vskill 0.2.68 → 0.2.70

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.
@@ -34,8 +34,8 @@ export async function runEvalGenerateAll(root, force) {
34
34
  try {
35
35
  const skillContent = readFileSync(skillMdPath, "utf-8");
36
36
  const prompt = buildEvalInitPrompt(skillContent);
37
- const raw = await client.generate("You generate eval test cases for AI skills. Output only valid JSON in a code fence.", prompt);
38
- const evalsFile = parseGeneratedEvals(raw);
37
+ const genResult = await client.generate("You generate eval test cases for AI skills. Output only valid JSON in a code fence.", prompt);
38
+ const evalsFile = parseGeneratedEvals(genResult.text);
39
39
  mkdirSync(join(skill.dir, "evals"), { recursive: true });
40
40
  writeFileSync(evalsPath, JSON.stringify(evalsFile, null, 2), "utf-8");
41
41
  generated++;
@@ -1 +1 @@
1
- {"version":3,"file":"generate-all.js","sourceRoot":"","sources":["../../../src/commands/eval/generate-all.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,uEAAuE;AACvE,8EAA8E;AAE9E,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,mBAAmB,EAAE,mBAAmB,EAAE,MAAM,8BAA8B,CAAC;AACxF,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,MAAM,uBAAuB,CAAC;AAEtE,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,IAAY,EACZ,KAAc;IAEd,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,IAAI,CAAC,CAAC;IAEtC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,qBAAqB,GAAG,IAAI,CAAC,CAAC,CAAC;QAC/C,OAAO;IACT,CAAC;IAED,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;IACjC,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,MAAM,WAAW,GAAa,EAAE,CAAC;IAEjC,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC3B,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,OAAO,EAAE,YAAY,CAAC,CAAC;QAEzD,8CAA8C;QAC9C,IAAI,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,EAAE,CAAC;YAC7B,OAAO,EAAE,CAAC;YACV,SAAS;QACX,CAAC;QAED,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;QAChD,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;YAC7B,MAAM,EAAE,CAAC;YACT,WAAW,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,KAAK,gBAAgB,CAAC,CAAC;YACjE,SAAS;QACX,CAAC;QAED,IAAI,CAAC;YACH,MAAM,YAAY,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;YACxD,MAAM,MAAM,GAAG,mBAAmB,CAAC,YAAY,CAAC,CAAC;YAEjD,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,QAAQ,CAC/B,qFAAqF,EACrF,MAAM,CACP,CAAC;YAEF,MAAM,SAAS,GAAG,mBAAmB,CAAC,GAAG,CAAC,CAAC;YAE3C,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,OAAO,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YACzD,aAAa,CAAC,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;YAEtE,SAAS,EAAE,CAAC;YACZ,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,gBAAgB,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;YAElE,iEAAiE;YACjE,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;QAChD,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,EAAE,CAAC;YACT,WAAW,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;YACnD,OAAO,CAAC,KAAK,CACX,GAAG,CAAC,aAAa,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,KAAK,KAAK,CAAC;gBAChD,GAAG,CAAE,GAAa,CAAC,OAAO,CAAC,CAC9B,CAAC;QACJ,CAAC;IACH,CAAC;IAED,gBAAgB;IAChB,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,4BAA4B,CAAC,CAAC,CAAC;IAChD,OAAO,CAAC,GAAG,CAAC,cAAc,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IAC3C,OAAO,CAAC,GAAG,CAAC,KAAK,KAAK,CAAC,cAAc,SAAS,EAAE,CAAC,EAAE,CAAC,CAAC;IACrD,OAAO,CAAC,GAAG,CAAC,KAAK,MAAM,CAAC,YAAY,OAAO,EAAE,CAAC,EAAE,CAAC,CAAC;IAClD,OAAO,CAAC,GAAG,CAAC,KAAK,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,MAAM,EAAE,CAAC,CAAC,CAAC,CAAC,WAAW,MAAM,EAAE,EAAE,CAAC,CAAC;IAEhF,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC3B,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,CAAC;QACrC,KAAK,MAAM,IAAI,IAAI,WAAW,EAAE,CAAC;YAC/B,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QAClC,CAAC;IACH,CAAC;AACH,CAAC"}
1
+ {"version":3,"file":"generate-all.js","sourceRoot":"","sources":["../../../src/commands/eval/generate-all.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,uEAAuE;AACvE,8EAA8E;AAE9E,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,UAAU,EAAE,MAAM,6BAA6B,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,mBAAmB,EAAE,mBAAmB,EAAE,MAAM,8BAA8B,CAAC;AACxF,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,MAAM,uBAAuB,CAAC;AAEtE,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,IAAY,EACZ,KAAc;IAEd,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,IAAI,CAAC,CAAC;IAEtC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,qBAAqB,GAAG,IAAI,CAAC,CAAC,CAAC;QAC/C,OAAO;IACT,CAAC;IAED,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;IACjC,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,MAAM,WAAW,GAAa,EAAE,CAAC;IAEjC,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;QAC3B,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,OAAO,EAAE,YAAY,CAAC,CAAC;QAEzD,8CAA8C;QAC9C,IAAI,KAAK,CAAC,QAAQ,IAAI,CAAC,KAAK,EAAE,CAAC;YAC7B,OAAO,EAAE,CAAC;YACV,SAAS;QACX,CAAC;QAED,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;QAChD,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;YAC7B,MAAM,EAAE,CAAC;YACT,WAAW,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,KAAK,gBAAgB,CAAC,CAAC;YACjE,SAAS;QACX,CAAC;QAED,IAAI,CAAC;YACH,MAAM,YAAY,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;YACxD,MAAM,MAAM,GAAG,mBAAmB,CAAC,YAAY,CAAC,CAAC;YAEjD,MAAM,SAAS,GAAG,MAAM,MAAM,CAAC,QAAQ,CACrC,qFAAqF,EACrF,MAAM,CACP,CAAC;YAEF,MAAM,SAAS,GAAG,mBAAmB,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;YAEtD,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,OAAO,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YACzD,aAAa,CAAC,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;YAEtE,SAAS,EAAE,CAAC;YACZ,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,gBAAgB,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;YAElE,iEAAiE;YACjE,MAAM,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC;QAChD,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,EAAE,CAAC;YACT,WAAW,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;YACnD,OAAO,CAAC,KAAK,CACX,GAAG,CAAC,aAAa,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,KAAK,KAAK,CAAC;gBAChD,GAAG,CAAE,GAAa,CAAC,OAAO,CAAC,CAC9B,CAAC;QACJ,CAAC;IACH,CAAC;IAED,gBAAgB;IAChB,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,4BAA4B,CAAC,CAAC,CAAC;IAChD,OAAO,CAAC,GAAG,CAAC,cAAc,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;IAC3C,OAAO,CAAC,GAAG,CAAC,KAAK,KAAK,CAAC,cAAc,SAAS,EAAE,CAAC,EAAE,CAAC,CAAC;IACrD,OAAO,CAAC,GAAG,CAAC,KAAK,MAAM,CAAC,YAAY,OAAO,EAAE,CAAC,EAAE,CAAC,CAAC;IAClD,OAAO,CAAC,GAAG,CAAC,KAAK,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,WAAW,MAAM,EAAE,CAAC,CAAC,CAAC,CAAC,WAAW,MAAM,EAAE,EAAE,CAAC,CAAC;IAEhF,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC3B,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,CAAC;QACrC,KAAK,MAAM,IAAI,IAAI,WAAW,EAAE,CAAC;YAC/B,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QAClC,CAAC;IACH,CAAC;AACH,CAAC"}
@@ -24,8 +24,8 @@ export async function runEvalInit(skillDir, force) {
24
24
  const prompt = buildEvalInitPrompt(skillContent);
25
25
  try {
26
26
  const client = createLlmClient();
27
- const raw = await client.generate("You generate eval test cases for AI skills. Output only valid JSON in a code fence.", prompt);
28
- const evalsFile = parseGeneratedEvals(raw);
27
+ const genResult = await client.generate("You generate eval test cases for AI skills. Output only valid JSON in a code fence.", prompt);
28
