@vercel/agent-eval 0.0.2 → 0.0.4

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/dist/cli.js CHANGED
@@ -18,7 +18,7 @@ const program = new Command();
18
18
  program
19
19
  .name('agent-eval')
20
20
  .description('Framework for testing AI coding agents in isolated sandboxes')
21
- .version('0.0.1');
21
+ .version('0.0.4');
22
22
  /**
23
23
  * Resolve config path shorthand.
24
24
  * - "cc" -> "experiments/cc.ts"
@@ -149,7 +149,7 @@ IMPORTANT: Do not run npm, pnpm, yarn, or any package manager commands. Dependen
149
149
  }
150
150
  // Upload test files for validation
151
151
  await sandbox.uploadFiles(testFiles);
152
- // Create vitest config for EVAL.ts
152
+ // Create vitest config for EVAL.ts/tsx
153
153
  await createVitestConfig(sandbox);
154
154
  // Capture the Claude Code transcript
155
155
  const transcript = await captureTranscript(sandbox);
@@ -1 +1 @@
1
- {"version":3,"file":"claude-code.js","sourceRoot":"","sources":["../../../src/lib/agents/claude-code.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,cAAc,EACd,iBAAiB,GAClB,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,aAAa,EACb,qBAAqB,EACrB,kBAAkB,EAClB,UAAU,EACV,gBAAgB,GACjB,MAAM,aAAa,CAAC;AAErB;;;GAGG;AACH,KAAK,UAAU,iBAAiB,CAAC,OAAuB;IACtD,IAAI,CAAC;QACH,6DAA6D;QAC7D,MAAM,OAAO,GAAG,OAAO,CAAC,mBAAmB,EAAE,CAAC;QAC9C,4DAA4D;QAC5D,MAAM,WAAW,GAAG,OAAO,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;QAChD,MAAM,gBAAgB,GAAG,sBAAsB,WAAW,EAAE,CAAC;QAE7D,oDAAoD;QACpD,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,QAAQ,CACvC,SAAS,gBAAgB,gCAAgC,CAC1D,CAAC;QAEF,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;YAC3D,OAAO,SAAS,CAAC;QACnB,CAAC;QAED,MAAM,cAAc,GAAG,UAAU,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;QAChD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;QACvD,OAAO,OAAO,CAAC;IACjB,CAAC;IAAC,MAAM,CAAC;QACP,oCAAoC;QACpC,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB,CAAC,EAAE,kBAAkB,EAAmC;IAC3F,OAAO;QACL,IAAI,EAAE,kBAAkB,CAAC,CAAC,CAAC,+BAA+B,CAAC,CAAC,CAAC,aAAa;QAC1E,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC,iCAAiC,CAAC,CAAC,CAAC,aAAa;QAEnF,eAAe;YACb,OAAO,kBAAkB,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,gBAAgB,CAAC,YAAY,CAAC;QACtF,CAAC;QAED,eAAe;YACb,OAAO,MAAM,CAAC;QAChB,CAAC;QAED,KAAK,CAAC,GAAG,CAAC,WAAmB,EAAE,OAAwB;YACvD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,IAAI,OAAO,GAA0B,IAAI,CAAC;YAC1C,IAAI,WAAW,GAAG,EAAE,CAAC;YACrB,IAAI,OAAO,GAAG,KAAK,CAAC;YACpB,IAAI,cAAc,GAAG,KAAK,CAAC;YAE3B,sBAAsB;YACtB,MAAM,YAAY,GAAG,GAAG,EAAE;gBACxB,OAAO,GAAG,IAAI,CAAC;gBACf,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;gBACjC,CAAC;YACH,CAAC,CAAC;YAEF,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACnB,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;oBAC3B,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,sBAAsB;wBAC7B,QAAQ,EAAE,CAAC;qBACZ,CAAC;gBACJ,CAAC;gBACD,OAAO,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;YACzD,CAAC;YAED,IAAI,CAAC;gBACH,6BAA6B;gBAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,WAAW,CAAC,CAAC;gBACtD,MAAM,EAAE,cAAc,EAAE,SAAS,EAAE,GAAG,cAAc,CAAC,QAAQ,CAAC,CAAC;gBAE/D,8CAA8C;gBAC9C,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;qBACjC,CAAC;gBACJ,CAAC;gBAED,iBAAiB;gBACjB,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,CAAC;oBACpC,OAAO,EAAE,OAAO,CAAC,OAAO;oBACxB,OAAO,EAAE,QAAQ;iBAClB,CAAC,CAAC;gBAEH,8EAA8E;gBAC9E,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,2CAA2C;gBAC3C,MAAM,OAAO,CAAC,WAAW,CAAC,cAAc,CAAC,CAAC;gBAE1C,iCAAiC;gBACjC,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;oBAClB,MAAM,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;gBAC/B,CAAC;gBAED,uBAAuB;gBACvB,MAAM,aAAa,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;gBACnE,IAAI,aAAa,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACjC,MAAM,IAAI,KAAK,CAAC,uBAAuB,aAAa,CAAC,MAAM,EAAE,CAAC,CAAC;gBACjE,CAAC;gBAED,mCAAmC;gBACnC,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE;oBACjD,SAAS;oBACT,IAAI;oBACJ,2BAA2B;iBAC5B,CAAC,CAAC;gBACH,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC9B,MAAM,IAAI,KAAK,CAAC,+BAA+B,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;gBACtE,CAAC;gBAED,kCAAkC;gBAClC,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEjC,0BAA0B;gBAC1B,MAAM,cAAc,GAAG,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE;;6LAEwI,CAAC;gBAExL,kDAAkD;gBAClD,MAAM,YAAY,GAAG,MAAM,OAAO,CAAC,UAAU,CAC3C,QAAQ,EACR,CAAC,SAAS,EAAE,SAAS,EAAE,OAAO,CAAC,KAAK,EAAE,gCAAgC,EAAE,cAAc,CAAC,EACvF;oBACE,GAAG,EAAE,kBAAkB;wBACrB,CAAC,CAAC;4BACE,2CAA2C;4BAC3C,kBAAkB,EAAE,UAAU,CAAC,OAAO;4BACtC,oBAAoB,EAAE,OAAO,CAAC,MAAM;4BACpC,iBAAiB,EAAE,EAAE;yBACtB;wBACH,CAAC,CAAC;4BACE,uBAAuB;4BACvB,iBAAiB,EAAE,OAAO,CAAC,MAAM;yBAClC;iBACN,CACF,CAAC;gBAEF,WAAW,GAAG,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC;gBAExD,IAAI,YAAY,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAChC,kFAAkF;oBAClF,MAAM,UAAU,GAAG,WAAW,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBACvE,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,UAAU,IAAI,gCAAgC,YAAY,CAAC,QAAQ,EAAE;wBAC5E,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,mCAAmC;gBACnC,MAAM,OAAO,CAAC,WAAW,CAAC,SAAS,CAAC,CAAC;gBAErC,mCAAmC;gBACnC,MAAM,kBAAkB,CAAC,OAAO,CAAC,CAAC;gBAElC,qCAAqC;gBACrC,MAAM,UAAU,GAAG,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEpD,yBAAyB;gBACzB,MAAM,iBAAiB,GAAG,MAAM,aAAa,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;gBAE9E,0BAA0B;gBAC1B,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,OAAO,CAAC,CAAC;gBAE5D,OAAO;oBACL,OAAO,EAAE,iBAAiB,CAAC,SAAS;oBACpC,MAAM,EAAE,WAAW;oBACnB,UAAU;oBACV,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,UAAU,EAAE,iBAAiB,CAAC,IAAI;oBAClC,cAAc,EAAE,iBAAiB,CAAC,OAAO;oBACzC,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,cAAc;iBACf,CAAC;YACJ,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,6BAA6B;gBAC7B,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;qBAC9B,CAAC;gBACJ,CAAC;gBACD,OAAO;oBACL,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE,WAAW;oBACnB,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;oBAC7D,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;iBAC9B,CAAC;YACJ,CAAC;oBAAS,CAAC;gBACT,0BAA0B;gBAC1B,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;oBACnB,OAAO,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;gBAC5D,CAAC;gBACD,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,MAAM,OAAO,CAAC,IAAI,EAAE,CAAC;gBACvB,CAAC;YACH,CAAC;QACH,CAAC;KACF,CAAC;AACF,CAAC"}
