACT,UAAU;;EAEV,QAAQ;;EAER;;EAEA,WAAW;;YAGD;EACV;EACA,YAAY;EACZ;;EAEA;;YAGU;EACV;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX,QAAQ;EACR;EACA,QAAQ;EACR;IAAoB;IAAoB;;;;;;;;;;;;;qBAa7B;YACD,4BAA4B;;YAE5B,aAAa;YAEb;EACV,SAAS;EACT,UAAU;EACV,cAAc;EACd;EACA;EACA;;YAGU;;EAEV;EACA;EACA;;;YAIU;EACV;;EAEA;;EAEA;;YAGU;EACV,QAAQ;EACR;EACA;EACA,UAAU;EACV,YAAY;EACZ,cAAc;EACd,QAAQ;;YAGE;WACD;EACT,YAAY,OAAO,kBAAkB;EACrC,YAAY,SAAS,aAAa;;EAElC,cAAc,cAAc,MAAM,aAAa;EAC/C,SAAS,SAAS,aAAa;;YAGrB;EACV,SACE,UAAU,aACV;IAAY,UAAU;MACrB,QAAQ;;;YAID;EACV;EACA,QAAQ;EACR;EACA,QAAQ;EACR,QAAQ;EACR;EACA;EACA,UAAU;EACV,QAAQ;;wBAYM,YAAY,iBAAiB;;;KCp2BjC;EACV;;EAEA;;;wBAIc,QAAQ,UAAS,eAAoB"}
|
|
1
|
+
{"version":3,"file":"index.d.mts","names":["TreeRecord","AssistantBlock","ModelRef","UserBlock","Transcript","EntryRecord","TreeRecord","MessageRecord","Transcript","ToolCallBlock","UserContent","AssistantBlock","ModelRef","ModelContext","Transcript"],"sources":["../../core/src/identity.ts","../../core/src/dataset.ts","../../core/src/usage.ts","../../core/src/agent-transcript.ts","../../core/src/isolation.ts","../../core/src/limits.ts","../../core/src/transcript.ts","../../core/src/judge.ts","../../core/src/catalog.ts","../../core/src/canonical-json.ts","../../core/src/matrix.ts","../../core/src/config.ts","../../core/src/json.ts","../../core/src/definition-site.ts","../../core/src/schema.ts","../../core/src/index.ts","../../agents/src/harness.ts","../../agents/src/pi.ts","../../agents/src/codex.ts","../../agents/src/opencode.ts","../../agents/src/claude-code.ts"],"mappings":";;;qBAEa,oBAAkB,OAAA;YACnB,sBAAsB,mBAAmB;;wBAGrC,SAAS;;qBAOZ,YAAU,OAAA,cAAA,OAAA;YAKX,cAAc,WAAW;YAEzB,YAAY,cAAc,eAAe,2BACxC;qBAEA,mBAAiB,OAAA,cAAA,OAAA;wBAMd,kBAAkB,cAAc,cAAc,MAAM,QAAK,OAAA,cAAA,OAAA;wBAWzD,YAAY,cAAc,cACxC,MAAM,OACN,MAAM,OACL,YAAY;wBAIC,iBAAiB,cAAc,cAC7C,eACA,eAAe;EACZ,MAAM;EAAO,MAAM;EAAM,KAAK,YAAY;;wBAY/B,aAAa,eAAe,eAAe,eAAe;;;;YCvD9D;;YAGA,6BAA6B;EAAe;;;;;;YAM5C,wBAAwB,SAAS,eAAe;;YAGhD;EACV,QAAQ;;EAER;;YAGU;;EAEV;EACA,QAAQ;;YAGE;EACV;EACA;;EAEA,QAAQ;;EAER;;EAEA,YAAY;EACZ,QAAQ;;KAGL;EACH;EACA;;EAEA,aAAa;;EAEb;;YAGU,0BAA0B;EACpC,QAAQ;;EAER;;EAEA;EACA;EACA;;YAGU,0BAA0B,gBAAgB,kBACpD;;EAEE,SAAS,MAAM,uBAAuB,UAAU,QAAQ;;EAExD,SACE,QAAQ,SACR,KAAK,sBACF,iBAAiB,QAAQ;EAC9B;EACA;EACA;;YAGQ;EAEN;EACA,QAAQ;EACR;EACA;;EAGA;EACA,SACE,MAAM,uBACH,iBAAiB,QAAQ;EAC9B,SACE,QAAQ,gBACR,KAAK,sBACF,iBAAiB,QAAQ;;;YAIxB;WACD;WACA;WACA;WACA,YAAY;;WAEZ;WACA,QAAQ;;;wBAIH,oBAAoB,eAAe;;;;;;;wBAuBnC,cAAc,gBAAgB,gBAC5C,YAAY,0BAA0B,0BAA0B,WAC/D;wBAmFa,UAAU,iBAAiB,SAAS;;;;;;;;;;;KC1MxC;EACV;EACA;EACA;EACA;EACA;;EAEA;;;;KCRG,iBAAiB,GAAG,UAAU,eAAe,oBAC9C,KAAK,GAAG;;YAIA,cAAc,iBACxBA;EAEI;EAAa;EAAe;;;YAGtB;;EAEV,kBAAkBC;;EAElB,iBAAiBC;;EAEjB,QAAQ;EACR,OAAO;EACP;IAAU;IAAiB;;;;;;;;;YASjB;;EAEV,KAAK,kBAAkBC,eAAa;IAAY,SAAS;;EACzD,UAAU,SAAS;;;;;;EAMnB,WACE,gBACA,iBACA;IAAY;IAAe;;;EAG7B,OAAO,QAAQ;;;;;;;;EAQf,OAAO,YAAYC;;;;;EAKnB,KACE,wBACA;IACE;IACA;IACA;MAAU;MAAiB;;;;;;;;;;EAU/B,SAAS,YAAYA;;YAGX;EACV;EACA;;EAEA;EACA;;;YAIU;;EAIN;EACA,qBAAqB;EACrB,QAAQ;;EAER,iBAAiBF;;;;;;;;;;YC/FX;EAAgB;;;;;;YAMhB;;YAGA;EAAmB;EAAc;;;YAGjC;EACV;;EAEA;EACA,SAAS;;EAET;;EAEA;EACA,iBAAiB;;;YAIP;EACV;;EAEA;;EAEA;;EAEA;;;;;;EAMA;;YAGU;EACV;EACA;EACA;EACA;;YAGU,YAAY,cAAc,qBAAqB;YAE/C;EACV;;EAEA,UAAU;EACV;EACA;;EAEA,mBAAmB;;;wBAOL,QAAQ;;;;;;;wBAUR,UAAU,SAAS,mBAAmB;;;;;;;wBAsDtC,YACd,UAAS,qBACR;;;;;;YAoBS;;EAEV;;EAEA;;EAEA;;EAEA;;;YAIU;;EAEV;;;YAIU;EACV;;EAEA;;EAEA;;YAGU,cAAc;EAAc;EAAgB;;YAE5C;;EAEV;;EAEA,MAAM,SAAS;;EAEf,SAAS;;EAET;;EAEA;;;YAIU;;WAED;;EAET;EACA;;EAEA,SACE,WAAW,eAAe;;EAG5B,QAAQ,QAAQ;;EAEhB,KAAK,kBAAkB;;;;;;;;YASb;WACD,WAAW;;WAEX;;;;;;WAMA;EACT,KACE,qCACA,UAAU,eACT,QAAQ;EACX,MACE,qCACA,UAAU,KAAK,6BACd;;EAEH,SAAS,eAAe;EACxB,UAAU,cAAc,kBAAkB,aAAa;;EAEvD,KAAK,gBAAgB;;;;;KClPX;;EAEV;;EAEA;;EAEA;;KAGU,kBAAkB;cAEjB,sBAAsB;;KAGvB;EACV;EACA;;EAEA;;;KAIU;EAAiB,OAAO;EAAW;EAAe;;;iBAU9C,cAAc,wBAAwB;;iBAiBtC,YACd,OAAO,yBACP,gBACC;;iBAwBa,iBAAiB,iBAAiB,WAAW;;cAKhD;;iBAGG,qBACd,YAAY,SAAS,uCACpB;;iBASa,eAAe,eAAe;;iBAe9B,YAAY,MAAM,WAAW;;;;cCpFhC;iBAiBG,aAAa,QAAQG,gBAAc,UAAUC;KAIjD,mBAAmBC;EAAkB;;KACrC,cAAcA;EAAkB;;;;;;iBAM5B,WAAW,YAAYC,eAAaF;;iBAsBpC,mBACd,YAAYE,cACZ,OAAOD,0BACNA;;KAQS;;EAEV,SAAS;EACT,MAAME;EACN,SAAS;;;KAIC;EACV;EACA;;EAEA,WAAW;;EAEX;;EAEA;;EAEA;;EAEA;;EAEA;;;iBAIc,oBACd,YAAYD,cACZ,gBACC;;iBAiBa,QAAQ,kBAAkBF,iBAAe;;iBAqBzC,YAAY,SAASI,gBAAcC;;iBAQnC,gBAAgB,YAAYH;;;;;iBAS5B,gBAAgB,QAAQ;;iBAOxB,UAAU,OAAOI;;iBAQjB,WAAW,gBAAgBA,aAAWA;;iBAStC,gBAAgB,QAAQJ,yBAAuBA;;;;;;;;;;KC3LnD;EACN;;EACA;EAAiB,SAAS;;EAC1B;EAAkB,UAAU;;;KAGtB;EAAgB;EAAa;EAAe;;;KAG5C;EACV;EACA;EACA;;;;;;wBAkBc,QACd,SAAS,yBAAyB,eAAe,kBAChD;;;;;KA6BS,kBAAkB;EAEjB;EAAiB;;;;;;;wBAQd,SAAS,SAAS,kBAAkB;;;;;;;;KA4BxC,kBAAkB,mBAAmB;KAErC;EACV;;EAEA;;EAEA;;EAEA;;EAEA,WAAW;;KAGD;EACV;EACA;EACA;EACA;;;KAIU,kBACP,KAAK;EAAkC;KACxC;;;;;wBAkNY,SACd,4BACA,SAAS;;;;;;;;;YChUC;;EAEV;;EAEA,QAAQ;;YAGE;EACV;EACA;EACA;EACA;EACA,UAAU;IAAQ,MAAM;IAAkB;;;EAE1C;;;YAIU;EACN;EAAe;;EACf;EAAgB;;EAChB;;YAEM;EACV;EACA;EACA,YAAY;EACZ;EACA;;EAEA;IAAU;;IAAwB;;;EAElC;IAAU,OAAO;;IAAiB;;;EAElC;IAAU;IAAc;;EACxB;;YAGU;EACV;EACA,MAAM;;EAEN;EACA;EACA;;EAEA;;EAEA,QAAQ;;EAER;;EAEA,SAAS;;EAET;EACA;;YAGU;;;EAEN;EAAc;;EACd;EAAe,aAAa;;EAC5B;EAA0B;EAAc,YAAY;;;;EAEpD;EAAoB,QAAQ;;YAEtB;EACV;EACA;EACA;EACA;EACA,OAAO;;EAEP,aAAa;EACb,UAAU;;EAEV;EACA;IAAU;;;EAEV,SAAS;EACT,YAAY;EACZ,SAAS;EACT,WAAW;;EAEX;;wBAGc,aACd,YAAY,gBACZ,OAAO,uBACP,UAAS,iBACR;;;;wBC9Ga,oBAAoB,OAAO;;;KCM/B;;EAEV;;EAEA,aAAa,SAAS,wBAAwB;;EAE9C,YAAY;;KAEF;EACV;;EAEA;EACA,MAAM;EACN,YAAY;;KAEF;EACV;EACA;EACA,gBAAgB;EAChB,YAAY,SAAS,wBAAwB;EAC7C,WAAW;;EAEX,iBAAiB;;EAEjB,mBAAmB;;;;;EAKnB,SAAS;IACP;IACA;IACA;;;KAGQ,aAAa;EACvB;EACA,MAAM,YAAY;EAClB,MAAM,YAAY,kBAAkB,iBAAiB;;wBA+BvC,uBAAuB,UAAU,sBAC/C,OAAO,IACN,aAAa;;;KC7EJ;;EAEV;EACA;