+ const evalsFile = parseGeneratedEvals(genResult.text);
29
29
  mkdirSync(evalsDir, { recursive: true });
30
30
  writeFileSync(evalsPath, JSON.stringify(evalsFile, null, 2), "utf-8");
31
31
  console.log(green(`Created ${evalsPath}`));
@@ -1 +1 @@
1
- {"version":3,"file":"init.js","sourceRoot":"","sources":["../../../src/commands/eval/init.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,gEAAgE;AAChE,8EAA8E;AAE9E,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,mBAAmB,EAAE,mBAAmB,EAAE,MAAM,8BAA8B,CAAC;AACxF,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,GAAG,EAAE,MAAM,EAAE,MAAM,uBAAuB,CAAC;AAEhE,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,QAAgB,EAChB,KAAc;IAEd,MAAM,WAAW,GAAG,IAAI,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;IAC/C,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IACzC,MAAM,SAAS,GAAG,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC;IAE/C,wBAAwB;IACxB,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;QAC7B,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,yBAAyB,WAAW,EAAE,CAAC,CAAC,CAAC;QAC3D,OAAO;IACT,CAAC;IAED,4BAA4B;IAC5B,IAAI,UAAU,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC;QACpC,OAAO,CAAC,GAAG,CACT,MAAM,CAAC,qDAAqD,CAAC,CAC9D,CAAC;QACF,OAAO;IACT,CAAC;IAED,MAAM,YAAY,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;IACxD,MAAM,MAAM,GAAG,mBAAmB,CAAC,YAAY,CAAC,CAAC;IAEjD,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,QAAQ,CAC/B,qFAAqF,EACrF,MAAM,CACP,CAAC;QAEF,MAAM,SAAS,GAAG,mBAAmB,CAAC,GAAG,CAAC,CAAC;QAE3C,SAAS,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACzC,aAAa,CAAC,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;QAEtE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,WAAW,SAAS,EAAE,CAAC,CAAC,CAAC;QAC3C,OAAO,CAAC,GAAG,CACT,GAAG,CAAC,KAAK,SAAS,CAAC,KAAK,CAAC,MAAM,gBAAgB,SAAS,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC,aAAa,CAAC,CAC9H,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,OAAO,CAAC,KAAK,CACX,GAAG,CAAC,4BAA4B,CAAC,GAAG,GAAG,CAAE,GAAa,CAAC,OAAO,CAAC,CAChE,CAAC;IACJ,CAAC;AACH,CAAC"}
1
+ {"version":3,"file":"init.js","sourceRoot":"","sources":["../../../src/commands/eval/init.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,gEAAgE;AAChE,8EAA8E;AAE9E,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,mBAAmB,EAAE,mBAAmB,EAAE,MAAM,8BAA8B,CAAC;AACxF,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,GAAG,EAAE,MAAM,EAAE,MAAM,uBAAuB,CAAC;AAEhE,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,QAAgB,EAChB,KAAc;IAEd,MAAM,WAAW,GAAG,IAAI,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;IAC/C,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IACzC,MAAM,SAAS,GAAG,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,CAAC;IAE/C,wBAAwB;IACxB,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;QAC7B,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,yBAAyB,WAAW,EAAE,CAAC,CAAC,CAAC;QAC3D,OAAO;IACT,CAAC;IAED,4BAA4B;IAC5B,IAAI,UAAU,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC;QACpC,OAAO,CAAC,GAAG,CACT,MAAM,CAAC,qDAAqD,CAAC,CAC9D,CAAC;QACF,OAAO;IACT,CAAC;IAED,MAAM,YAAY,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;IACxD,MAAM,MAAM,GAAG,mBAAmB,CAAC,YAAY,CAAC,CAAC;IAEjD,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,SAAS,GAAG,MAAM,MAAM,CAAC,QAAQ,CACrC,qFAAqF,EACrF,MAAM,CACP,CAAC;QAEF,MAAM,SAAS,GAAG,mBAAmB,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;QAEtD,SAAS,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACzC,aAAa,CAAC,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;QAEtE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,WAAW,SAAS,EAAE,CAAC,CAAC,CAAC;QAC3C,OAAO,CAAC,GAAG,CACT,GAAG,CAAC,KAAK,SAAS,CAAC,KAAK,CAAC,MAAM,gBAAgB,SAAS,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC,aAAa,CAAC,CAC9H,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,OAAO,CAAC,KAAK,CACX,GAAG,CAAC,4BAA4B,CAAC,GAAG,GAAG,CAAE,GAAa,CAAC,OAAO,CAAC,CAChE,CAAC;IACJ,CAAC;AACH,CAAC"}
@@ -54,13 +54,13 @@ export async function runEvalRun(skillDir) {
54
54
  try {
55
55
  // Step 1: Send prompt to LLM
56
56
  process.stdout.write(dim(`[${i + 1}/${total}] ${evalCase.name} — generating...`));
57
- const output = await client.generate(systemPrompt, evalCase.prompt);
57
+ const genResult = await client.generate(systemPrompt, evalCase.prompt);
58
58
  process.stdout.write(dim(` judging ${evalCase.assertions.length} assertions...`));
59
59
  // Step 2: Judge each assertion
60
60
  const assertionResults = [];
61
61
  let passCount = 0;
62
62
  for (const assertion of evalCase.assertions) {
63
- const result = await judgeAssertion(output, assertion, client);
63
+ const result = await judgeAssertion(genResult.text, assertion, client);
64
64
  assertionResults.push(result);
65
65
  if (result.pass)
66
66
  passCount++;
@@ -1 +1 @@
1
- {"version":3,"file":"run.js","sourceRoot":"","sources":["../../../src/commands/eval/run.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,6DAA6D;AAC7D,8EAA8E;AAE9E,OAAO,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,oBAAoB,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AACjF,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AACrD,OAAO,EAAE,cAAc,EAAE,MAAM,yBAAyB,CAAC;AAEzD,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC;AAE7E,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,QAAgB;IAC/C,+BAA+B;IAC/B,IAAI,SAAS,CAAC;IACd,IAAI,CAAC;QACH,SAAS,GAAG,oBAAoB,CAAC,QAAQ,CAAC,CAAC;IAC7C,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,mBAAmB,EAAE,CAAC;YACvC,MAAM,QAAQ,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,IAAI,EAAE,CAAC;YAC9C,IAAI,QAAQ,CAAC,QAAQ,CAAC,eAAe,CAAC,EAAE,CAAC;gBACvC,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,0BAA0B,QAAQ,mBAAmB,CAAC,CAAC,CAAC;YAC5E,CAAC;iBAAM,CAAC;gBACN,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,uBAAuB,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YAC3D,CAAC;QACH,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,wBAAyB,GAAa,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACvE,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAChB,OAAO;IACT,CAAC;IAED,8CAA8C;IAC9C,MAAM,WAAW,GAAG,IAAI,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;IAC/C,IAAI,YAAY,GAAG,EAAE,CAAC;IACtB,IAAI,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;QAC5B,YAAY,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;IACpD,CAAC;SAAM,CAAC;QACN,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,iCAAiC,WAAW,wCAAwC,CAAC,CAAC,CAAC;IAC9G,CAAC;IAED,MAAM,YAAY,GAAG,YAAY;QAC/B,CAAC,CAAC,8HAA8H,YAAY,OAAO;QACnJ,CAAC,CAAC,sDAAsD,CAAC;IAE3D,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;IACjC,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;IAC3B,MAAM,KAAK,GAAG,SAAS,CAAC,KAAK,CAAC,MAAM,CAAC;IACrC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,aAAa,KAAK,MAAM,KAAK,aAAa,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;IACrF,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,UAAU,YAAY,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC;IAEtE,MAAM,cAAc,GAAoB,EAAE,CAAC;IAC3C,MAAM,SAAS,GAAe,EAAE,CAAC;IAEjC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAChD,MAAM,QAAQ,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QACpC,IAAI,CAAC;YACH,6BAA6B;YAC7B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,KAAK,KAAK,QAAQ,CAAC,IAAI,kBAAkB,CAAC,CAAC,CAAC;YAClF,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,YAAY,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;YACpE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,YAAY,QAAQ,CAAC,UAAU,CAAC,MAAM,gBAAgB,CAAC,CAAC,CAAC;YAElF,+BAA+B;YAC/B,MAAM,gBAAgB,GAAG,EAAE,CAAC;YAC5B,IAAI,SAAS,GAAG,CAAC,CAAC;YAElB,KAAK,MAAM,SAAS,IAAI,QAAQ,CAAC,UAAU,EAAE,CAAC;gBAC5C,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;gBAC/D,gBAAgB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBAC9B,IAAI,MAAM,CAAC,IAAI;oBAAE,SAAS,EAAE,CAAC;gBAE7B,MAAM,aAAa,GACjB,SAAS,CAAC,IAAI,CAAC,MAAM,GAAG,EAAE;oBACxB,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,KAAK;oBACrC,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC;gBAErB,SAAS,CAAC,IAAI,CAAC;oBACb,QAAQ,CAAC,IAAI;oBACb,SAAS,CAAC,EAAE;oBACZ,aAAa;oBACb,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC;iBAC1C,CAAC,CAAC;YACL,CAAC;YAED,MAAM,QAAQ,GAAG,QAAQ,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC;gBAC7C,CAAC,CAAC,SAAS,GAAG,QAAQ,CAAC,UAAU,CAAC,MAAM;gBACxC,CAAC,CAAC,CAAC,CAAC;YACN,MAAM,SAAS,GAAG,SAAS,KAAK,QAAQ,CAAC,UAAU,CAAC,MAAM,CAAC;YAC3D,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,SAAS,IAAI,QAAQ,CAAC,UAAU,CAAC,MAAM,SAAS,CAAC,CAAC,CAAC;YAEpG,cAAc,CAAC,IAAI,CAAC;gBAClB,OAAO,EAAE,QAAQ,CAAC,EAAE;gBACpB,SAAS,EAAE,QAAQ,CAAC,IAAI;gBACxB,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;gBACnC,aAAa,EAAE,IAAI;gBACnB,SAAS,EAAE,QAAQ;gBACnB,UAAU,EAAE,gBAAgB;aAC7B,CAAC,CAAC;QACL,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC;YAC9B,cAAc,CAAC,IAAI,CAAC;gBAClB,OAAO,EAAE,QAAQ,CAAC,EAAE;gBACpB,SAAS,EAAE,QAAQ,CAAC,IAAI;gBACxB,MAAM,EAAE,OAAO;gBACf,aAAa,EAAG,GAAa,CAAC,OAAO;gBACrC,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,EAAE;aACf,CAAC,CAAC;YAEH,SAAS,CAAC,IAAI,CAAC;gBACb,QAAQ,CAAC,IAAI;gBACb,GAAG;gBACH,GAAG,CAAC,SAAS,GAAI,GAAa,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBACpD,MAAM,CAAC,OAAO,CAAC;aAChB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,sBAAsB;IACtB,MAAM,OAAO,GAAG,CAAC,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC;IACxD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,mBAAmB,SAAS,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC;IAC/D,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC;IAEvC,kBAAkB;IAClB,MAAM,MAAM,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,MAAM,CAAC;IACxE,MAAM,MAAM,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,MAAM,CAAC;IACxE,MAAM,MAAM,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IACzE,OAAO,CAAC,GAAG,CACT,KAAK,KAAK,CAAC,GAAG,MAAM,SAAS,CAAC,IAAI,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,EAAE,CACrI,CAAC;IAEF,uBAAuB;IACvB,MAAM,SAAS,GAAoB;QACjC,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACnC,KAAK;QACL,UAAU,EAAE,SAAS,CAAC,UAAU;QAChC,KAAK,EAAE,cAAc;KACtB,CAAC;IAEF,MAAM,cAAc,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IAC1C,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,0BAA0B,QAAQ,uBAAuB,CAAC,CAAC,CAAC;AAC9E,CAAC"}
1
+ {"version":3,"file":"run.js","sourceRoot":"","sources":["../../../src/commands/eval/run.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,6DAA6D;AAC7D,8EAA8E;AAE9E,OAAO,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,oBAAoB,EAAE,mBAAmB,EAAE,MAAM,sBAAsB,CAAC;AACjF,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AACrD,OAAO,EAAE,cAAc,EAAE,MAAM,yBAAyB,CAAC;AAEzD,OAAO,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC;AAE7E,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,QAAgB;IAC/C,+BAA+B;IAC/B,IAAI,SAAS,CAAC;IACd,IAAI,CAAC;QACH,SAAS,GAAG,oBAAoB,CAAC,QAAQ,CAAC,CAAC;IAC7C,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,mBAAmB,EAAE,CAAC;YACvC,MAAM,QAAQ,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,IAAI,EAAE,CAAC;YAC9C,IAAI,QAAQ,CAAC,QAAQ,CAAC,eAAe,CAAC,EAAE,CAAC;gBACvC,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,0BAA0B,QAAQ,mBAAmB,CAAC,CAAC,CAAC;YAC5E,CAAC;iBAAM,CAAC;gBACN,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,uBAAuB,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;YAC3D,CAAC;QACH,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,wBAAyB,GAAa,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QACvE,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAChB,OAAO;IACT,CAAC;IAED,8CAA8C;IAC9C,MAAM,WAAW,GAAG,IAAI,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;IAC/C,IAAI,YAAY,GAAG,EAAE,CAAC;IACtB,IAAI,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;QAC5B,YAAY,GAAG,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;IACpD,CAAC;SAAM,CAAC;QACN,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,iCAAiC,WAAW,wCAAwC,CAAC,CAAC,CAAC;IAC9G,CAAC;IAED,MAAM,YAAY,GAAG,YAAY;QAC/B,CAAC,CAAC,8HAA8H,YAAY,OAAO;QACnJ,CAAC,CAAC,sDAAsD,CAAC;IAE3D,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;IACjC,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC;IAC3B,MAAM,KAAK,GAAG,SAAS,CAAC,KAAK,CAAC,MAAM,CAAC;IACrC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,aAAa,KAAK,MAAM,KAAK,aAAa,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;IACrF,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,UAAU,YAAY,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,CAAC;IAEtE,MAAM,cAAc,GAAoB,EAAE,CAAC;IAC3C,MAAM,SAAS,GAAe,EAAE,CAAC;IAEjC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAChD,MAAM,QAAQ,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QACpC,IAAI,CAAC;YACH,6BAA6B;YAC7B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,KAAK,KAAK,QAAQ,CAAC,IAAI,kBAAkB,CAAC,CAAC,CAAC;YAClF,MAAM,SAAS,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,YAAY,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;YACvE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,YAAY,QAAQ,CAAC,UAAU,CAAC,MAAM,gBAAgB,CAAC,CAAC,CAAC;YAElF,+BAA+B;YAC/B,MAAM,gBAAgB,GAAG,EAAE,CAAC;YAC5B,IAAI,SAAS,GAAG,CAAC,CAAC;YAElB,KAAK,MAAM,SAAS,IAAI,QAAQ,CAAC,UAAU,EAAE,CAAC;gBAC5C,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,SAAS,CAAC,IAAI,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;gBACvE,gBAAgB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBAC9B,IAAI,MAAM,CAAC,IAAI;oBAAE,SAAS,EAAE,CAAC;gBAE7B,MAAM,aAAa,GACjB,SAAS,CAAC,IAAI,CAAC,MAAM,GAAG,EAAE;oBACxB,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,KAAK;oBACrC,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC;gBAErB,SAAS,CAAC,IAAI,CAAC;oBACb,QAAQ,CAAC,IAAI;oBACb,SAAS,CAAC,EAAE;oBACZ,aAAa;oBACb,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC;iBAC1C,CAAC,CAAC;YACL,CAAC;YAED,MAAM,QAAQ,GAAG,QAAQ,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC;gBAC7C,CAAC,CAAC,SAAS,GAAG,QAAQ,CAAC,UAAU,CAAC,MAAM;gBACxC,CAAC,CAAC,CAAC,CAAC;YACN,MAAM,SAAS,GAAG,SAAS,KAAK,QAAQ,CAAC,UAAU,CAAC,MAAM,CAAC;YAC3D,OAAO,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,SAAS,IAAI,QAAQ,CAAC,UAAU,CAAC,MAAM,SAAS,CAAC,CAAC,CAAC;YAEpG,cAAc,CAAC,IAAI,CAAC;gBAClB,OAAO,EAAE,QAAQ,CAAC,EAAE;gBACpB,SAAS,EAAE,QAAQ,CAAC,IAAI;gBACxB,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;gBACnC,aAAa,EAAE,IAAI;gBACnB,SAAS,EAAE,QAAQ;gBACnB,UAAU,EAAE,gBAAgB;aAC7B,CAAC,CAAC;QACL,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC;YAC9B,cAAc,CAAC,IAAI,CAAC;gBAClB,OAAO,EAAE,QAAQ,CAAC,EAAE;gBACpB,SAAS,EAAE,QAAQ,CAAC,IAAI;gBACxB,MAAM,EAAE,OAAO;gBACf,aAAa,EAAG,GAAa,CAAC,OAAO;gBACrC,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,EAAE;aACf,CAAC,CAAC;YAEH,SAAS,CAAC,IAAI,CAAC;gBACb,QAAQ,CAAC,IAAI;gBACb,GAAG;gBACH,GAAG,CAAC,SAAS,GAAI,GAAa,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;gBACpD,MAAM,CAAC,OAAO,CAAC;aAChB,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,sBAAsB;IACtB,MAAM,OAAO,GAAG,CAAC,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,QAAQ,CAAC,CAAC;IACxD,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,mBAAmB,SAAS,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC;IAC/D,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC;IAEvC,kBAAkB;IAClB,MAAM,MAAM,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,MAAM,CAAC;IACxE,MAAM,MAAM,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,MAAM,CAAC;IACxE,MAAM,MAAM,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IACzE,OAAO,CAAC,GAAG,CACT,KAAK,KAAK,CAAC,GAAG,MAAM,SAAS,CAAC,IAAI,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,MAAM,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,EAAE,CACrI,CAAC;IAEF,uBAAuB;IACvB,MAAM,SAAS,GAAoB;QACjC,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACnC,KAAK;QACL,UAAU,EAAE,SAAS,CAAC,UAAU;QAChC,KAAK,EAAE,cAAc;KACtB,CAAC;IAEF,MAAM,cAAc,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC;IAC1C,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,0BAA0B,QAAQ,uBAAuB,CAAC,CAAC,CAAC;AAC9E,CAAC"}