1
+ {"version":3,"file":"claude-code.js","sourceRoot":"","sources":["../../../src/lib/agents/claude-code.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,cAAc,EACd,iBAAiB,GAClB,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,aAAa,EACb,qBAAqB,EACrB,kBAAkB,EAClB,UAAU,EACV,gBAAgB,GACjB,MAAM,aAAa,CAAC;AAErB;;;GAGG;AACH,KAAK,UAAU,iBAAiB,CAAC,OAAuB;IACtD,IAAI,CAAC;QACH,6DAA6D;QAC7D,MAAM,OAAO,GAAG,OAAO,CAAC,mBAAmB,EAAE,CAAC;QAC9C,4DAA4D;QAC5D,MAAM,WAAW,GAAG,OAAO,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;QAChD,MAAM,gBAAgB,GAAG,sBAAsB,WAAW,EAAE,CAAC;QAE7D,oDAAoD;QACpD,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,QAAQ,CACvC,SAAS,gBAAgB,gCAAgC,CAC1D,CAAC;QAEF,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;YAC3D,OAAO,SAAS,CAAC;QACnB,CAAC;QAED,MAAM,cAAc,GAAG,UAAU,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;QAChD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;QACvD,OAAO,OAAO,CAAC;IACjB,CAAC;IAAC,MAAM,CAAC;QACP,oCAAoC;QACpC,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB,CAAC,EAAE,kBAAkB,EAAmC;IAC3F,OAAO;QACL,IAAI,EAAE,kBAAkB,CAAC,CAAC,CAAC,+BAA+B,CAAC,CAAC,CAAC,aAAa;QAC1E,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC,iCAAiC,CAAC,CAAC,CAAC,aAAa;QAEnF,eAAe;YACb,OAAO,kBAAkB,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,gBAAgB,CAAC,YAAY,CAAC;QACtF,CAAC;QAED,eAAe;YACb,OAAO,MAAM,CAAC;QAChB,CAAC;QAED,KAAK,CAAC,GAAG,CAAC,WAAmB,EAAE,OAAwB;YACvD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,IAAI,OAAO,GAA0B,IAAI,CAAC;YAC1C,IAAI,WAAW,GAAG,EAAE,CAAC;YACrB,IAAI,OAAO,GAAG,KAAK,CAAC;YACpB,IAAI,cAAc,GAAG,KAAK,CAAC;YAE3B,sBAAsB;YACtB,MAAM,YAAY,GAAG,GAAG,EAAE;gBACxB,OAAO,GAAG,IAAI,CAAC;gBACf,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;gBACjC,CAAC;YACH,CAAC,CAAC;YAEF,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACnB,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;oBAC3B,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,sBAAsB;wBAC7B,QAAQ,EAAE,CAAC;qBACZ,CAAC;gBACJ,CAAC;gBACD,OAAO,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;YACzD,CAAC;YAED,IAAI,CAAC;gBACH,6BAA6B;gBAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,WAAW,CAAC,CAAC;gBACtD,MAAM,EAAE,cAAc,EAAE,SAAS,EAAE,GAAG,cAAc,CAAC,QAAQ,CAAC,CAAC;gBAE/D,8CAA8C;gBAC9C,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;qBACjC,CAAC;gBACJ,CAAC;gBAED,iBAAiB;gBACjB,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,CAAC;oBACpC,OAAO,EAAE,OAAO,CAAC,OAAO;oBACxB,OAAO,EAAE,QAAQ;iBAClB,CAAC,CAAC;gBAEH,8EAA8E;gBAC9E,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,2CAA2C;gBAC3C,MAAM,OAAO,CAAC,WAAW,CAAC,cAAc,CAAC,CAAC;gBAE1C,iCAAiC;gBACjC,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;oBAClB,MAAM,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;gBAC/B,CAAC;gBAED,uBAAuB;gBACvB,MAAM,aAAa,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;gBACnE,IAAI,aAAa,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACjC,MAAM,IAAI,KAAK,CAAC,uBAAuB,aAAa,CAAC,MAAM,EAAE,CAAC,CAAC;gBACjE,CAAC;gBAED,mCAAmC;gBACnC,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE;oBACjD,SAAS;oBACT,IAAI;oBACJ,2BAA2B;iBAC5B,CAAC,CAAC;gBACH,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC9B,MAAM,IAAI,KAAK,CAAC,+BAA+B,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;gBACtE,CAAC;gBAED,kCAAkC;gBAClC,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEjC,0BAA0B;gBAC1B,MAAM,cAAc,GAAG,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE;;6LAEwI,CAAC;gBAExL,kDAAkD;gBAClD,MAAM,YAAY,GAAG,MAAM,OAAO,CAAC,UAAU,CAC3C,QAAQ,EACR,CAAC,SAAS,EAAE,SAAS,EAAE,OAAO,CAAC,KAAK,EAAE,gCAAgC,EAAE,cAAc,CAAC,EACvF;oBACE,GAAG,EAAE,kBAAkB;wBACrB,CAAC,CAAC;4BACE,2CAA2C;4BAC3C,kBAAkB,EAAE,UAAU,CAAC,OAAO;4BACtC,oBAAoB,EAAE,OAAO,CAAC,MAAM;4BACpC,iBAAiB,EAAE,EAAE;yBACtB;wBACH,CAAC,CAAC;4BACE,uBAAuB;4BACvB,iBAAiB,EAAE,OAAO,CAAC,MAAM;yBAClC;iBACN,CACF,CAAC;gBAEF,WAAW,GAAG,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC;gBAExD,IAAI,YAAY,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAChC,kFAAkF;oBAClF,MAAM,UAAU,GAAG,WAAW,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBACvE,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,UAAU,IAAI,gCAAgC,YAAY,CAAC,QAAQ,EAAE;wBAC5E,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,mCAAmC;gBACnC,MAAM,OAAO,CAAC,WAAW,CAAC,SAAS,CAAC,CAAC;gBAErC,uCAAuC;gBACvC,MAAM,kBAAkB,CAAC,OAAO,CAAC,CAAC;gBAElC,qCAAqC;gBACrC,MAAM,UAAU,GAAG,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEpD,yBAAyB;gBACzB,MAAM,iBAAiB,GAAG,MAAM,aAAa,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;gBAE9E,0BAA0B;gBAC1B,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,OAAO,CAAC,CAAC;gBAE5D,OAAO;oBACL,OAAO,EAAE,iBAAiB,CAAC,SAAS;oBACpC,MAAM,EAAE,WAAW;oBACnB,UAAU;oBACV,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,UAAU,EAAE,iBAAiB,CAAC,IAAI;oBAClC,cAAc,EAAE,iBAAiB,CAAC,OAAO;oBACzC,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,cAAc;iBACf,CAAC;YACJ,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,6BAA6B;gBAC7B,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;qBAC9B,CAAC;gBACJ,CAAC;gBACD,OAAO;oBACL,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE,WAAW;oBACnB,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;oBAC7D,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;iBAC9B,CAAC;YACJ,CAAC;oBAAS,CAAC;gBACT,0BAA0B;gBAC1B,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;oBACnB,OAAO,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;gBAC5D,CAAC;gBACD,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,MAAM,OAAO,CAAC,IAAI,EAAE,CAAC;gBACvB,CAAC;YACH,CAAC;QACH,CAAC;KACF,CAAC;AACF,CAAC"}
@@ -193,7 +193,7 @@ IMPORTANT: Do not run npm, pnpm, yarn, or any package manager commands. Dependen
193
193
  }
194
194
  // Upload test files for validation
195
195
  await sandbox.uploadFiles(testFiles);
196
- // Create vitest config for EVAL.ts
196
+ // Create vitest config for EVAL.ts/tsx
197
197
  await createVitestConfig(sandbox);
198
198
  // Extract transcript from the Codex JSON output (--json flag outputs JSONL)
199
199
  const transcript = extractTranscriptFromOutput(agentOutput);
@@ -1 +1 @@
1