EACA;EACA;;IAEE;;;EAGF,iBAAiB;EACjB,WAAW;;EAEX,SAAS,KAAK;IAA0C;;EACxD;IAAc;IAAsB;IAAiB;;;;;;;EAMrD,SAAS;;;;;;EAMT,YAAY;;;;;;EAMZ,WAAW,SAAS,eAAe;EACnC;;EAEA;;;;;;;IAOE;;IAEA,cAAc;;;;;;;EAOhB;;EAEA,gBAAgB;EAChB,QAAQ;;;;;;EAMR,QAAQ;;KAGE;KAEA;;EAEV;;EAEA;;qBAGW,2BAA2B,SAAS;;qBAMpC;;KAWD,eAAe;KAEf;;EAEV;;;;;;EAMA,UAAU,sBAAsB,eAAe,QAAQ;;;wBAIzC,mBAAmB,UAAU,eAC3C,eAAe,IACd;;;;qBC3GU,qBAAqB;WAGrB;WACA;WAHO;EAClB,YACW,cACA;;;;;;;;wBAYG,OAAO,gBAAgB,gBAAiB;;wBA4CxC,aACd,gBACA;GACI,cAAc;;;;;;;;;;;KC3DR;EAAmB;EAAc;;wBAqB7B,eAAe,iBAAiB;;;;qBCtBnC,iBAAe,OAAA;;qBAUf;qBAEA,qBAAmB,OAAA;;;;;qBAMnB,mBAAiB,OAAA;;;;;;qBAQjB,iBAAiB,OAAO,OAAO;qBAoB/B,iBAAe,OAAA;;;;;;;;;;;;;;qBAiBf,aAAW,OAAA;;;;;qBAMX,kBAAgB,OAAA;;;;;;qBAOhB,cAAY,OAAA;;;;;qBAMZ,oBAAkB,OAAA;;;;;qBAMlB,oBAAkB,OAAA,OAAA,OAAA;;;;;;;qBASlB,mBAAiB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAsBjB,mBAAiB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAMjB,aAAW,OAAA;;;;;;;;;;;;;;;;;;;;qBAuBX,kBAAgB,OAAA,QAAA,YAAA;qBAUhB,oBAAkB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBA4ClB,mBAAiB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAajB,wBAAsB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAiBtB,oBAAkB,OAAA;;;;;;;;;;;;qBAQlB,kBAAgB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBAUhB,qBAAmB,OAAA;;;;;qBAKnB,uBAAqB,OAAA;;;;;qBAKrB,kBAAgB,OAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;qBA4FhB,gBAAgB,OAAO,OAAO;qBAI9B,gBAAc,OAAA;;;;;;YC1Sf;YACA,YACV,gBAAgB;GAAiB,cAAc;;YACrC;GAAgB,cAAc;;YAE9B;;EAEV;EACA;EACA;EACA;;YAGU;;;;;YAMA;EACV;EACA,gBAAgB;;YAGN,gBAAgB,QAAQ,OAAO,kBAAkB;YAKjD;EACV;EACA;EACA;;YAGU;EACN;EAAqB;;EAErB;EACA;;EAEA;EACA;;EAEA;EAAmC;EAAc;;EACjD;EAAiC;EAAc;;EAE/C;EACA;EACA;EACA;;EAGA;EACA;EACA;EACA;;EAGA;EACA;EACA,OAAO;EACP;;EAEA;EAAwB;EAAgB;;EAExC;EACA;;EAEA;EACA;EACA;;EAGA;EACA;EACA,OAAO;EACP;;EAEC;EAAuB;IAAsB;EAC9C;EAAuB;;EACvB;EAAe,OAAO;EAAe;;;;;;;;;;;;YAY/B;EACV,YAAY;;EAEZ,WAAW,eAAe;EAC1B,WAAW;;YAGD,WAAW;EAAgB;;YAE3B;;EAEV;;YAGU,qBAAqB;YACrB,qBAAqB;YAErB;EACV;EACA;;EAEA,QAAQ;EACR,UAAU;;EAEV,aAAa;EACb,UAAU;;EAEV,MAAM;;;;;EAKN,QAAQ;;EAER,QAAQ;;;;;EAKR,OAAO;;EAEP,UAAU,SAAS,eAAe;;EAElC,WAAW;;EAEX,oBAAoB;;YAGV,WAAW;WACZ,WAAW;;WAEX,WAAW;EACpB,MAAM;IACJ,SAAS;IACT,UAAU;;;;;IAKV,YAAY;;;;;;;;IAQZ,SAAS,cAAc,kBAAkB,eAAe;MACtD,QAAQ,WAAW;;YAGb,WAAW;;;;;EAKrB,KAAK,SAAS,WAAW,eAAe;;EAExC,UAAU;;;qBAIC,wBAAwB;;wBAKrB,cAAc,OAAO,kBAAkB;;;;;wBAUvC,kBAAkB,UAAU,YAAY,OAAO,IAAI;;YAMvD;;EAEV,YAAY;;EAEZ,MAAM;;EAEN;;EAEA,QAAQ;;YAGE;EACV;;EAEA,oBAAoB,OAAO;;YAGjB;EACV;EACA,aAAa;;;YAIH;EACV;;EAEA,kBAAkB;EAClB,4BAA4B;EAC5B;;IAEE;IACA;IACA;IACA,WAAW;IACX;IACA,SAAS;;EAEX;;YAGU;EACV;EACA;EACA;IAAY;IAAc,YAAY;;;YAE5B,iBACV,WAAW,YAAY,cAAc;;wBAGvB,eACd,UAAU,iCACT;;;;;wBAUa,KAAK,SAAS,sBAAsB;wBAIpC,MAAM,aAAa,aAAa;;;;;;YASpC;EACV,QAAQ;;EAER;;EAEA,QAAQ;;EAER;;EAEA,QAAQ;;EAER;;EAEA,aAAa;;;YAIH;EAAiB;EAAe;;YAEhC,WAAW,+BAEnB;;EAGE;EACA;EACA;;;;;EAKA,WAAW;;;;;