@@ -4,7 +4,10 @@ function mockClient(responses) {
4
4
  let i = 0;
5
5
  return {
6
6
  model: "test-model",
7
- generate: vi.fn(async () => responses[i++] ?? ""),
7
+ generate: vi.fn(async () => {
8
+ const text = responses[i++] ?? "";
9
+ return { text, durationMs: 100, inputTokens: null, outputTokens: null };
10
+ }),
8
11
  };
9
12
  }
10
13
  const PROMPTS = [
@@ -1 +1 @@
1
- {"version":3,"file":"activation-tester.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/activation-tester.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AAClD,OAAO,EAAE,cAAc,EAAE,MAAM,yBAAyB,CAAC;AAIzD,SAAS,UAAU,CAAC,SAAmB;IACrC,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,OAAO;QACL,KAAK,EAAE,YAAY;QACnB,QAAQ,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE,CAAC,SAAS,CAAC,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC;KAClD,CAAC;AACJ,CAAC;AAED,MAAM,OAAO,GAAuB;IAClC,EAAE,MAAM,EAAE,wBAAwB,EAAE,QAAQ,EAAE,iBAAiB,EAAE;IACjE,EAAE,MAAM,EAAE,sBAAsB,EAAE,QAAQ,EAAE,qBAAqB,EAAE;CACpE,CAAC;AAEF,QAAQ,CAAC,gBAAgB,EAAE,GAAG,EAAE;IAC9B,EAAE,CAAC,0DAA0D,EAAE,KAAK,IAAI,EAAE;QACxE,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,cAAc,EAAE,CAAC;YACjF,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,aAAa,EAAE,CAAC;SAClF,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,wBAAwB,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAChF,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClC,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC/B,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACpC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8DAA8D,EAAE,KAAK,IAAI,EAAE;QAC5E,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,QAAQ,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;YAC1E,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,SAAS,EAAE,mBAAmB,EAAE,CAAC;SACtF,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACpC,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,qDAAqD,EAAE,KAAK,IAAI,EAAE;QACnE,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,QAAQ,EAAE,CAAC;YAC5E,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,SAAS,EAAE,CAAC;SAC9E,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+BAA+B,EAAE,KAAK,IAAI,EAAE;QAC7C,MAAM,MAAM,GAAc;YACxB,KAAK,EAAE,MAAM;YACb,QAAQ,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE;gBACzB,MAAM,IAAI,KAAK,CAAC,aAAa,CAAC,CAAC;YACjC,CAAC,CAAC;SACH,CAAC;QAEF,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC9B,8EAA8E;QAC9E,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC;IAChE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;QACvD,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;YACxE,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;SACzE,CAAC,CAAC;QAEH,MAAM,OAAO,GAAU,EAAE,CAAC;QAC1B,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;QACtE,MAAM,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;QAChC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC7C,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC/C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uCAAuC,EAAE,KAAK,IAAI,EAAE;QACrD,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,yFAAyF;YACzF,iFAAiF;SAClF,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC7B,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6BAA6B,EAAE,KAAK,IAAI,EAAE;QAC3C,MAAM,MAAM,GAAG,UAAU,CAAC,EAAE,CAAC,CAAC;QAC9B,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;QACzD,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC9B,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClC,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC/B,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACtC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
1
+ {"version":3,"file":"activation-tester.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/activation-tester.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AAClD,OAAO,EAAE,cAAc,EAAE,MAAM,yBAAyB,CAAC;AAIzD,SAAS,UAAU,CAAC,SAAmB;IACrC,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,OAAO;QACL,KAAK,EAAE,YAAY;QACnB,QAAQ,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE;YACzB,MAAM,IAAI,GAAG,SAAS,CAAC,CAAC,EAAE,CAAC,IAAI,EAAE,CAAC;YAClC,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,WAAW,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;QAC1E,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,MAAM,OAAO,GAAuB;IAClC,EAAE,MAAM,EAAE,wBAAwB,EAAE,QAAQ,EAAE,iBAAiB,EAAE;IACjE,EAAE,MAAM,EAAE,sBAAsB,EAAE,QAAQ,EAAE,qBAAqB,EAAE;CACpE,CAAC;AAEF,QAAQ,CAAC,gBAAgB,EAAE,GAAG,EAAE;IAC9B,EAAE,CAAC,0DAA0D,EAAE,KAAK,IAAI,EAAE;QACxE,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,cAAc,EAAE,CAAC;YACjF,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,aAAa,EAAE,CAAC;SAClF,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,wBAAwB,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAChF,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClC,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC/B,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACpC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8DAA8D,EAAE,KAAK,IAAI,EAAE;QAC5E,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,QAAQ,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;YAC1E,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,SAAS,EAAE,mBAAmB,EAAE,CAAC;SACtF,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACpC,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,qDAAqD,EAAE,KAAK,IAAI,EAAE;QACnE,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,QAAQ,EAAE,CAAC;YAC5E,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,SAAS,EAAE,CAAC;SAC9E,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,+BAA+B,EAAE,KAAK,IAAI,EAAE;QAC7C,MAAM,MAAM,GAAc;YACxB,KAAK,EAAE,MAAM;YACb,QAAQ,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE;gBACzB,MAAM,IAAI,KAAK,CAAC,aAAa,CAAC,CAAC;YACjC,CAAC,CAAC;SACH,CAAC;QAEF,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC9B,8EAA8E;QAC9E,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC;IAChE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;QACvD,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;YACxE,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;SACzE,CAAC,CAAC;QAEH,MAAM,OAAO,GAAU,EAAE,CAAC;QAC1B,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;QACtE,MAAM,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;QAChC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC7C,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC/C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uCAAuC,EAAE,KAAK,IAAI,EAAE;QACrD,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,yFAAyF;YACzF,iFAAiF;SAClF,CAAC,CAAC;QAEH,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3B,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC7B,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6BAA6B,EAAE,KAAK,IAAI,EAAE;QAC3C,MAAM,MAAM,GAAG,UAAU,CAAC,EAAE,CAAC,CAAC;QAC9B,MAAM,OAAO,GAAG,MAAM,cAAc,CAAC,MAAM,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC;QACzD,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC9B,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClC,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC/B,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACtC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