- {"version":3,"file":"codex.js","sourceRoot":"","sources":["../../../src/lib/agents/codex.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,cAAc,EACd,iBAAiB,GAClB,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,aAAa,EACb,qBAAqB,EACrB,kBAAkB,EAClB,UAAU,EACV,aAAa,GACd,MAAM,aAAa,CAAC;AAErB;;;GAGG;AACH,SAAS,2BAA2B,CAAC,MAAc;IACjD,IAAI,CAAC,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;QAC9B,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,8DAA8D;IAC9D,2DAA2D;IAC3D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE;QAC7C,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QAC5B,OAAO,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;IAEH,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED;;;GAGG;AACH,SAAS,mBAAmB,CAAC,KAAa,EAAE,kBAA2B,EAAE,MAAe;IACtF,MAAM,SAAS,GAAG,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,UAAU,KAAK,EAAE,CAAC;IAElE,IAAI,kBAAkB,EAAE,CAAC;QACvB,OAAO;;;;;cAKG,UAAU,CAAC,aAAa;aACzB,UAAU,CAAC,YAAY;;;;;WAKzB,SAAS;CACnB,CAAC;IACA,CAAC;SAAM,CAAC;QACN,wDAAwD;QACxD,6DAA6D;QAC7D,OAAO;;;;;cAKG,aAAa,CAAC,OAAO;aACtB,MAAM;;;;;WAKR,SAAS;CACnB,CAAC;IACA,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAE,kBAAkB,EAAmC;IACtF,OAAO;QACL,IAAI,EAAE,kBAAkB,CAAC,CAAC,CAAC,yBAAyB,CAAC,CAAC,CAAC,OAAO;QAC9D,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC,kCAAkC,CAAC,CAAC,CAAC,cAAc;QAErF,eAAe;YACb,OAAO,kBAAkB,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,aAAa,CAAC,YAAY,CAAC;QACnF,CAAC;QAED,eAAe;YACb,OAAO,sBAAsB,CAAC;QAChC,CAAC;QAED,KAAK,CAAC,GAAG,CAAC,WAAmB,EAAE,OAAwB;YACvD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,IAAI,OAAO,GAA0B,IAAI,CAAC;YAC1C,IAAI,WAAW,GAAG,EAAE,CAAC;YACrB,IAAI,OAAO,GAAG,KAAK,CAAC;YACpB,IAAI,cAAc,GAAG,KAAK,CAAC;YAE3B,sBAAsB;YACtB,MAAM,YAAY,GAAG,GAAG,EAAE;gBACxB,OAAO,GAAG,IAAI,CAAC;gBACf,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;gBACjC,CAAC;YACH,CAAC,CAAC;YAEF,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACnB,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;oBAC3B,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,sBAAsB;wBAC7B,QAAQ,EAAE,CAAC;qBACZ,CAAC;gBACJ,CAAC;gBACD,OAAO,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;YACzD,CAAC;YAED,IAAI,CAAC;gBACH,6BAA6B;gBAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,WAAW,CAAC,CAAC;gBACtD,MAAM,EAAE,cAAc,EAAE,SAAS,EAAE,GAAG,cAAc,CAAC,QAAQ,CAAC,CAAC;gBAE/D,8CAA8C;gBAC9C,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;qBACjC,CAAC;gBACJ,CAAC;gBAED,iBAAiB;gBACjB,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,CAAC;oBACpC,OAAO,EAAE,OAAO,CAAC,OAAO;oBACxB,OAAO,EAAE,QAAQ;iBAClB,CAAC,CAAC;gBAEH,8EAA8E;gBAC9E,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,2CAA2C;gBAC3C,MAAM,OAAO,CAAC,WAAW,CAAC,cAAc,CAAC,CAAC;gBAE1C,iCAAiC;gBACjC,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;oBAClB,MAAM,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;gBAC/B,CAAC;gBAED,uBAAuB;gBACvB,MAAM,aAAa,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;gBACnE,IAAI,aAAa,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACjC,MAAM,IAAI,KAAK,CAAC,uBAAuB,aAAa,CAAC,MAAM,EAAE,CAAC,CAAC;gBACjE,CAAC;gBAED,6BAA6B;gBAC7B,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE;oBACjD,SAAS;oBACT,IAAI;oBACJ,eAAe;iBAChB,CAAC,CAAC;gBACH,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC9B,MAAM,IAAI,KAAK,CAAC,6BAA6B,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;gBACpE,CAAC;gBAED,gDAAgD;gBAChD,MAAM,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAC,CAAC;gBAC5C,MAAM,aAAa,GAAG,mBAAmB,CACvC,OAAO,CAAC,KAAK,EACb,kBAAkB,EAClB,kBAAkB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAChD,CAAC;gBACF,MAAM,OAAO,CAAC,QAAQ,CAAC;EAC3B,aAAa;IACX,CAAC,CAAC;gBAEA,kCAAkC;gBAClC,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEjC,0BAA0B;gBAC1B,MAAM,cAAc,GAAG,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE;;6LAEwI,CAAC;gBAExL,8DAA8D;gBAC9D,yFAAyF;gBACzF,6FAA6F;gBAC7F,oEAAoE;gBACpE,MAAM,WAAW,GAAG,MAAM,OAAO,CAAC,UAAU,CAC1C,OAAO,EACP;oBACE,MAAM;oBACN,4CAA4C;oBAC5C,QAAQ;oBACR,uBAAuB;oBACvB,cAAc;iBACf,EACD;oBACE,GAAG,EAAE,kBAAkB;wBACrB,CAAC,CAAC;4BACE,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;yBAC1C;wBACH,CAAC,CAAC;4BACE,CAAC,aAAa,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;yBAC7C;iBACN,CACF,CAAC;gBAEF,WAAW,GAAG,WAAW,CAAC,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC;gBAEtD,IAAI,WAAW,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC/B,kFAAkF;oBAClF,MAAM,UAAU,GAAG,WAAW,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBACvE,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,UAAU,IAAI,8BAA8B,WAAW,CAAC,QAAQ,EAAE;wBACzE,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,mCAAmC;gBACnC,MAAM,OAAO,CAAC,WAAW,CAAC,SAAS,CAAC,CAAC;gBAErC,mCAAmC;gBACnC,MAAM,kBAAkB,CAAC,OAAO,CAAC,CAAC;gBAElC,4EAA4E;gBAC5E,MAAM,UAAU,GAAG,2BAA2B,CAAC,WAAW,CAAC,CAAC;gBAE5D,yBAAyB;gBACzB,MAAM,iBAAiB,GAAG,MAAM,aAAa,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;gBAE9E,0BAA0B;gBAC1B,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,OAAO,CAAC,CAAC;gBAE5D,OAAO;oBACL,OAAO,EAAE,iBAAiB,CAAC,SAAS;oBACpC,MAAM,EAAE,WAAW;oBACnB,UAAU;oBACV,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,UAAU,EAAE,iBAAiB,CAAC,IAAI;oBAClC,cAAc,EAAE,iBAAiB,CAAC,OAAO;oBACzC,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,cAAc;iBACf,CAAC;YACJ,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,6BAA6B;gBAC7B,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;qBAC9B,CAAC;gBACJ,CAAC;gBACD,OAAO;oBACL,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE,WAAW;oBACnB,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;oBAC7D,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;iBAC9B,CAAC;YACJ,CAAC;oBAAS,CAAC;gBACT,0BAA0B;gBAC1B,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;oBACnB,OAAO,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;gBAC5D,CAAC;gBACD,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,MAAM,OAAO,CAAC,IAAI,EAAE,CAAC;gBACvB,CAAC;YACH,CAAC;QACH,CAAC;KACF,CAAC;AACF,CAAC"}
1