EAKX,WAAW;;EAEX,QAAQ;;;;;;wBAOE,KAAK,iBAAiB;YAK1B;EACV;EACA;EACA;EACA;;EAEA;EACA;EACA,WAAW;;EAEX;EACA,WAAW;;EAEX;EACA,QAAQ;EACR;EACA,QAAQ;;EAER;;YAGU;EACV,WAAW;EACX,WAAW;;YAGD;EACV,SAAS;;;;;;EAMT,YAAY;;EAEZ,WAAW,SAAS,eAAe;;EAEnC,oBAAoB;EACpB,WAAW;;EAEX,OAAO;;EAEP;;EAEA,MAAM;;;;;;EAMN,sBAAsB;;;;;;;EAOtB,WAAW,UACT,MAAM,gBAAgB,YACrB,kBAAkB;EACrB,WACE,MAAM,YAAY,qBACjB,kBAAkB;;;;;;EAMrB,OAAO,cAAc,kBAAkB,aAAa;;;;;;;YAQ1C,kBAAkB,YAAY,KAAK;EAC7C,WAAW;;YAGD,gBAAgB;EAC1B;EACA;;EAEA;EACA;;EAEA;;;;;;EAMA,WAAW;EACX,IACE,SAAS,2BACE,WAAW,YAAY,kBAAkB,WAAW;;;YAIvD,cACV,SAAS,mBACN,wBAAwB,QAAQ;;;;;;;YAQzB;EACV;EACA;EACA;;EAEA;;;;;;EAMA,QAAQ;;;;;EAKR,SAAS,SAAS,eAAe;;EAEjC,QAAQ;;EAER;;EAEA;;YAGU,cAAc,kBAAkB;;wBAG5B,UAAU,oBACxB,cACA,KAAK,gBAAgB,kBACrB,UAAS,KACP;;EAIA,WAAW;IAEW;IAAiB;;IAGxC,gBAAgB;;;;;;YA6BP;EACV;EACA;;EAEA;;EAEA;EACA,IACE,SAAS,sCACa;;;wBAIV,aAAa,qBAC3B,MAAM,aAAa,4BAA4B,MAC/C,KAAK,eACL,UAAS,KAAK,6BACb;;YAMS;EACV,MAAM;;EAEN;EACA;EACA;;EAEA;EACA,QAAQ;;wBAKM,MACd,cACA,SAAS,KAAK;EACZ,SAAS;IAEV;YA+BS;;EAEV;;YAGU,eAAe,eAAe,aAAa;EACrD;EACA;EACA,OAAO;;EAEP,oBAAoB;EACpB,YAAY;EACZ,SAAS;;EAET,mBAAmB;EACnB,SAAS;;;;;EAKT;;;;;EAKA,SAAS;;EAET,WAAW;EACX,YAAY;;;YAIF;;;;;;;;;EASV;;wBAGc,iBAAiB,UAAU,gBAAgB,YAAY,IAAI;;wBA8C3D,wBACd,YAAY,KAAK;EACf;GAEF,eAAe;YAgBL,UACV,wBAAwB,4BAA4B;EAEpD;EACA;EACA,OAAO;;wBAGO,kBAAkB,eAAe,WAC/C,OAAO,SACN;YAKS,mBAAmB,iBAAiB,YAAY;wBAE5C,YAAY;EAAS;;YAkBzB;;EAEV,gBAAgB;EAChB,iBAAiB;EACjB,mBAAmB;EACnB,IAAI,cAAc;EAClB,SAAS,cAAc;EACvB,YAAY;EACZ,iBAAiB;;EAEjB,QAAQ,UAAU,iBAAiB;;YAGzB;EACV;EACA;EACA;;YAGU;EACV;EACA;EACA;;;;;;wBAOc,oBACR,gCAAgC,oBACtC,eAAe,iBAAiB;;;;;;;YA6EtB;;YAIA;EACV;;EAEA;EACA;;;;;;YAOU;;EAEV;;EAEA;;EAEA;EACA;EACA;;EAEA;EACA;IAAO;IAAkB;IAAmB;IAAc;;;;YAIhD;EACV,SAAS;EACT,OAAO;EACP,QAAQ;;;YAIE;EAAe;EAAa;;YAE5B;EACV;EACA;EACA,QAAQ;EACR;EACA;EACA,aAAa;EACb;IAAW;IAAY;;EACvB,MAAM;EACN;;EAEA,SAAS;EACT,UAAU;;EAEV,QAAQ;;EAER;;EAEA,WAAW;;EAEX,SAAS;;EAET,YAAY;;;YAIF;EACN;EAAmB;EAAe,SAAS;;EAC3C;;YAEM;EACV;EACA,YAAY;EACZ;;EAEA;;YAGU;EACV;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;EACA,WAAW;;EAEX;EACA,QAAQ;EACR;EACA,QAAQ;;EAER;IAAoB;IAAmB;;;;;;;;;;;;;;;;;qBAiB5B;YAUD,4BAA4B;;YAE5B,aAAa;YAEb;EACV,SAAS;EACT,UAAU;EACV,cAAc;EACd;EACA;EACA;;YAGU;;EAEV;EACA;EACA;;;YAIU;EACV;;EAEA;;EAEA;;YAGU;EACV,QAAQ;EACR;EACA;EACA,UAAU;EACV,YAAY;EACZ,cAAc;EACd,QAAQ;;EAER,QAAQ;;YAGE;WACD;;EAET,YAAY,OAAO,WAAW;;EAE9B,iBACE,MAAM,uBAAuB,gCAC5B;;EAEH,cAAc,YAAY,eAAe;EACzC,YAAY,SAAS,aAAa;;EAElC,cAAc,cAAc,MAAM,aAAa;EAC/C,SAAS,SAAS,aAAa;;YAGrB;EACV,SACE,UAAU,aACV;IAAY,UAAU;MACrB,QAAQ;;;YAID;EACV;EACA,QAAQ;EACR;EACA,QAAQ;EACR,QAAQ;EACR;EACA;EACA,UAAU;EACV,QAAQ;;EAER,QAAQ;;wBAgBM,YAAY,iBAAiB;;wBAa7B,SACd,SAAS,KAAK,+BACd,OAAO,wBACN;;YAWS;EACV;EACA,MAAM;;EAEN;EACA;EACA,QAAQ;;;wBAIM,UACd,SAAS,KAAK,+BACd,gBACC;;wBAKa,UAAU,SAAS,KAAK;;wBAKxB,MACd,SAAS,KAAK,gCACb;;;;;;wBASa,aACd,SAAS,KAAK,gCACbK;;;;KCroCS;;EAEV;;;;;EAKA,mBAAmB