@@ -4,7 +4,10 @@ function mockClient(responses) {
4
4
  let callIndex = 0;
5
5
  return {
6
6
  model: "test-model",
7
- generate: vi.fn(async () => responses[callIndex++] ?? ""),
7
+ generate: vi.fn(async () => {
8
+ const text = responses[callIndex++] ?? "";
9
+ return { text, durationMs: 100, inputTokens: 50, outputTokens: 100 };
10
+ }),
8
11
  };
9
12
  }
10
13
  describe("generateComparisonOutputs", () => {
@@ -1 +1 @@
1
- {"version":3,"file":"comparator.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/comparator.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AAClD,OAAO,EACL,yBAAyB,EACzB,eAAe,EACf,aAAa,GACd,MAAM,kBAAkB,CAAC;AAG1B,SAAS,UAAU,CAAC,SAAmB;IACrC,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,OAAO;QACL,KAAK,EAAE,YAAY;QACnB,QAAQ,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE,CAAC,SAAS,CAAC,SAAS,EAAE,CAAC,IAAI,EAAE,CAAC;KAC1D,CAAC;AACJ,CAAC;AAED,QAAQ,CAAC,2BAA2B,EAAE,GAAG,EAAE;IACzC,EAAE,CAAC,mDAAmD,EAAE,KAAK,IAAI,EAAE;QACjE,MAAM,MAAM,GAAG,UAAU,CAAC,CAAC,gBAAgB,EAAE,mBAAmB,CAAC,CAAC,CAAC;QACnE,MAAM,MAAM,GAAG,MAAM,yBAAyB,CAAC,aAAa,EAAE,iBAAiB,EAAE,MAAM,CAAC,CAAC;QAEzF,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;QAClD,MAAM,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;QACxD,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;QACzD,MAAM,CAAC,MAAM,CAAC,kBAAkB,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;QAC5D,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,CAAC;QAEjD,0CAA0C;QAC1C,MAAM,SAAS,GAAI,MAAM,CAAC,QAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QACzD,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;QAEhD,gCAAgC;QAChC,MAAM,UAAU,GAAI,MAAM,CAAC,QAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QAC1D,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,sBAAsB,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;IAC/B,EAAE,CAAC,sCAAsC,EAAE,KAAK,IAAI,EAAE;QACpD,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,QAAQ;gBAChB,SAAS,EAAE,aAAa;aACzB,CAAC;SACH,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,UAAU,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;QAC/E,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACvC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6BAA6B,EAAE,KAAK,IAAI,EAAE;QAC3C,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,6HAA6H;SAC9H,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;QAC5D,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAClC,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACvC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4BAA4B,EAAE,KAAK,IAAI,EAAE;QAC1C,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,EAAE;gBACrB,eAAe,EAAE,CAAC,CAAC;gBACnB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,OAAO;aAChB,CAAC;SACH,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;QAC5D,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACzC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gCAAgC,EAAE,KAAK,IAAI,EAAE;QAC9C,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,SAAS;aAClB,CAAC;SACH,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;QAC5D,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,eAAe,EAAE,GAAG,EAAE;IAC7B,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC5D,oEAAoE;QACpE,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,mBAAmB;YACnB,sBAAsB;YACtB,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,OAAO;gBACf,SAAS,EAAE,aAAa;aACzB,CAAC;SACH,CAAC,CAAC;QAEH,wCAAwC;QACxC,EAAE,CAAC,KAAK,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC,CAAC,qBAAqB;QAEpE,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,aAAa,EAAE,eAAe,EAAE,MAAM,CAAC,CAAC;QAE3E,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QAC1C,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;QACrD,MAAM,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,sBAAsB,CAAC,CAAC;QAC3D,qCAAqC;QACrC,MAAM,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACzC,MAAM,CAAC,MAAM,CAAC,mBAAmB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3C,MAAM,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5C,MAAM,CAAC,MAAM,CAAC,sBAAsB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC9C,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAEpC,EAAE,CAAC,eAAe,EAAE,CAAC;IACvB,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;QACxD,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,WAAW;YACX,cAAc;YACd,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,QAAQ;aACjB,CAAC;SACH,CAAC,CAAC;QAEH,qBAAqB;QACrB,EAAE,CAAC,KAAK,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC;QAE9C,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,GAAG,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;QACrD,6CAA6C;QAC7C,MAAM,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACzC,MAAM,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5C,8BAA8B;QAC9B,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAEpC,EAAE,CAAC,eAAe,EAAE,CAAC;IACvB,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
1
+ {"version":3,"file":"comparator.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/comparator.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AAClD,OAAO,EACL,yBAAyB,EACzB,eAAe,EACf,aAAa,GACd,MAAM,kBAAkB,CAAC;AAG1B,SAAS,UAAU,CAAC,SAAmB;IACrC,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,OAAO;QACL,KAAK,EAAE,YAAY;QACnB,QAAQ,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE;YACzB,MAAM,IAAI,GAAG,SAAS,CAAC,SAAS,EAAE,CAAC,IAAI,EAAE,CAAC;YAC1C,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,WAAW,EAAE,EAAE,EAAE,YAAY,EAAE,GAAG,EAAE,CAAC;QACvE,CAAC,CAAC;KACH,CAAC;AACJ,CAAC;AAED,QAAQ,CAAC,2BAA2B,EAAE,GAAG,EAAE;IACzC,EAAE,CAAC,mDAAmD,EAAE,KAAK,IAAI,EAAE;QACjE,MAAM,MAAM,GAAG,UAAU,CAAC,CAAC,gBAAgB,EAAE,mBAAmB,CAAC,CAAC,CAAC;QACnE,MAAM,MAAM,GAAG,MAAM,yBAAyB,CAAC,aAAa,EAAE,iBAAiB,EAAE,MAAM,CAAC,CAAC;QAEzF,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;QAClD,MAAM,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;QACxD,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;QACzD,MAAM,CAAC,MAAM,CAAC,kBAAkB,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;QAC5D,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,CAAC;QAEjD,0CAA0C;QAC1C,MAAM,SAAS,GAAI,MAAM,CAAC,QAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QACzD,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;QAEhD,gCAAgC;QAChC,MAAM,UAAU,GAAI,MAAM,CAAC,QAAgB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QAC1D,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,sBAAsB,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;IAC/B,EAAE,CAAC,sCAAsC,EAAE,KAAK,IAAI,EAAE;QACpD,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,QAAQ;gBAChB,SAAS,EAAE,aAAa;aACzB,CAAC;SACH,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,UAAU,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;QAC/E,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACvC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6BAA6B,EAAE,KAAK,IAAI,EAAE;QAC3C,