+ {"version":3,"file":"codex.js","sourceRoot":"","sources":["../../../src/lib/agents/codex.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,cAAc,EACd,iBAAiB,GAClB,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,aAAa,EACb,qBAAqB,EACrB,kBAAkB,EAClB,UAAU,EACV,aAAa,GACd,MAAM,aAAa,CAAC;AAErB;;;GAGG;AACH,SAAS,2BAA2B,CAAC,MAAc;IACjD,IAAI,CAAC,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;QAC9B,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,8DAA8D;IAC9D,2DAA2D;IAC3D,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE;QAC7C,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;QAC5B,OAAO,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;IAEH,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED;;;GAGG;AACH,SAAS,mBAAmB,CAAC,KAAa,EAAE,kBAA2B,EAAE,MAAe;IACtF,MAAM,SAAS,GAAG,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,UAAU,KAAK,EAAE,CAAC;IAElE,IAAI,kBAAkB,EAAE,CAAC;QACvB,OAAO;;;;;cAKG,UAAU,CAAC,aAAa;aACzB,UAAU,CAAC,YAAY;;;;;WAKzB,SAAS;CACnB,CAAC;IACA,CAAC;SAAM,CAAC;QACN,wDAAwD;QACxD,6DAA6D;QAC7D,OAAO;;;;;cAKG,aAAa,CAAC,OAAO;aACtB,MAAM;;;;;WAKR,SAAS;CACnB,CAAC;IACA,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAE,kBAAkB,EAAmC;IACtF,OAAO;QACL,IAAI,EAAE,kBAAkB,CAAC,CAAC,CAAC,yBAAyB,CAAC,CAAC,CAAC,OAAO;QAC9D,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC,kCAAkC,CAAC,CAAC,CAAC,cAAc;QAErF,eAAe;YACb,OAAO,kBAAkB,CAAC,CAAC,CAAC,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,aAAa,CAAC,YAAY,CAAC;QACnF,CAAC;QAED,eAAe;YACb,OAAO,sBAAsB,CAAC;QAChC,CAAC;QAED,KAAK,CAAC,GAAG,CAAC,WAAmB,EAAE,OAAwB;YACvD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,IAAI,OAAO,GAA0B,IAAI,CAAC;YAC1C,IAAI,WAAW,GAAG,EAAE,CAAC;YACrB,IAAI,OAAO,GAAG,KAAK,CAAC;YACpB,IAAI,cAAc,GAAG,KAAK,CAAC;YAE3B,sBAAsB;YACtB,MAAM,YAAY,GAAG,GAAG,EAAE;gBACxB,OAAO,GAAG,IAAI,CAAC;gBACf,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,OAAO,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;gBACjC,CAAC;YACH,CAAC,CAAC;YAEF,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;gBACnB,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;oBAC3B,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,sBAAsB;wBAC7B,QAAQ,EAAE,CAAC;qBACZ,CAAC;gBACJ,CAAC;gBACD,OAAO,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;YACzD,CAAC;YAED,IAAI,CAAC;gBACH,6BAA6B;gBAC7B,MAAM,QAAQ,GAAG,MAAM,iBAAiB,CAAC,WAAW,CAAC,CAAC;gBACtD,MAAM,EAAE,cAAc,EAAE,SAAS,EAAE,GAAG,cAAc,CAAC,QAAQ,CAAC,CAAC;gBAE/D,8CAA8C;gBAC9C,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;qBACjC,CAAC;gBACJ,CAAC;gBAED,iBAAiB;gBACjB,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,CAAC;oBACpC,OAAO,EAAE,OAAO,CAAC,OAAO;oBACxB,OAAO,EAAE,QAAQ;iBAClB,CAAC,CAAC;gBAEH,8EAA8E;gBAC9E,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,EAAE;wBACV,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,2CAA2C;gBAC3C,MAAM,OAAO,CAAC,WAAW,CAAC,cAAc,CAAC,CAAC;gBAE1C,iCAAiC;gBACjC,IAAI,OAAO,CAAC,KAAK,EAAE,CAAC;oBAClB,MAAM,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;gBAC/B,CAAC;gBAED,uBAAuB;gBACvB,MAAM,aAAa,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,SAAS,CAAC,CAAC,CAAC;gBACnE,IAAI,aAAa,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACjC,MAAM,IAAI,KAAK,CAAC,uBAAuB,aAAa,CAAC,MAAM,EAAE,CAAC,CAAC;gBACjE,CAAC;gBAED,6BAA6B;gBAC7B,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE;oBACjD,SAAS;oBACT,IAAI;oBACJ,eAAe;iBAChB,CAAC,CAAC;gBACH,IAAI,UAAU,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC9B,MAAM,IAAI,KAAK,CAAC,6BAA6B,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;gBACpE,CAAC;gBAED,gDAAgD;gBAChD,MAAM,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAC,CAAC;gBAC5C,MAAM,aAAa,GAAG,mBAAmB,CACvC,OAAO,CAAC,KAAK,EACb,kBAAkB,EAClB,kBAAkB,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAChD,CAAC;gBACF,MAAM,OAAO,CAAC,QAAQ,CAAC;EAC3B,aAAa;IACX,CAAC,CAAC;gBAEA,kCAAkC;gBAClC,MAAM,iBAAiB,CAAC,OAAO,CAAC,CAAC;gBAEjC,0BAA0B;gBAC1B,MAAM,cAAc,GAAG,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE;;6LAEwI,CAAC;gBAExL,8DAA8D;gBAC9D,yFAAyF;gBACzF,6FAA6F;gBAC7F,oEAAoE;gBACpE,MAAM,WAAW,GAAG,MAAM,OAAO,CAAC,UAAU,CAC1C,OAAO,EACP;oBACE,MAAM;oBACN,4CAA4C;oBAC5C,QAAQ;oBACR,uBAAuB;oBACvB,cAAc;iBACf,EACD;oBACE,GAAG,EAAE,kBAAkB;wBACrB,CAAC,CAAC;4BACE,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;yBAC1C;wBACH,CAAC,CAAC;4BACE,CAAC,aAAa,CAAC,YAAY,CAAC,EAAE,OAAO,CAAC,MAAM;yBAC7C;iBACN,CACF,CAAC;gBAEF,WAAW,GAAG,WAAW,CAAC,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC;gBAEtD,IAAI,WAAW,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBAC/B,kFAAkF;oBAClF,MAAM,UAAU,GAAG,WAAW,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBACvE,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,UAAU,IAAI,8BAA8B,WAAW,CAAC,QAAQ,EAAE;wBACzE,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,CAAC,SAAS;qBAC7B,CAAC;gBACJ,CAAC;gBAED,mCAAmC;gBACnC,MAAM,OAAO,CAAC,WAAW,CAAC,SAAS,CAAC,CAAC;gBAErC,uCAAuC;gBACvC,MAAM,kBAAkB,CAAC,OAAO,CAAC,CAAC;gBAElC,4EAA4E;gBAC5E,MAAM,UAAU,GAAG,2BAA2B,CAAC,WAAW,CAAC,CAAC;gBAE5D,yBAAyB;gBACzB,MAAM,iBAAiB,GAAG,MAAM,aAAa,CAAC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;gBAE9E,0BAA0B;gBAC1B,MAAM,cAAc,GAAG,MAAM,qBAAqB,CAAC,OAAO,CAAC,CAAC;gBAE5D,OAAO;oBACL,OAAO,EAAE,iBAAiB,CAAC,SAAS;oBACpC,MAAM,EAAE,WAAW;oBACnB,UAAU;oBACV,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,UAAU,EAAE,iBAAiB,CAAC,IAAI;oBAClC,cAAc,EAAE,iBAAiB,CAAC,OAAO;oBACzC,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,cAAc;iBACf,CAAC;YACJ,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,6BAA6B;gBAC7B,IAAI,OAAO,EAAE,CAAC;oBACZ,OAAO;wBACL,OAAO,EAAE,KAAK;wBACd,MAAM,EAAE,WAAW;wBACnB,KAAK,EAAE,SAAS;wBAChB,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;wBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;qBAC9B,CAAC;gBACJ,CAAC;gBACD,OAAO;oBACL,OAAO,EAAE,KAAK;oBACd,MAAM,EAAE,WAAW;oBACnB,KAAK,EAAE,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;oBAC7D,QAAQ,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAChC,SAAS,EAAE,OAAO,EAAE,SAAS;iBAC9B,CAAC;YACJ,CAAC;oBAAS,CAAC;gBACT,0BAA0B;gBAC1B,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;oBACnB,OAAO,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,YAAY,CAAC,CAAC;gBAC5D,CAAC;gBACD,IAAI,OAAO,IAAI,CAAC,cAAc,EAAE,CAAC;oBAC/B,cAAc,GAAG,IAAI,CAAC;oBACtB,MAAM,OAAO,CAAC,IAAI,EAAE,CAAC;gBACvB,CAAC;YACH,CAAC;QACH,CAAC;KACF,CAAC;AACF,CAAC"}
@@ -20,7 +20,7 @@ export declare function runValidation(sandbox: SandboxManager, scripts: string[]
20
20
  */
21
21
  export declare function captureGeneratedFiles(sandbox: SandboxManager): Promise<Record<string, string>>;
22
22
  /**
23
- * Create vitest config for running EVAL.ts.
23
+ * Create vitest config for running EVAL.ts or EVAL.tsx.