,YAAY;;EAE/B;;EAEA,MAAM,SAAS;;;;;;;;;EASf;;;;;;;EAOA,cAAc,SAAS,wBAAwB;;;KAIrC;;EAEV;;EAEA;;EAEA;;EAEA;;EAEA;EACA;;;KAIU;EACV;EACA;EACA;;EAEA,UAAU,oBAAoB;;EAE9B;;EAEA,KAAK,MAAM;;EAEX,mBAAmB;;;;;EAKnB,YACE,SAAS;IAAgB;IAAiB;;;EAG5C,OAAO,eAAeC;;;;;;;;wBASR,aACd,MAAM,aACN,SAAS,iBACR;;;KC7ES,YAAY;;EAEtB;;EAEA;;EAEA;;;;;;;wBAQc,QAAQ,UAAS,YAAiB;;;KCpBtC,eAAe;;;;;;;wBAQX,WAAW,UAAS,eAAoB;;;KCH5C,kBAAkB;;EAE5B;;;;;;;;;;;;wBAac,cAAc,UAAS,kBAAuB;;;KCnBlD,oBAAoB;;EAE9B;;;;;;;wBAQc,gBAAgB,UAAS,oBAAyB"}
|
package/dist/index.mjs
CHANGED
|
@@ -1,71 +1,260 @@
|
|
|
1
|
-
import { $ as
|
|
2
|
-
|
|
3
|
-
|
|
4
|
-
|
|
5
|
-
|
|
1
|
+
import { $ as DEFAULT_ARTIFACT_EXCLUDE, A as JudgeRunInfoSchema, At as definitionSite, B as RunMetadataSchema, C as ApiErrorSchema, D as BatchSchema, Dt as defineDataset, E as AutIdentitySchema, F as RecordedErrorSchema, Ft as parseResourceUri, G as ScoreResultSchema, H as RunSourceSchema, I as RunDatasetSchema, It as resourceUri, J as container, K as TranscriptSchema, L as RunEventSchema, Lt as resourceUriSchema, M as LimitsSchema, Mt as ResourceUriSchema, N as MetricResultSchema, Nt as UuidSchema, O as CheckpointResultSchema, Ot as isDataset, P as REPORT_SCHEMA_VERSION, Pt as authorId, Q as canonicalParameters, R as RunIsolationSchema, Rt as resourceUuid, S as user, T as AttachmentEntrySchema, U as RunStatusSchema, V as RunScoringSchema, W as RunSummarySchema, X as sandboxExec, Y as none, Z as defineEvalMatrix, _ as recordError, a as assertJudgesIndependent, at as choices, b as toolCalls, c as defineEval, d as judge, et as DEFAULT_ATTACHMENT_LIMITS, f as lastReply, ft as NotJsonError, g as predicate, h as modelContext, i as agentRuntimes, j as LimitReachedSchema, jt as ResourceKindSchema, k as JsonValueSchema, kt as relativeDatasetPath, l as defineSuite, lt as passMark, m as metric, mt as toJsonObject, n as DEFAULT_AGENT_RUNTIMES, nt as catalogEntry, o as authoringId, ot as criteria, p as messages, pt as toJson, q as UsageSchema, r as agent, s as defineAgent, t as BUILT_IN_METRICS, tt as defineConfig, u as expectToolCall, v as registerEvals, w as ArtifactEntrySchema, x as usage, y as skip, z as RunManifestSchema } from "./src-D6Ax2t9b.mjs";
|
|
2
|
+
import { randomUUID } from "node:crypto";
|
|
3
|
+
import pi from "@leostera/atxf/pi";
|
|
4
|
+
import codex from "@leostera/atxf/codex";
|
|
5
|
+
import opencode from "@leostera/atxf/opencode";
|
|
6
|
+
import claudeCode from "@leostera/atxf/claude-code";
|
|
7
|
+
//#region ../agents/src/harness.ts
|
|
8
|
+
/**
|
|
9
|
+
* An adapter that runs a coding-agent CLI through `context.shell`, one
|
|
10
|
+
* non-interactive invocation per user message, continuing one session.
|
|
11
|
+
* After each turn it imports the CLI's own session record through ATXF,
|
|
12
|
+
* so the run's transcript has every message, tool call and token count.