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,6HAA6H;SAC9H,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;QAC5D,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAClC,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACvC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4BAA4B,EAAE,KAAK,IAAI,EAAE;QAC1C,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,EAAE;gBACrB,eAAe,EAAE,CAAC,CAAC;gBACnB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,OAAO;aAChB,CAAC;SACH,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;QAC5D,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACvC,MAAM,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACzC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gCAAgC,EAAE,KAAK,IAAI,EAAE;QAC9C,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,SAAS;aAClB,CAAC;SACH,CAAC,CAAC;QAEH,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;QAC5D,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,eAAe,EAAE,GAAG,EAAE;IAC7B,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC5D,oEAAoE;QACpE,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,mBAAmB;YACnB,sBAAsB;YACtB,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,OAAO;gBACf,SAAS,EAAE,aAAa;aACzB,CAAC;SACH,CAAC,CAAC;QAEH,wCAAwC;QACxC,EAAE,CAAC,KAAK,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC,CAAC,qBAAqB;QAEpE,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,aAAa,EAAE,eAAe,EAAE,MAAM,CAAC,CAAC;QAE3E,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QAC1C,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,mBAAmB,CAAC,CAAC;QACrD,MAAM,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC,IAAI,CAAC,sBAAsB,CAAC,CAAC;QAC3D,qCAAqC;QACrC,MAAM,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACzC,MAAM,CAAC,MAAM,CAAC,mBAAmB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC3C,MAAM,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5C,MAAM,CAAC,MAAM,CAAC,sBAAsB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC9C,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAEpC,EAAE,CAAC,eAAe,EAAE,CAAC;IACvB,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;QACxD,MAAM,MAAM,GAAG,UAAU,CAAC;YACxB,WAAW;YACX,cAAc;YACd,IAAI,CAAC,SAAS,CAAC;gBACb,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,eAAe,EAAE,CAAC;gBAClB,iBAAiB,EAAE,CAAC;gBACpB,MAAM,EAAE,QAAQ;aACjB,CAAC;SACH,CAAC,CAAC;QAEH,qBAAqB;QACrB,EAAE,CAAC,KAAK,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC;QAE9C,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,GAAG,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC;QACrD,6CAA6C;QAC7C,MAAM,CAAC,MAAM,CAAC,iBAAiB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACzC,MAAM,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAC5C,8BAA8B;QAC9B,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAEpC,EAAE,CAAC,eAAe,EAAE,CAAC;IACvB,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
@@ -3,8 +3,11 @@ import { judgeAssertion } from "../judge.js";
3
3
  // ---------------------------------------------------------------------------
4
4
  // Helpers
5
5
  // ---------------------------------------------------------------------------
6
+ function mockResult(text) {
7
+ return { text, durationMs: 100, inputTokens: null, outputTokens: null };
8
+ }
6
9
  function mockClient(response) {
7
- return { generate: vi.fn().mockResolvedValue(response), model: "test-model" };
10
+ return { generate: vi.fn().mockResolvedValue(mockResult(response)), model: "test-model" };
8
11
  }
9
12
  const ASSERTION = {
10
13
  id: "assert-1",
@@ -1 +1 @@
1
- {"version":3,"file":"judge.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/judge.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAc,MAAM,QAAQ,CAAC;AAG9D,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAE7C,8EAA8E;AAC9E,UAAU;AACV,8EAA8E;AAE9E,SAAS,UAAU,CAAC,QAAgB;IAClC,OAAO,EAAE,QAAQ,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,QAAQ,CAAC,EAAE,KAAK,EAAE,YAAY,EAAE,CAAC;AAChF,CAAC;AAED,MAAM,SAAS,GAAc;IAC3B,EAAE,EAAE,UAAU;IACd,IAAI,EAAE,6BAA6B;IACnC,IAAI,EAAE,SAAS;CAChB,CAAC;AAEF,8EAA8E;AAC9E,QAAQ;AACR,8EAA8E;AAE9E,QAAQ,CAAC,gBAAgB,EAAE,GAAG,EAAE;IAC9B,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC5D,MAAM,MAAM,GAAG,UAAU,CACvB,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,2BAA2B,EAAE,CAAC,CACvE,CAAC;QAEF,MAAM,MAAM,GAAG,MAAM,cAAc,CACjC,6CAA6C,EAC7C,SAAS,EACT,MAAM,CACP,CAAC;QAEF,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC/B,MAAM,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;QAC3D,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QACnC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,6BAA6B,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC5D,MAAM,MAAM,GAAG,UAAU,CACvB,IAAI,CAAC,SAAS,CAAC;YACb,IAAI,EAAE,KAAK;YACX,SAAS,EAAE,8BAA8B;SAC1C,CAAC,CACH,CAAC;QAEF,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,aAAa,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;QAEtE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAChC,MAAM,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,8BAA8B,CAAC,CAAC;IAChE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oCAAoC,EAAE,KAAK,IAAI,EAAE;QAClD,MAAM,MAAM,GAAG,UAAU,CAAC,kBAAkB,CAAC,CAAC;QAE9C,MAAM,MAAM,CACV,cAAc,CAAC,aAAa,EAAE,SAAS,EAAE,MAAM,CAAC,CACjD,CAAC,OAAO,CAAC,OAAO,CAAC,uBAAuB,CAAC,CAAC;IAC7C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oCAAoC,EAAE,KAAK,IAAI,EAAE;QAClD,MAAM,MAAM,GAAG,UAAU,CACvB,yDAAyD,CAC1D,CAAC;QAEF,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,aAAa,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;QACtE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
1
+ {"version":3,"file":"judge.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/judge.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAc,MAAM,QAAQ,CAAC;AAG9D,OAAO,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAE7C,8EAA8E;AAC9E,UAAU;AACV,8EAA8E;AAE9E,SAAS,UAAU,CAAC,IAAY;IAC9B,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,WAAW,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;AAC1E,CAAC;AAED,SAAS,UAAU,CAAC,QAAgB;IAClC,OAAO,EAAE,QAAQ,EAAE,EAAE,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC,EAAE,KAAK,EAAE,YAAY,EAAE,CAAC;AAC5F,CAAC;AAED,MAAM,SAAS,GAAc;IAC3B,EAAE,EAAE,UAAU;IACd,IAAI,EAAE,6BAA6B;IACnC,IAAI,EAAE,SAAS;CAChB,CAAC;AAEF,8EAA8E;AAC9E,QAAQ;AACR,8EAA8E;AAE9E,QAAQ,CAAC,gBAAgB,EAAE,GAAG,EAAE;IAC9B,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC5D,MAAM,MAAM,GAAG,UAAU,CACvB,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,SAAS,EAAE,2BAA2B,EAAE,CAAC,CACvE,CAAC;QAEF,MAAM,MAAM,GAAG,MAAM,cAAc,CACjC,6CAA6C,EAC7C,SAAS,EACT,MAAM,CACP,CAAC;QAEF,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC/B,MAAM,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;QAC3D,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QACnC,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,6BAA6B,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC5D,MAAM,MAAM,GAAG,UAAU,CACvB,IAAI,CAAC,SAAS,CAAC;YACb,IAAI,EAAE,KAAK;YACX,SAAS,EAAE,8BAA8B;SAC1C,CAAC,CACH,CAAC;QAEF,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,aAAa,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;QAEtE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAChC,MAAM,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,8BAA8B,CAAC,CAAC;IAChE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oCAAoC,EAAE,KAAK,IAAI,EAAE;QAClD,MAAM,MAAM,GAAG,UAAU,CAAC,kBAAkB,CAAC,CAAC;QAE9C,MAAM,MAAM,CACV,cAAc,CAAC,aAAa,EAAE,SAAS,EAAE,MAAM,CAAC,CACjD,CAAC,OAAO,CAAC,OAAO,CAAC,uBAAuB,CAAC,CAAC;IAC7C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oCAAoC,EAAE,KAAK,IAAI,EAAE;QAClD,MAAM,MAAM,GAAG,UAAU,CACvB,yDAAyD,CAC1D,CAAC;QAEF,MAAM,MAAM,GAAG,MAAM,cAAc,CAAC,aAAa,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;QACtE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