24
24
  */
25
25
  export declare function createVitestConfig(sandbox: SandboxManager): Promise<void>;
26
26
  /**
@@ -1 +1 @@
1
- {"version":3,"file":"shared.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/shared.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAC/C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAEpD;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,SAAS,EAAE,OAAO,CAAC;IACnB,IAAI,CAAC,EAAE,YAAY,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;CACvC;AAED;;GAEG;AACH,wBAAsB,aAAa,CACjC,OAAO,EAAE,cAAc,EACvB,OAAO,EAAE,MAAM,EAAE,GAChB,OAAO,CAAC,iBAAiB,CAAC,CAgC5B;AAED;;GAEG;AACH,wBAAsB,qBAAqB,CAAC,OAAO,EAAE,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CA2BpG;AAED;;GAEG;AACH,wBAAsB,kBAAkB,CAAC,OAAO,EAAE,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAY/E;AAED;;GAEG;AACH,eAAO,MAAM,UAAU;;;;CAIb,CAAC;AAEX;;GAEG;AACH,eAAO,MAAM,gBAAgB;;CAEnB,CAAC;AAEX;;GAEG;AACH,eAAO,MAAM,aAAa;;;CAGhB,CAAC"}
1
+ {"version":3,"file":"shared.d.ts","sourceRoot":"","sources":["../../../src/lib/agents/shared.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAC/C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAEpD;;GAEG;AACH,MAAM,WAAW,iBAAiB;IAChC,SAAS,EAAE,OAAO,CAAC;IACnB,IAAI,CAAC,EAAE,YAAY,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;CACvC;AAgCD;;GAEG;AACH,wBAAsB,aAAa,CACjC,OAAO,EAAE,cAAc,EACvB,OAAO,EAAE,MAAM,EAAE,GAChB,OAAO,CAAC,iBAAiB,CAAC,CAmC5B;AAED;;GAEG;AACH,wBAAsB,qBAAqB,CAAC,OAAO,EAAE,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CA2BpG;AAED;;GAEG;AACH,wBAAsB,kBAAkB,CAAC,OAAO,EAAE,cAAc,GAAG,OAAO,CAAC,IAAI,CAAC,CAe/E;AAED;;GAEG;AACH,eAAO,MAAM,UAAU;;;;CAIb,CAAC;AAEX;;GAEG;AACH,eAAO,MAAM,gBAAgB;;CAEnB,CAAC;AAEX;;GAEG;AACH,eAAO,MAAM,aAAa;;;CAGhB,CAAC"}
@@ -1,6 +1,33 @@
1
1
  /**
2
2
  * Shared utilities for agent implementations.
3
3
  */
4
+ /**
5
+ * Detect which eval file exists in the sandbox (EVAL.ts or EVAL.tsx).
6
+ * Case-sensitive: Only matches exact uppercase filenames.
7
+ * Returns the filename if found, or 'EVAL.ts' as fallback.
8
+ */
9
+ async function detectEvalFile(sandbox) {
10
+ try {
11
+ // List files in current directory and check for exact case match
12
+ const lsResult = await sandbox.runShell('ls -1');
13
+ if (lsResult.exitCode === 0) {
14
+ const files = lsResult.stdout.split('\n').map((f) => f.trim());
15
+ // Check for EVAL.tsx first (prefer JSX if both exist)
16
+ if (files.includes('EVAL.tsx')) {
17
+ return 'EVAL.tsx';
18
+ }
19
+ // Check for EVAL.ts
20
+ if (files.includes('EVAL.ts')) {
21
+ return 'EVAL.ts';
22
+ }
23
+ }
24
+ }
25
+ catch {
26
+ // Ignore errors
27
+ }
28
+ // Default to EVAL.ts (will fail later if it doesn't exist)
29
+ return 'EVAL.ts';
30
+ }
4
31
  /**
5
32
  * Run validation scripts in the sandbox.
6
33
  */
@@ -9,8 +36,10 @@ export async function runValidation(sandbox, scripts) {
9
36
  allPassed: true,
10
37
  scripts: {},
11
38
  };
12
- // Always run vitest for EVAL.ts (explicitly specify the file)
13
- const testResult = await sandbox.runCommand('npx', ['vitest', 'run', 'EVAL.ts']);
39
+ // Detect which eval file exists (EVAL.ts or EVAL.tsx)
40
+ const evalFile = await detectEvalFile(sandbox);
41
+ // Always run vitest for the eval file (explicitly specify the file)
42
+ const testResult = await sandbox.runCommand('npx', ['vitest', 'run', evalFile]);
14
43
  results.test = {
15
44
  success: testResult.exitCode === 0,
16
45
  output: testResult.stdout + testResult.stderr,
@@ -60,15 +89,17 @@ export async function captureGeneratedFiles(sandbox) {
60
89
  return files;
61
90
  }
62
91
  /**
63
- * Create vitest config for running EVAL.ts.
92
+ * Create vitest config for running EVAL.ts or EVAL.tsx.
64
93
  */
65
94
  export async function createVitestConfig(sandbox) {
95
+ // Detect which eval file exists
96
+ const evalFile = await detectEvalFile(sandbox);
66
97
  await sandbox.writeFiles({
67
98
  'vitest.config.ts': `
68
99
  import { defineConfig } from 'vitest/config';
69
100
  export default defineConfig({
70
101
  test: {
71
- include: ['EVAL.ts'],
102
+ include: ['${evalFile}'],
72
103
  globals: false,
73
104
  },
74
105
  });
@@ -1 +1 @@
1
- {"version":3,"file":"shared.js","sourceRoot":"","sources":["../../../src/lib/agents/shared.ts"],"names":[],"mappings":"AAAA;;GAEG;AAcH;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,OAAuB,EACvB,OAAiB;IAEjB,MAAM,OAAO,GAAsB;QACjC,SAAS,EAAE,IAAI;QACf,OAAO,EAAE,EAAE;KACZ,CAAC;IAEF,8DAA8D;IAC9D,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;IACjF,OAAO,CAAC,IAAI,GAAG;QACb,OAAO,EAAE,UAAU,CAAC,QAAQ,KAAK,CAAC;QAClC,MAAM,EAAE,UAAU,CAAC,MAAM,GAAG,UAAU,CAAC,MAAM;KAC9C,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;QAC1B,OAAO,CAAC,SAAS,GAAG,KAAK,CAAC;IAC5B,CAAC;IAED,yBAAyB;IACzB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,YAAY,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC;QACtE,MAAM,MAAM,GAAiB;YAC3B,OAAO,EAAE,YAAY,CAAC,QAAQ,KAAK,CAAC;YACpC,MAAM,EAAE,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM;SAClD,CAAC;QAEF,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC;QAEjC,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,OAAO,CAAC,SAAS,GAAG,KAAK,CAAC;QAC5B,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAAC,OAAuB;IACjE,MAAM,KAAK,GAA2B,EAAE,CAAC;IAEzC,IAAI,CAAC;QACH,wBAAwB;QACxB,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,QAAQ,CACvC,+HAA+H,CAChI,CAAC;QAEF,MAAM,SAAS,GAAG,UAAU,CAAC,MAAM;aAChC,IAAI,EAAE;aACN,KAAK,CAAC,IAAI,CAAC;aACX,MAAM,CAAC,OAAO,CAAC,CAAC;QAEnB,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,IAAI,CAAC;gBACH,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;gBACjD,KAAK,CAAC,QAAQ,CAAC,GAAG,OAAO,CAAC;YAC5B,CAAC;YAAC,MAAM,CAAC;gBACP,wBAAwB;YAC1B,CAAC;QACH,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,wCAAwC;IAC1C,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,OAAuB;IAC9D,MAAM,OAAO,CAAC,UAAU,CAAC;QACvB,kBAAkB,EAAE;;;;;;;;CAQvB;KACE,CAAC,CAAC;AACL,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,MAAM,UAAU,GAAG;IACxB,OAAO,EAAE,8BAA8B;IACvC,aAAa,EAAE,iCAAiC;IAChD,YAAY,EAAE,oBAAoB;CAC1B,CAAC;AAEX;;GAEG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAG;IAC9B,YAAY,EAAE,mBAAmB;CACzB,CAAC;AAEX;;GAEG;AACH,MAAM,CAAC,MAAM,aAAa,GAAG;IAC3B,OAAO,EAAE,2BAA2B;IACpC,YAAY,EAAE,gBAAgB;CACtB,CAAC"}
1