|
|
13
|
+
*/
|
|
14
|
+
function harnessAgent(spec, options) {
|
|
15
|
+
const command = options.command ?? spec.command;
|
|
6
16
|
return defineAgent({
|
|
7
17
|
identity: {
|
|
8
|
-
name:
|
|
18
|
+
name: spec.name,
|
|
9
19
|
kind: "process",
|
|
10
|
-
id:
|
|
20
|
+
id: spec.id
|
|
11
21
|
},
|
|
12
22
|
runtimes: { local: {
|
|
13
23
|
kind: "process",
|
|
14
|
-
configuration: {
|
|
24
|
+
configuration: {
|
|
25
|
+
command,
|
|
26
|
+
...typeof options.model === "string" ? { model: options.model } : {},
|
|
27
|
+
config: options.config ?? "private"
|
|
28
|
+
}
|
|
15
29
|
} },
|
|
16
|
-
async start({ context,
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
30
|
+
async start({ context, transcript, attach }) {
|
|
31
|
+
const { shell } = context;
|
|
32
|
+
const stateDir = joinPath(shell.home, spec.id);
|
|
33
|
+
const configDir = (options.config ?? "private") === "private" ? joinPath(stateDir, "config") : void 0;
|
|
34
|
+
await shell.writeFile(joinPath(stateDir, ".evalkit"), "");
|
|
35
|
+
if (configDir) {
|
|
36
|
+
await shell.writeFile(joinPath(configDir, ".evalkit"), "");
|
|
37
|
+
for (const [file, content] of Object.entries(options.configFiles ?? {})) await shell.writeFile(joinPath(configDir, file), content);
|
|
38
|
+
} else if (options.configFiles) throw new Error(`${spec.name}: configFiles needs config: 'private', which is the default`);
|
|
39
|
+
const env = {
|
|
40
|
+
...configDir ? spec.configEnv(configDir) : {},
|
|
41
|
+
...options.env
|
|
42
|
+
};
|
|
43
|
+
const model = typeof options.model === "function" ? options.model(context.parameters ?? {}) : options.model;
|
|
44
|
+
let session = spec.newSession?.();
|
|
45
|
+
let turns = 0;
|
|
46
|
+
const log = [];
|
|
22
47
|
return {
|
|
23
48
|
async send(message) {
|
|
24
|
-
turn
|
|
25
|
-
await onEvent({
|
|
26
|
-
kind: "turn-started",
|
|
27
|
-
turn,
|
|
28
|
-
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
29
|
-
});
|
|
30
|
-
const process = Bun.spawn([
|
|
49
|
+
const turn = {
|
|
31
50
|
command,
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
new Response(process.stdout).text(),
|
|
44
|
-
new Response(process.stderr).text(),
|
|
45
|
-
process.exited
|
|
46
|
-
]);
|
|
47
|
-
if (exitCode !== 0) throw new Error(`Pi exited with ${exitCode}: ${stderr.trim()}`);
|
|
48
|
-
await onEvent({
|
|
49
|
-
kind: "message",
|
|
50
|
-
role: "assistant",
|
|
51
|
-
content: stdout.trim(),
|
|
52
|
-
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
53
|
-
});
|
|
54
|
-
await onEvent({
|
|
55
|
-
kind: "turn-completed",
|
|
56
|
-
turn,
|
|
57
|
-
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
51
|
+
model,
|
|
52
|
+
configDir,
|
|
53
|
+
stateDir,
|
|
54
|
+
session,
|
|
55
|
+
first: turns === 0
|
|
56
|
+
};
|
|
57
|
+
turns += 1;
|
|
58
|
+
const argv = [...spec.argv(turn), ...options.args ?? []];
|
|
59
|
+
const result = await shell.exec(argv, {
|
|
60
|
+
input: message,
|
|
61
|
+
env
|
|
58
62
|
});
|
|
63
|
+
log.push(`$ ${argv.join(" ")} # turn ${turns}, exit ${result.exitCode}`, result.stdout, result.stderr);
|
|
64
|
+
session ??= spec.sessionFromOutput?.(result.stdout);
|
|
65
|
+
const imported = session ? await importSession(shell, spec, {
|
|
66
|
+
...turn,
|
|
67
|
+
session
|
|
68
|
+
}, env).then((transcript_) => {
|
|
69
|
+
transcript.import(transcript_);
|
|
70
|
+
}, (error) => error) : /* @__PURE__ */ new Error(`${spec.name} did not report its session`);
|
|
71
|
+
if (result.exitCode !== 0 && !context.signal.aborted) throw new Error(`${spec.name} exited with ${result.exitCode}: ${tail(result.stderr || result.stdout)}`);
|
|
72
|
+
if (imported && !context.signal.aborted) throw new Error(`Could not read ${spec.name}'s session: ${imported instanceof Error ? imported.message : String(imported)}`);
|
|
59
73
|
},
|
|
60
74
|
async close() {
|
|
61
|
-
await
|
|
62
|
-
kind: "completed",
|
|
63
|
-
timestamp: (/* @__PURE__ */ new Date()).toISOString()
|
|
64
|
-
});
|
|
75
|
+
if (log.length) await attach(`logs/${spec.id}.log`, log.join("\n"));
|
|
65
76
|
}
|
|
66
77
|
};
|
|
67
78
|
}
|
|
68
79
|
});
|
|
69
80
|
}
|
|
81
|
+
async function importSession(shell, spec, turn, env) {
|
|
82
|
+
const target = joinPath(turn.stateDir, "session");
|
|
83
|
+
const saved = await shell.exec(spec.saveSession({
|
|
84
|
+
...turn,
|
|
85
|
+
target
|
|
86
|
+
}), {
|
|
87
|
+
env,
|
|
88
|
+
signal: new AbortController().signal,
|
|
89
|
+
timeoutMs: 3e4
|
|
90
|
+
});
|
|
91
|
+
if (saved.exitCode !== 0) throw new Error(tail(saved.stderr || saved.stdout) || `exit ${saved.exitCode}`);
|
|
92
|
+
return spec.toAtxf(await shell.readFile(target));
|
|
93
|
+
}
|
|
94
|
+
/** Joins shell paths, which are POSIX inside containers. */
|
|
95
|
+
const joinPath = (...parts) => parts.join("/").replace(/\/+/g, "/");
|
|
96
|
+
/** A shell-quoted word. */
|
|
97
|
+
const quoteWord = (word) => `'${word.replace(/'/g, `'\\''`)}'`;
|
|
98
|
+
const tail = (text) => text.trim().slice(-2e3);
|
|
99
|
+
//#endregion
|
|
100
|
+
//#region ../agents/src/pi.ts
|
|
101
|
+
/**
|
|
102
|
+
* Pi (https://pi.dev) as the agent under test: `pi --print` with its own
|
|
103
|
+
* tools, system prompt and context handling, one session per run. Its
|
|
104
|
+
* session file becomes the run's transcript.