@@ -124,10 +124,14 @@ describe("createLlmClient", () => {
124
124
  it("returns text content on successful generate call", async () => {
125
125
  mockCreate.mockResolvedValue({
126
126
  content: [{ type: "text", text: "Generated response" }],
127
+ usage: { input_tokens: 50, output_tokens: 100 },
127
128
  });
128
129
  const client = createLlmClient();
129
130
  const result = await client.generate("system prompt", "user prompt");
130
- expect(result).toBe("Generated response");
131
+ expect(result.text).toBe("Generated response");
132
+ expect(result.durationMs).toBeGreaterThanOrEqual(0);
133
+ expect(result.inputTokens).toBe(50);
134
+ expect(result.outputTokens).toBe(100);
131
135
  expect(mockCreate).toHaveBeenCalledOnce();
132
136
  });
133
137
  it("uses default model claude-sonnet-4-20250514", async () => {
@@ -180,7 +184,7 @@ describe("createLlmClient", () => {
180
184
  mockSpawn.mockReturnValue(createFakeProc("CLI response\n"));
181
185
  const client = createLlmClient();
182
186
  const result = await client.generate("system prompt", "user prompt");
183
- expect(result).toBe("CLI response");
187
+ expect(result.text).toBe("CLI response");
184
188
  expect(mockSpawn).toHaveBeenCalledWith("claude", ["-p", "--model", "sonnet"], expect.objectContaining({ stdio: ["pipe", "pipe", "pipe"] }));
185
189
  // Verify prompt was piped via stdin
186
190
  const proc = mockSpawn.mock.results[0].value;
@@ -245,7 +249,7 @@ describe("createLlmClient", () => {
245
249
  const mockFetch = vi.spyOn(globalThis, "fetch").mockResolvedValue(new Response(JSON.stringify({ response: "Ollama reply" }), { status: 200 }));
246
250
  const client = createLlmClient();
247
251
  const result = await client.generate("system prompt", "user prompt");
248
- expect(result).toBe("Ollama reply");
252
+ expect(result.text).toBe("Ollama reply");
249
253
  expect(mockFetch).toHaveBeenCalledWith("http://localhost:11434/api/generate", expect.objectContaining({
250
254
  method: "POST",
251
255
  body: expect.stringContaining('"model":"llama3.1:8b"'),
@@ -1 +1 @@
1
- {"version":3,"file":"llm.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/llm.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,QAAQ,CAAC;AAEzE,8EAA8E;AAC9E,QAAQ;AACR,8EAA8E;AAE9E,MAAM,UAAU,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;AAC7C,MAAM,SAAS,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;AAE5C,EAAE,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,EAAE,CAAC,CAAC;IAClC,OAAO,EAAE,MAAM,aAAa;QAC1B,QAAQ,GAAG,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC;KACnC;CACF,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,oBAAoB,EAAE,GAAG,EAAE,CAAC,CAAC;IACnC,KAAK,EAAE,SAAS;CACjB,CAAC,CAAC,CAAC;AAEJ,8EAA8E;AAC9E,yDAAyD;AACzD,8EAA8E;AAC9E,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAE3C,SAAS,cAAc,CAAC,UAAkB,EAAE,QAAQ,GAAG,CAAC,EAAE,UAAU,GAAG,EAAE;IACvE,MAAM,IAAI,GAAG,IAAI,YAAY,EAAS,CAAC;IACvC,IAAI,CAAC,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;IACjC,IAAI,CAAC,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;IACjC,IAAI,CAAC,KAAK,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC;IAC9C,IAAI,CAAC,IAAI,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;IAEpB,4CAA4C;IAC5C,UAAU,CAAC,GAAG,EAAE;QACd,IAAI,UAAU;YAAE,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;QAClE,IAAI,UAAU;YAAE,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;QAClE,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;IAC/B,CAAC,EAAE,CAAC,CAAC,CAAC;IAEN,OAAO,IAAI,CAAC;AACd,CAAC;AAED,8EAA8E;AAC9E,uCAAuC;AACvC,8EAA8E;AAE9E,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,WAAW,CAAC,CAAC;AAEtD,8EAA8E;AAC9E,QAAQ;AACR,8EAA8E;AAE9E,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;IAC/B,MAAM,OAAO,GAAG,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC;IAEnC,UAAU,CAAC,GAAG,EAAE;QACd,EAAE,CAAC,aAAa,EAAE,CAAC;QACnB,OAAO,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC;QACxC,OAAO,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;QACrC,OAAO,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;QACrC,OAAO,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC;QACnC,OAAO,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC;IAChC,CAAC,CAAC,CAAC;IAEH,SAAS,CAAC,GAAG,EAAE;QACb,OAAO,CAAC,GAAG,GAAG,EAAE,GAAG,OAAO,EAAE,CAAC;IAC/B,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,iBAAiB;IACjB,4EAA4E;IAE5E,EAAE,CAAC,8CAA8C,EAAE,GAAG,EAAE;QACtD,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IAC7C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gDAAgD,EAAE,GAAG,EAAE;QACxD,OAAO,CAAC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;QAC7B,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,sDAAsD,EAAE,GAAG,EAAE;QAC9D,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACxD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,qEAAqE,EAAE,GAAG,EAAE;QAC7E,OAAO,CAAC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;QAC7B,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wDAAwD,EAAE,GAAG,EAAE;QAChE,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,QAAQ,CAAC;QAC5C,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4BAA4B,EAAE,GAAG,EAAE;QACpC,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,WAAW,CAAC;QAC/C,MAAM,CAAC,GAAG,EAAE,CAAC,eAAe,EAAE,CAAC,CAAC,OAAO,CAAC,2CAA2C,CAAC,CAAC;IACvF,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,kBAAkB;IAClB,4EAA4E;IAE5E,EAAE,CAAC,iDAAiD,EAAE,GAAG,EAAE;QACzD,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,QAAQ,CAAC;QAC5C,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,WAAW,EAAE,CAAC,CAAC;QAC1D,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACxD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8CAA8C,EAAE,GAAG,EAAE;QACtD,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,wBAAwB,EAAE,CAAC,CAAC;QAC3F,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,wBAAwB,CAAC,CAAC;IACtD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gDAAgD,EAAE,GAAG,EAAE;QACxD,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAC1E,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4CAA4C,EAAE,GAAG,EAAE;QACpD,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,aAAa,EAAE,CAAC,CAAC;QAC7E,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gEAAgE,EAAE,GAAG,EAAE;QACxE,OAAO,CAAC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;QAC7B,8CAA8C;QAC9C,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,CAAC,CAAC;QAC3D,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IAC7C,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,qBAAqB;IACrB,4EAA4E;IAE5E,QAAQ,CAAC,oBAAoB,EAAE,GAAG,EAAE;QAClC,UAAU,CAAC,GAAG,EAAE;YACd,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,WAAW,CAAC;YAC/C,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC7C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kDAAkD,EAAE,KAAK,IAAI,EAAE;YAChE,UAAU,CAAC,iBAAiB,CAAC;gBAC3B,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,oBAAoB,EAAE,CAAC;aACxD,CAAC,CAAC;YAEH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,eAAe,EAAE,aAAa,CAAC,CAAC;YAErE,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;YAC1C,MAAM,CAAC,UAAU,CAAC,CAAC,oBAAoB,EAAE,CAAC;QAC5C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,6CAA6C,EAAE,KAAK,IAAI,EAAE;YAC3D,UAAU,CAAC,