+ {"version":3,"file":"shared.js","sourceRoot":"","sources":["../../../src/lib/agents/shared.ts"],"names":[],"mappings":"AAAA;;GAEG;AAcH;;;;GAIG;AACH,KAAK,UAAU,cAAc,CAAC,OAAuB;IACnD,IAAI,CAAC;QACH,iEAAiE;QACjE,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QACjD,IAAI,QAAQ,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;YAC5B,MAAM,KAAK,GAAG,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;YAE/D,sDAAsD;YACtD,IAAI,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,EAAE,CAAC;gBAC/B,OAAO,UAAU,CAAC;YACpB,CAAC;YAED,oBAAoB;YACpB,IAAI,KAAK,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;gBAC9B,OAAO,SAAS,CAAC;YACnB,CAAC;QACH,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,gBAAgB;IAClB,CAAC;IAED,2DAA2D;IAC3D,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,OAAuB,EACvB,OAAiB;IAEjB,MAAM,OAAO,GAAsB;QACjC,SAAS,EAAE,IAAI;QACf,OAAO,EAAE,EAAE;KACZ,CAAC;IAEF,sDAAsD;IACtD,MAAM,QAAQ,GAAG,MAAM,cAAc,CAAC,OAAO,CAAC,CAAC;IAE/C,oEAAoE;IACpE,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE,KAAK,EAAE,QAAQ,CAAC,CAAC,CAAC;IAChF,OAAO,CAAC,IAAI,GAAG;QACb,OAAO,EAAE,UAAU,CAAC,QAAQ,KAAK,CAAC;QAClC,MAAM,EAAE,UAAU,CAAC,MAAM,GAAG,UAAU,CAAC,MAAM;KAC9C,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;QAC1B,OAAO,CAAC,SAAS,GAAG,KAAK,CAAC;IAC5B,CAAC;IAED,yBAAyB;IACzB,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,YAAY,GAAG,MAAM,OAAO,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC;QACtE,MAAM,MAAM,GAAiB;YAC3B,OAAO,EAAE,YAAY,CAAC,QAAQ,KAAK,CAAC;YACpC,MAAM,EAAE,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM;SAClD,CAAC;QAEF,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC;QAEjC,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,OAAO,CAAC,SAAS,GAAG,KAAK,CAAC;QAC5B,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAAC,OAAuB;IACjE,MAAM,KAAK,GAA2B,EAAE,CAAC;IAEzC,IAAI,CAAC;QACH,wBAAwB;QACxB,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,QAAQ,CACvC,+HAA+H,CAChI,CAAC;QAEF,MAAM,SAAS,GAAG,UAAU,CAAC,MAAM;aAChC,IAAI,EAAE;aACN,KAAK,CAAC,IAAI,CAAC;aACX,MAAM,CAAC,OAAO,CAAC,CAAC;QAEnB,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,IAAI,CAAC;gBACH,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;gBACjD,KAAK,CAAC,QAAQ,CAAC,GAAG,OAAO,CAAC;YAC5B,CAAC;YAAC,MAAM,CAAC;gBACP,wBAAwB;YAC1B,CAAC;QACH,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,wCAAwC;IAC1C,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,OAAuB;IAC9D,gCAAgC;IAChC,MAAM,QAAQ,GAAG,MAAM,cAAc,CAAC,OAAO,CAAC,CAAC;IAE/C,MAAM,OAAO,CAAC,UAAU,CAAC;QACvB,kBAAkB,EAAE;;;;iBAIP,QAAQ;;;;CAIxB;KACE,CAAC,CAAC;AACL,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,MAAM,UAAU,GAAG;IACxB,OAAO,EAAE,8BAA8B;IACvC,aAAa,EAAE,iCAAiC;IAChD,YAAY,EAAE,oBAAoB;CAC1B,CAAC;AAEX;;GAEG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAG;IAC9B,YAAY,EAAE,mBAAmB;CACzB,CAAC;AAEX;;GAEG;AACH,MAAM,CAAC,MAAM,aAAa,GAAG;IAC3B,OAAO,EAAE,2BAA2B;IACpC,YAAY,EAAE,gBAAgB;CACtB,CAAC"}
@@ -15,9 +15,10 @@ export declare class FixtureValidationError extends Error {
15
15
  */
16
16
  export declare function discoverFixtures(evalsDir: string): string[];
17
17
  /**
18
- * Validates that a fixture has all required files.
18
+ * Validates that a fixture has all required files with correct case.
19
19
  * Returns an array of missing file names, or empty array if valid.
20
20
  * Note: Accepts either EVAL.ts or EVAL.tsx for the eval file.
21
+ * Case-sensitive: 'prompt.md' will fail even on Mac/Windows.
21
22
  */
22
23
  export declare function validateFixtureFiles(fixturePath: string): string[];
23
24
  /**
@@ -1 +1 @@
1
- {"version":3,"file":"fixture.d.ts","sourceRoot":"","sources":["../../src/lib/fixture.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAG9C;;GAEG;AACH,qBAAa,sBAAuB,SAAQ,KAAK;IAEtC,WAAW,EAAE,MAAM;gBAAnB,WAAW,EAAE,MAAM,EAC1B,OAAO,EAAE,MAAM;CAKlB;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,EAAE,CAyB3D;AAED;;;;GAIG;AACH,wBAAgB,oBAAoB,CAAC,WAAW,EAAE,MAAM,GAAG,MAAM,EAAE,CAoBlE;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,WAAW,EAAE,MAAM,GAAG;IAAE,QAAQ,EAAE,OAAO,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,CAwB9F;AAED;;GAEG;AACH,wBAAgB,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,WAAW,CAgCvE;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,MAAM,GAAG;IACjD,QAAQ,EAAE,WAAW,EAAE,CAAC;IACxB,MAAM,EAAE,sBAAsB,EAAE,CAAC;CAClC,CAmBA;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAC7B,WAAW,EAAE,MAAM,EACnB,eAAe,GAAE,SAAS,MAAM,EAAmB,GAClD,MAAM,EAAE,CA2BV;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAC9B,WAAW,EAAE,MAAM,EACnB,eAAe,CAAC,EAAE,SAAS,MAAM,EAAE,GAClC,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAUrB"}
1
+ {"version":3,"file":"fixture.d.ts","sourceRoot":"","sources":["../../src/lib/fixture.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAG9C;;GAEG;AACH,qBAAa,sBAAuB,SAAQ,KAAK;IAEtC,WAAW,EAAE,MAAM;gBAAnB,WAAW,EAAE,MAAM,EAC1B,OAAO,EAAE,MAAM;CAKlB;AAeD;;;GAGG;AACH,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,EAAE,CAyB3D;AAED;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,WAAW,EAAE,MAAM,GAAG,MAAM,EAAE,CAmBlE;AAED;;;GAGG;AACH,wBAAgB,mBAAmB,CAAC,WAAW,EAAE,MAAM,GAAG;IAAE,QAAQ,EAAE,OAAO,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,CAwB9F;AAED;;GAEG;AACH,wBAAgB,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,WAAW,CAmCvE;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAAC,QAAQ,EAAE,MAAM,GAAG;IACjD,QAAQ,EAAE,WAAW,EAAE,CAAC;IACxB,MAAM,EAAE,sBAAsB,EAAE,CAAC;CAClC,CAmBA;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAC7B,WAAW,EAAE,MAAM,EACnB,eAAe,GAAE,SAAS,MAAM,EAAmB,GAClD,MAAM,EAAE,CA2BV;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAC9B,WAAW,EAAE,MAAM,EACnB,eAAe,CAAC,EAAE,SAAS,MAAM,EAAE,GAClC,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAUrB"}
@@ -15,6 +15,19 @@ export class FixtureValidationError extends Error {
15
15
  this.name = 'FixtureValidationError';
16
16
  }
17
17
  }
18
+ /**
19
+ * Check if a file exists with exact case match (case-sensitive even on Mac/Windows).
20
+ * Returns true only if the file exists AND the case matches exactly.
21
+ */
22
+ function existsWithExactCase(dirPath, fileName) {
23
+ try {
24
+ const files = readdirSync(dirPath);
25
+ return files.includes(fileName);
26
+ }
27
+ catch {
28
+ return false;
29
+ }
30
+ }
18
31
  /**
19
32
  * Discovers all eval fixtures in a directory.
20
33
  * Each subdirectory is considered a potential eval fixture.
@@ -40,24 +53,24 @@ export function discoverFixtures(evalsDir) {
40
53
  return fixtures.sort();
41
54
  }
42
55
  /**
43
- * Validates that a fixture has all required files.
56
+ * Validates that a fixture has all required files with correct case.