|
|
105
|
+
*/
|
|
106
|
+
function piAgent(options = {}) {
|
|
107
|
+
return harnessAgent({
|
|
108
|
+
id: "pi",
|
|
109
|
+
name: "Pi",
|
|
110
|
+
command: "pi",
|
|
111
|
+
configEnv: (configDir) => ({ PI_CODING_AGENT_DIR: configDir }),
|
|
112
|
+
newSession: () => randomUUID(),
|
|
113
|
+
argv: (turn) => [
|
|
114
|
+
turn.command,
|
|
115
|
+
"--print",
|
|
116
|
+
"--session-dir",
|
|
117
|
+
joinPath(turn.stateDir, "sessions"),
|
|
118
|
+
"--session-id",
|
|
119
|
+
turn.session ?? "",
|
|
120
|
+
...turn.configDir ? [
|
|
121
|
+
"--no-extensions",
|
|
122
|
+
"--no-skills",
|
|
123
|
+
"--no-prompt-templates",
|
|
124
|
+
"--no-themes"
|
|
125
|
+
] : [],
|
|
126
|
+
...options.tools === false ? ["--no-tools"] : [],
|
|
127
|
+
...turn.model ? ["--model", turn.model] : [],
|
|
128
|
+
...(options.skills ?? []).flatMap((skill) => ["--skill", skill]),
|
|
129
|
+
...options.appendSystemPrompt ? ["--append-system-prompt", options.appendSystemPrompt] : []
|
|
130
|
+
],
|
|
131
|
+
saveSession: ({ stateDir, session, target }) => [
|
|
132
|
+
"sh",
|
|
133
|
+
"-c",
|
|
134
|
+
`cp ${quoteWord(joinPath(stateDir, "sessions"))}/*_${session}.jsonl ${quoteWord(target)}`
|
|
135
|
+
],
|
|
136
|
+
toAtxf: (text) => pi.toAtxf(text)
|
|
137
|
+
}, options);
|
|
138
|
+
}
|
|
139
|
+
//#endregion
|
|
140
|
+
//#region ../agents/src/codex.ts
|
|
141
|
+
/**
|
|
142
|
+
* Codex as the agent under test: `codex exec` with approvals and its own
|
|
143
|
+
* sandbox off (EvalKit's isolation is the sandbox; Codex's cannot nest in
|
|
144
|
+
* `sandboxExec()`), one session per run. Its rollout file becomes the
|
|
145
|
+
* run's transcript.
|
|
146
|
+
*/
|
|
147
|
+
function codexAgent(options = {}) {
|
|
148
|
+
return harnessAgent({
|
|
149
|
+
id: "codex",
|
|
150
|
+
name: "Codex",
|
|
151
|
+
command: "codex",
|
|
152
|
+
configEnv: (configDir) => ({ CODEX_HOME: configDir }),
|
|
153
|
+
argv: (turn) => [
|
|
154
|
+
turn.command,
|
|
155
|
+
"exec",
|
|
156
|
+
...turn.session ? ["resume", turn.session] : [],
|
|
157
|
+
"--json",
|
|
158
|
+
"--skip-git-repo-check",
|
|
159
|
+
"--dangerously-bypass-approvals-and-sandbox",
|
|
160
|
+
...turn.model ? ["--model", turn.model] : [],
|
|
161
|
+
"-"
|
|
162
|
+
],
|
|
163
|
+
sessionFromOutput: (stdout) => {
|
|
164
|
+
for (const line of stdout.split("\n")) {
|
|
165
|
+
const event = parseLine(line);
|
|
166
|
+
if (event?.type === "thread.started" && typeof event.thread_id === "string") return event.thread_id;
|
|
167
|
+
}
|
|
168
|
+
},
|
|
169
|
+
saveSession: ({ session, target }) => [
|
|
170
|
+
"sh",
|
|
171
|
+
"-c",
|
|
172
|
+
`cp "\${CODEX_HOME:-$HOME/.codex}"/sessions/*/*/*/rollout-*-${session}.jsonl ${quoteWord(target)}`
|
|
173
|
+
],
|
|
174
|
+
toAtxf: (text) => codex.toAtxf(text)
|
|
175
|
+
}, options);
|
|
176
|
+
}
|
|
177
|
+
function parseLine(line) {
|
|
178
|
+
try {
|
|
179
|
+
const value = JSON.parse(line);
|
|
180
|
+
return value && typeof value === "object" ? value : void 0;
|
|
181
|
+
} catch {
|
|
182
|
+
return;
|
|
183
|
+
}
|
|
184
|
+
}
|
|
185
|
+
//#endregion
|
|
186
|
+
//#region ../agents/src/opencode.ts
|
|
187
|
+
/**
|
|
188
|
+
* opencode as the agent under test: `opencode run --standalone`, with a
|
|
189
|
+
* private server per command and permissions auto-approved, one session
|
|
190
|
+
* per run. Its session export becomes the run's transcript.
|
|
191
|
+
*
|
|
192
|
+
* With `config: 'private'`, opencode's config, data and state directories
|
|
193
|
+
* (`XDG_CONFIG_HOME`, `XDG_DATA_HOME`, `XDG_STATE_HOME`) are in the config
|
|
194
|
+
* directory: put `config/opencode/opencode.json` and
|
|
195
|
+
* `data/opencode/auth.json` in `configFiles`.