44
57
  * Returns an array of missing file names, or empty array if valid.
45
58
  * Note: Accepts either EVAL.ts or EVAL.tsx for the eval file.
59
+ * Case-sensitive: 'prompt.md' will fail even on Mac/Windows.
46
60
  */
47
61
  export function validateFixtureFiles(fixturePath) {
48
62
  const missing = [];
49
63
  for (const file of REQUIRED_EVAL_FILES) {
50
- // Special case: Accept either EVAL.ts or EVAL.tsx
64
+ // Special case: Accept either EVAL.ts or EVAL.tsx (both case-sensitive)
51
65
  if (file === 'EVAL.ts') {
52
- const hasEvalTs = existsSync(join(fixturePath, 'EVAL.ts'));
53
- const hasEvalTsx = existsSync(join(fixturePath, 'EVAL.tsx'));
66
+ const hasEvalTs = existsWithExactCase(fixturePath, 'EVAL.ts');
67
+ const hasEvalTsx = existsWithExactCase(fixturePath, 'EVAL.tsx');
54
68
  if (!hasEvalTs && !hasEvalTsx) {
55
69
  missing.push('EVAL.ts or EVAL.tsx');
56
70
  }
57
71
  }
58
72
  else {
59
- const filePath = join(fixturePath, file);
60
- if (!existsSync(filePath)) {
73
+ if (!existsWithExactCase(fixturePath, file)) {
61
74
  missing.push(file);
62
75
  }
63
76
  }
@@ -109,7 +122,10 @@ export function loadFixture(evalsDir, name) {
109
122
  if (pkgValidation.error) {
110
123
  throw new FixtureValidationError(name, pkgValidation.error);
111
124
  }
112
- // Read prompt
125
+ // Read prompt (case-sensitive check)
126
+ if (!existsWithExactCase(fixturePath, 'PROMPT.md')) {
127
+ throw new FixtureValidationError(name, 'PROMPT.md not found (case-sensitive: must be uppercase)');
128
+ }
113
129
  const promptPath = join(fixturePath, 'PROMPT.md');
114
130
  const prompt = readFileSync(promptPath, 'utf-8');
115
131
  return {
@@ -1 +1 @@
1
- {"version":3,"file":"fixture.js","sourceRoot":"","sources":["../../src/lib/fixture.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AACrE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,MAAM,CAAC;AAErC,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAEjE;;GAEG;AACH,MAAM,OAAO,sBAAuB,SAAQ,KAAK;IAEtC;IADT,YACS,WAAmB,EAC1B,OAAe;QAEf,KAAK,CAAC,SAAS,WAAW,MAAM,OAAO,EAAE,CAAC,CAAC;QAHpC,gBAAW,GAAX,WAAW,CAAQ;QAI1B,IAAI,CAAC,IAAI,GAAG,wBAAwB,CAAC;IACvC,CAAC;CACF;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,QAAgB;IAC/C,MAAM,YAAY,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAEvC,IAAI,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,CAAC;QAC9B,MAAM,IAAI,KAAK,CAAC,8BAA8B,YAAY,EAAE,CAAC,CAAC;IAChE,CAAC;IAED,MAAM,OAAO,GAAG,WAAW,CAAC,YAAY,CAAC,CAAC;IAC1C,MAAM,QAAQ,GAAa,EAAE,CAAC;IAE9B,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC5B,MAAM,SAAS,GAAG,IAAI,CAAC,YAAY,EAAE,KAAK,CAAC,CAAC;QAE5C,oCAAoC;QACpC,IAAI,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YAC1B,SAAS;QACX,CAAC;QAED,4BAA4B;QAC5B,IAAI,QAAQ,CAAC,SAAS,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC;YACtC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACvB,CAAC;IACH,CAAC;IAED,OAAO,QAAQ,CAAC,IAAI,EAAE,CAAC;AACzB,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,oBAAoB,CAAC,WAAmB;IACtD,MAAM,OAAO,GAAa,EAAE,CAAC;IAE7B,KAAK,MAAM,IAAI,IAAI,mBAAmB,EAAE,CAAC;QACvC,kDAAkD;QAClD,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;YACvB,MAAM,SAAS,GAAG,UAAU,CAAC,IAAI,CAAC,WAAW,EAAE,SAAS,CAAC,CAAC,CAAC;YAC3D,MAAM,UAAU,GAAG,UAAU,CAAC,IAAI,CAAC,WAAW,EAAE,UAAU,CAAC,CAAC,CAAC;YAC7D,IAAI,CAAC,SAAS,IAAI,CAAC,UAAU,EAAE,CAAC;gBAC9B,OAAO,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;YACtC,CAAC;QACH,CAAC;aAAM,CAAC;YACN,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;YACzC,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;gBAC1B,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACrB,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,WAAmB;IACrD,MAAM,eAAe,GAAG,IAAI,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC;IAE1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,YAAY,CAAC,eAAe,EAAE,OAAO,CAAC,CAAC;QACvD,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAEhC,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YAC1B,OAAO;gBACL,QAAQ,EAAE,KAAK;gBACf,KAAK,EAAE,yCAAyC;aACjD,CAAC;QACJ,CAAC;QAED,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;IAC5B,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,WAAW,EAAE,CAAC;YACjC,OAAO;gBACL,QAAQ,EAAE,KAAK;gBACf,KAAK,EAAE,gCAAgC;aACxC,CAAC;QACJ,CAAC;QACD,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,WAAW,CAAC,QAAgB,EAAE,IAAY;IACxD,MAAM,WAAW,GAAG,OAAO,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IAE5C,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;QAC7B,MAAM,IAAI,sBAAsB,CAAC,IAAI,EAAE,wBAAwB,WAAW,EAAE,CAAC,CAAC;IAChF,CAAC;IAED,0BAA0B;IAC1B,MAAM,YAAY,GAAG,oBAAoB,CAAC,WAAW,CAAC,CAAC;IACvD,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,sBAAsB,CAC9B,IAAI,EACJ,2BAA2B,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CACrD,CAAC;IACJ,CAAC;IAED,wBAAwB;IACxB,MAAM,aAAa,GAAG,mBAAmB,CAAC,WAAW,CAAC,CAAC;IACvD,IAAI,aAAa,CAAC,KAAK,EAAE,CAAC;QACxB,MAAM,IAAI,sBAAsB,CAAC,IAAI,EAAE,aAAa,CAAC,KAAK,CAAC,CAAC;IAC9D,CAAC;IAED,cAAc;IACd,MAAM,UAAU,GAAG,IAAI,CAAC,WAAW,EAAE,WAAW,CAAC,CAAC;IAClD,MAAM,MAAM,GAAG,YAAY,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;IAEjD,OAAO;QACL,IAAI;QACJ,IAAI,EAAE,WAAW;QACjB,MAAM;QACN,QAAQ,EAAE,aAAa,CAAC,QAAQ;KACjC,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,QAAgB;IAI9C,MAAM,YAAY,GAAG,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IAChD,MAAM,QAAQ,GAAkB,EAAE,CAAC;IACnC,MAAM,MAAM,GAA6B,EAAE,CAAC;IAE5C,KAAK,MAAM,IAAI,IAAI,YAAY,EAAE,CAAC;QAChC,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,WAAW,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;YAC5C,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACzB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,KAAK,YAAY,sBAAsB,EAAE,CAAC;gBAC5C,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YACrB,CAAC;iBAAM,CAAC;gBACN,MAAM,KAAK,CAAC;YACd,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;AAC9B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAC7B,WAAmB,EACnB,kBAAqC,cAAc;IAEnD,MAAM,KAAK,GAAa,EAAE,CAAC;IAE3B,SAAS,IAAI,CAAC,GAAW,EAAE,WAAmB,EAAE;QAC9C,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;QAEjC,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,MAAM,YAAY,GAAG,QAAQ,CAAC,CAAC,CAAC,GAAG,QAAQ,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC;YAE/D,8BAA8B;YAC9B,IAAI,eAAe,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,YAAY,KAAK,OAAO,IAAI,KAAK,KAAK,OAAO,CAAC,EAAE,CAAC;gBACrF,SAAS;YACX,CAAC;YAED,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YAClC,MAAM,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC;YAEhC,IAAI,IAAI,CAAC,WAAW,EAAE,EAAE,CAAC;gBACvB,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC;YAC/B,CAAC;iBAAM,CAAC;gBACN,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;YAC3B,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,CAAC,WAAW,CAAC,CAAC;IAClB,OAAO,KAAK,CAAC,IAAI,EAAE,CAAC;AACtB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAC9B,WAAmB,EACnB,eAAmC;IAEnC,MAAM,KAAK,GAAG,eAAe,CAAC,WAAW,EAAE,eAAe,CAAC,CAAC;IAC5D,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;IAE3C,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;QACzC,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAC;IACtD,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC"}