|
|
196
|
+
*/
|
|
197
|
+
function opencodeAgent(options = {}) {
|
|
198
|
+
return harnessAgent({
|
|
199
|
+
id: "opencode",
|
|
200
|
+
name: "opencode",
|
|
201
|
+
command: "opencode",
|
|
202
|
+
configEnv: (configDir) => ({
|
|
203
|
+
XDG_CONFIG_HOME: joinPath(configDir, "config"),
|
|
204
|
+
XDG_DATA_HOME: joinPath(configDir, "data"),
|
|
205
|
+
XDG_STATE_HOME: joinPath(configDir, "state")
|
|
206
|
+
}),
|
|
207
|
+
argv: (turn) => [
|
|
208
|
+
turn.command,
|
|
209
|
+
"run",
|
|
210
|
+
"--standalone",
|
|
211
|
+
"--format",
|
|
212
|
+
"json",
|
|
213
|
+
"--auto",
|
|
214
|
+
...turn.session ? ["--session", turn.session] : [],
|
|
215
|
+
...turn.model ? ["--model", turn.model] : [],
|
|
216
|
+
...options.agent ? ["--agent", options.agent] : []
|
|
217
|
+
],
|
|
218
|
+
sessionFromOutput: (stdout) => /"sessionID":"([^"]+)"/.exec(stdout)?.[1],
|
|
219
|
+
saveSession: ({ command, session, target }) => [
|
|
220
|
+
"sh",
|
|
221
|
+
"-c",
|
|
222
|
+
`${quoteWord(command)} session export --standalone ${quoteWord(session)} > ${quoteWord(target)}`
|
|
223
|
+
],
|
|
224
|
+
toAtxf: (text) => opencode.toAtxf(text)
|
|
225
|
+
}, options);
|
|
226
|
+
}
|
|
227
|
+
//#endregion
|
|
228
|
+
//#region ../agents/src/claude-code.ts
|
|
229
|
+
/**
|
|
230
|
+
* Claude Code as the agent under test: `claude -p` with every tool allowed
|
|
231
|
+
* (EvalKit's isolation is the sandbox), one session per run. Its session
|
|
232
|
+
* file becomes the run's transcript.
|
|
233
|
+
*/
|
|
234
|
+
function claudeCodeAgent(options = {}) {
|
|
235
|
+
return harnessAgent({
|
|
236
|
+
id: "claude-code",
|
|
237
|
+
name: "Claude Code",
|
|
238
|
+
command: "claude",
|
|
239
|
+
configEnv: (configDir) => ({ CLAUDE_CONFIG_DIR: configDir }),
|
|
240
|
+
newSession: () => randomUUID(),
|
|
241
|
+
argv: (turn) => [
|
|
242
|
+
turn.command,
|
|
243
|
+
"--print",
|
|
244
|
+
"--output-format",
|
|
245
|
+
"json",
|
|
246
|
+
"--dangerously-skip-permissions",
|
|
247
|
+
...turn.first ? ["--session-id", turn.session ?? ""] : ["--resume", turn.session ?? ""],
|
|
248
|
+
...turn.model ? ["--model", turn.model] : [],
|
|
249
|
+
...options.appendSystemPrompt ? ["--append-system-prompt", options.appendSystemPrompt] : []
|
|
250
|
+
],
|
|
251
|
+
saveSession: ({ session, target }) => [
|
|
252
|
+
"sh",
|
|
253
|
+
"-c",
|
|
254
|
+
`cp "\${CLAUDE_CONFIG_DIR:-$HOME/.claude}"/projects/*/${session}.jsonl ${quoteWord(target)}`
|
|
255
|
+
],
|
|
256
|
+
toAtxf: (text) => claudeCode.toAtxf(text)
|
|
257
|
+
}, options);
|
|
258
|
+
}
|
|
70
259
|
//#endregion
|
|
71
|
-
export { ApiErrorSchema, ArtifactEntrySchema, AttachmentEntrySchema, AutIdentitySchema, BUILT_IN_METRICS, BatchSchema, CheckpointResultSchema, DEFAULT_AGENT_RUNTIMES, DEFAULT_ARTIFACT_EXCLUDE, DEFAULT_ATTACHMENT_LIMITS, JsonValueSchema, JudgeRunInfoSchema, MetricResultSchema, NotJsonError, REPORT_SCHEMA_VERSION, RecordedErrorSchema, ResourceKindSchema, ResourceUriSchema, RunDatasetSchema, RunManifestSchema, RunMetadataSchema, RunScoringSchema, RunSourceSchema, RunStatusSchema, RunSummarySchema, ScoreResultSchema,
|
|
260
|
+
export { ApiErrorSchema, ArtifactEntrySchema, AttachmentEntrySchema, AutIdentitySchema, BUILT_IN_METRICS, BatchSchema, CheckpointResultSchema, DEFAULT_AGENT_RUNTIMES, DEFAULT_ARTIFACT_EXCLUDE, DEFAULT_ATTACHMENT_LIMITS, JsonValueSchema, JudgeRunInfoSchema, LimitReachedSchema, LimitsSchema, MetricResultSchema, NotJsonError, REPORT_SCHEMA_VERSION, RecordedErrorSchema, ResourceKindSchema, ResourceUriSchema, RunDatasetSchema, RunEventSchema, RunIsolationSchema, RunManifestSchema, RunMetadataSchema, RunScoringSchema, RunSourceSchema, RunStatusSchema, RunSummarySchema, ScoreResultSchema, TranscriptSchema, UsageSchema, UuidSchema, agent, agentRuntimes, assertJudgesIndependent, authorId, authoringId, canonicalParameters, catalogEntry, choices, claudeCodeAgent, codexAgent, container, criteria, defineAgent, defineConfig, defineDataset, defineEval, defineEvalMatrix, defineSuite, definitionSite, expectToolCall, harnessAgent, isDataset, judge, lastReply, messages, metric, modelContext, none, opencodeAgent, parseResourceUri, passMark, piAgent, predicate, recordError, registerEvals, relativeDatasetPath, resourceUri, resourceUriSchema, resourceUuid, sandboxExec, skip, toJson, toJsonObject, toolCalls, usage, user };
|