1
+ {"version":3,"file":"fixture.js","sourceRoot":"","sources":["../../src/lib/fixture.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,YAAY,EAAE,WAAW,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AACrE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,MAAM,CAAC;AAErC,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAEjE;;GAEG;AACH,MAAM,OAAO,sBAAuB,SAAQ,KAAK;IAEtC;IADT,YACS,WAAmB,EAC1B,OAAe;QAEf,KAAK,CAAC,SAAS,WAAW,MAAM,OAAO,EAAE,CAAC,CAAC;QAHpC,gBAAW,GAAX,WAAW,CAAQ;QAI1B,IAAI,CAAC,IAAI,GAAG,wBAAwB,CAAC;IACvC,CAAC;CACF;AAED;;;GAGG;AACH,SAAS,mBAAmB,CAAC,OAAe,EAAE,QAAgB;IAC5D,IAAI,CAAC;QACH,MAAM,KAAK,GAAG,WAAW,CAAC,OAAO,CAAC,CAAC;QACnC,OAAO,KAAK,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,QAAgB;IAC/C,MAAM,YAAY,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC;IAEvC,IAAI,CAAC,UAAU,CAAC,YAAY,CAAC,EAAE,CAAC;QAC9B,MAAM,IAAI,KAAK,CAAC,8BAA8B,YAAY,EAAE,CAAC,CAAC;IAChE,CAAC;IAED,MAAM,OAAO,GAAG,WAAW,CAAC,YAAY,CAAC,CAAC;IAC1C,MAAM,QAAQ,GAAa,EAAE,CAAC;IAE9B,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC5B,MAAM,SAAS,GAAG,IAAI,CAAC,YAAY,EAAE,KAAK,CAAC,CAAC;QAE5C,oCAAoC;QACpC,IAAI,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;YAC1B,SAAS;QACX,CAAC;QAED,4BAA4B;QAC5B,IAAI,QAAQ,CAAC,SAAS,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC;YACtC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACvB,CAAC;IACH,CAAC;IAED,OAAO,QAAQ,CAAC,IAAI,EAAE,CAAC;AACzB,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,oBAAoB,CAAC,WAAmB;IACtD,MAAM,OAAO,GAAa,EAAE,CAAC;IAE7B,KAAK,MAAM,IAAI,IAAI,mBAAmB,EAAE,CAAC;QACvC,wEAAwE;QACxE,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;YACvB,MAAM,SAAS,GAAG,mBAAmB,CAAC,WAAW,EAAE,SAAS,CAAC,CAAC;YAC9D,MAAM,UAAU,GAAG,mBAAmB,CAAC,WAAW,EAAE,UAAU,CAAC,CAAC;YAChE,IAAI,CAAC,SAAS,IAAI,CAAC,UAAU,EAAE,CAAC;gBAC9B,OAAO,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;YACtC,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,mBAAmB,CAAC,WAAW,EAAE,IAAI,CAAC,EAAE,CAAC;gBAC5C,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACrB,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,mBAAmB,CAAC,WAAmB;IACrD,MAAM,eAAe,GAAG,IAAI,CAAC,WAAW,EAAE,cAAc,CAAC,CAAC;IAE1D,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,YAAY,CAAC,eAAe,EAAE,OAAO,CAAC,CAAC;QACvD,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAEhC,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;YAC1B,OAAO;gBACL,QAAQ,EAAE,KAAK;gBACf,KAAK,EAAE,yCAAyC;aACjD,CAAC;QACJ,CAAC;QAED,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;IAC5B,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,WAAW,EAAE,CAAC;YACjC,OAAO;gBACL,QAAQ,EAAE,KAAK;gBACf,KAAK,EAAE,gCAAgC;aACxC,CAAC;QACJ,CAAC;QACD,MAAM,KAAK,CAAC;IACd,CAAC;AACH,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,WAAW,CAAC,QAAgB,EAAE,IAAY;IACxD,MAAM,WAAW,GAAG,OAAO,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IAE5C,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;QAC7B,MAAM,IAAI,sBAAsB,CAAC,IAAI,EAAE,wBAAwB,WAAW,EAAE,CAAC,CAAC;IAChF,CAAC;IAED,0BAA0B;IAC1B,MAAM,YAAY,GAAG,oBAAoB,CAAC,WAAW,CAAC,CAAC;IACvD,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,sBAAsB,CAC9B,IAAI,EACJ,2BAA2B,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CACrD,CAAC;IACJ,CAAC;IAED,wBAAwB;IACxB,MAAM,aAAa,GAAG,mBAAmB,CAAC,WAAW,CAAC,CAAC;IACvD,IAAI,aAAa,CAAC,KAAK,EAAE,CAAC;QACxB,MAAM,IAAI,sBAAsB,CAAC,IAAI,EAAE,aAAa,CAAC,KAAK,CAAC,CAAC;IAC9D,CAAC;IAED,qCAAqC;IACrC,IAAI,CAAC,mBAAmB,CAAC,WAAW,EAAE,WAAW,CAAC,EAAE,CAAC;QACnD,MAAM,IAAI,sBAAsB,CAAC,IAAI,EAAE,yDAAyD,CAAC,CAAC;IACpG,CAAC;IACD,MAAM,UAAU,GAAG,IAAI,CAAC,WAAW,EAAE,WAAW,CAAC,CAAC;IAClD,MAAM,MAAM,GAAG,YAAY,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;IAEjD,OAAO;QACL,IAAI;QACJ,IAAI,EAAE,WAAW;QACjB,MAAM;QACN,QAAQ,EAAE,aAAa,CAAC,QAAQ;KACjC,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,QAAgB;IAI9C,MAAM,YAAY,GAAG,gBAAgB,CAAC,QAAQ,CAAC,CAAC;IAChD,MAAM,QAAQ,GAAkB,EAAE,CAAC;IACnC,MAAM,MAAM,GAA6B,EAAE,CAAC;IAE5C,KAAK,MAAM,IAAI,IAAI,YAAY,EAAE,CAAC;QAChC,IAAI,CAAC;YACH,MAAM,OAAO,GAAG,WAAW,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;YAC5C,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACzB,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,KAAK,YAAY,sBAAsB,EAAE,CAAC;gBAC5C,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YACrB,CAAC;iBAAM,CAAC;gBACN,MAAM,KAAK,CAAC;YACd,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;AAC9B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAC7B,WAAmB,EACnB,kBAAqC,cAAc;IAEnD,MAAM,KAAK,GAAa,EAAE,CAAC;IAE3B,SAAS,IAAI,CAAC,GAAW,EAAE,WAAmB,EAAE;QAC9C,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;QAEjC,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,MAAM,YAAY,GAAG,QAAQ,CAAC,CAAC,CAAC,GAAG,QAAQ,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC;YAE/D,8BAA8B;YAC9B,IAAI,eAAe,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,YAAY,KAAK,OAAO,IAAI,KAAK,KAAK,OAAO,CAAC,EAAE,CAAC;gBACrF,SAAS;YACX,CAAC;YAED,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YAClC,MAAM,IAAI,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC;YAEhC,IAAI,IAAI,CAAC,WAAW,EAAE,EAAE,CAAC;gBACvB,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC;YAC/B,CAAC;iBAAM,CAAC;gBACN,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;YAC3B,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,CAAC,WAAW,CAAC,CAAC;IAClB,OAAO,KAAK,CAAC,IAAI,EAAE,CAAC;AACtB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAC9B,WAAmB,EACnB,eAAmC;IAEnC,MAAM,KAAK,GAAG,eAAe,CAAC,WAAW,EAAE,eAAe,CAAC,CAAC;IAC5D,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;IAE3C,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,QAAQ,GAAG,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;QACzC,QAAQ,CAAC,GAAG,CAAC,IAAI,EAAE,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,CAAC;IACtD,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@vercel/agent-eval",
3
- "version": "0.0.2",
3
+ "version": "0.0.4",
4
4
  "description": "Framework for testing AI coding agents in isolated sandboxes",
5
5
  "repository": {
6
6
  "type": "git",