iBAAiB,CAAC;gBAC3B,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;aACxC,CAAC,CAAC;YAEH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAEpC,MAAM,CAAC,UAAU,CAAC,CAAC,oBAAoB,CACrC,MAAM,CAAC,gBAAgB,CAAC,EAAE,KAAK,EAAE,0BAA0B,EAAE,CAAC,EAC9D,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;YACxD,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,wBAAwB,CAAC;YACzD,UAAU,CAAC,iBAAiB,CAAC;gBAC3B,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;aACxC,CAAC,CAAC;YAEH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAEpC,MAAM,CAAC,UAAU,CAAC,CAAC,oBAAoB,CACrC,MAAM,CAAC,gBAAgB,CAAC,EAAE,KAAK,EAAE,wBAAwB,EAAE,CAAC,EAC5D,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,mCAAmC,EAAE,KAAK,IAAI,EAAE;YACjD,UAAU,CAAC,iBAAiB,CAAC,IAAI,KAAK,CAAC,oBAAoB,CAAC,CAAC,CAAC;YAE9D,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CACzD,oBAAoB,CACrB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;YACxD,UAAU,CAAC,iBAAiB,CAAC;gBAC3B,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;aACxC,CAAC,CAAC;YAEH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,QAAQ,CAAC,kBAAkB,EAAE,gBAAgB,CAAC,CAAC;YAE5D,MAAM,CAAC,UAAU,CAAC,CAAC,oBAAoB,CACrC,MAAM,CAAC,gBAAgB,CAAC;gBACtB,MAAM,EAAE,kBAAkB;gBAC1B,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,gBAAgB,EAAE,CAAC;gBACvD,UAAU,EAAE,IAAI;aACjB,CAAC,EACF,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;YAClD,OAAO,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;YACrC,MAAM,CAAC,GAAG,EAAE,CAAC,eAAe,EAAE,CAAC,CAAC,OAAO,CAAC,8BAA8B,CAAC,CAAC;QAC1E,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,sBAAsB;IACtB,4EAA4E;IAE5E,QAAQ,CAAC,qBAAqB,EAAE,GAAG,EAAE;QACnC,UAAU,CAAC,GAAG,EAAE;YACd,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,YAAY,CAAC;QAClD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,+DAA+D,EAAE,KAAK,IAAI,EAAE;YAC7E,SAAS,CAAC,eAAe,CAAC,cAAc,CAAC,gBAAgB,CAAC,CAAC,CAAC;YAE5D,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,eAAe,EAAE,aAAa,CAAC,CAAC;YAErE,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;YACpC,MAAM,CAAC,SAAS,CAAC,CAAC,oBAAoB,CACpC,QAAQ,EACR,CAAC,IAAI,EAAE,SAAS,EAAE,QAAQ,CAAC,EAC3B,MAAM,CAAC,gBAAgB,CAAC,EAAE,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC,CAC7D,CAAC;YACF,oCAAoC;YACpC,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC;YAC7C,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,oBAAoB,CAAC,8BAA8B,CAAC,CAAC;QAC9E,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0BAA0B,EAAE,GAAG,EAAE;YAClC,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QAC7C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,4CAA4C,EAAE,KAAK,IAAI,EAAE;YAC1D,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,MAAM,CAAC;YACvC,SAAS,CAAC,eAAe,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC;YAElD,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;YACzC,MAAM,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAEpC,MAAM,CAAC,SAAS,CAAC,CAAC,oBAAoB,CACpC,QAAQ,EACR,CAAC,IAAI,EAAE,SAAS,EAAE,MAAM,CAAC,EACzB,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,gDAAgD,EAAE,KAAK,IAAI,EAAE;YAC9D,MAAM,IAAI,GAAG,IAAI,YAAY,EAAS,CAAC;YACvC,IAAI,CAAC,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;YACjC,IAAI,CAAC,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;YACjC,IAAI,CAAC,KAAK,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC;YAC9C,IAAI,CAAC,IAAI,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;YACpB,SAAS,CAAC,eAAe,CAAC,IAAI,CAAC,CAAC;YAEhC,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,OAAO,GAAG,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAE9C,oBAAoB;YACpB,UAAU,CAAC,GAAG,EAAE;gBACd,MAAM,GAAG,GAAG,IAAI,KAAK,CAAC,qBAAqB,CAA0B,CAAC;gBACtE,GAAG,CAAC,IAAI,GAAG,QAAQ,CAAC;gBACpB,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC;YAC1B,CAAC,EAAE,CAAC,CAAC,CAAC;YAEN,MAAM,MAAM,CAAC,OAAO,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,sBAAsB,CAAC,CAAC;QAChE,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;YAC5D,SAAS,CAAC,eAAe,CAAC,cAAc,CAAC,EAAE,EAAE,CAAC,EAAE,sBAAsB,CAAC,CAAC,CAAC;YAEzE,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CACzD,qDAAqD,CACtD,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,4DAA4D,EAAE,GAAG,EAAE;YACpE,OAAO,CAAC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;YAC7B,MAAM,CAAC,GAAG,EAAE,CAAC,eAAe,EAAE,CAAC,CAAC,OAAO,CACrC,6DAA6D,CAC9D,CAAC;QACJ,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,kBAAkB;IAClB,4EAA4E;IAE5E,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;QAC/B,UAAU,CAAC,GAAG,EAAE;YACd,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,QAAQ,CAAC;QAC9C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,gCAAgC,EAAE,GAAG,EAAE;YACxC,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QAC3C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;YAClD,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,aAAa,CAAC;YAC9C,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QAC3C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,4CAA4C,EAAE,KAAK,IAAI,EAAE;YAC1D,MAAM,SAAS,GAAG,EAAE,CAAC,KAAK,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC,iBAAiB,CAC/D,IAAI,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,cAAc,EAAE,CAAC,EAAE,EAAE,MAAM,EAAE,GAAG,EAAE,CAAC,CAC5E,CAAC;YAEF,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,eAAe,EAAE,aAAa,CAAC,CAAC;YAErE,MAAM,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;YACpC,MAAM,CAAC,SAAS,CAAC,CAAC,oBAAoB,CACpC,qCAAqC,EACrC,MAAM,CAAC,gBAAgB,CAAC;gBACtB,MAAM,EAAE,MAAM;gBACd,IAAI,EAAE,MAAM,CAAC,gBAAgB,CAAC,uBAAuB,CAAC;aACvD,CAAC,CACH,CAAC;YAEF,SAAS,CAAC,WAAW,EAAE,CAAC;QAC1B,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,2CAA2C,EAAE,KAAK,IAAI,EAAE;YACzD,OAAO,CAAC,GAAG,CAAC,eAAe,GAAG,yBAAyB,CAAC;YAExD,MAAM,SAAS,GAAG,EAAE,CAAC,KAAK,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC,iBAAiB,CAC/D,IAAI,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,EAAE,EAAE,MAAM,EAAE,GAAG,EAAE,CAAC,CAClE,CAAC;YAEF,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAEpC,MAAM,CAAC,SAAS,CAAC,CAAC,oBAAoB,CACpC,sCAAsC,EACtC,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;YAEF,SAAS,CAAC,WAAW,EAAE,CAAC;QAC1B,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
1
+ {"version":3,"file":"llm.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/llm.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,QAAQ,CAAC;AAEzE,8EAA8E;AAC9E,QAAQ;AACR,8EAA8E;AAE9E,MAAM,UAAU,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;AAC7C,MAAM,SAAS,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;AAE5C,EAAE,CAAC,IAAI,CAAC,mBAAmB,EAAE,GAAG,EAAE,CAAC,CAAC;IAClC,OAAO,EAAE,MAAM,aAAa;QAC1B,QAAQ,GAAG,EAAE,MAAM,EAAE,UAAU,EAAE,CAAC;KACnC;CACF,CAAC,CAAC,CAAC;AAEJ,EAAE,CAAC,IAAI,CAAC,oBAAoB,EAAE,GAAG,EAAE,CAAC,CAAC;IACnC,KAAK,EAAE,SAAS;CACjB,CAAC,CAAC,CAAC;AAEJ,8EAA8E;AAC9E,yDAAyD;AACzD,8EAA8E;AAC9E,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAE3C,SAAS,cAAc,CAAC,UAAkB,EAAE,QAAQ,GAAG,CAAC,EAAE,UAAU,GAAG,EAAE;IACvE,MAAM,IAAI,GAAG,IAAI,YAAY,EAAS,CAAC;IACvC,IAAI,CAAC,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;IACjC,IAAI,CAAC,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;IACjC,IAAI,CAAC,KAAK,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC;IAC9C,IAAI,CAAC,IAAI,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;IAEpB,4CAA4C;IAC5C,UAAU,CAAC,GAAG,EAAE;QACd,IAAI,UAAU;YAAE,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;QAClE,IAAI,UAAU;YAAE,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;QAClE,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;IAC/B,CAAC,EAAE,CAAC,CAAC,CAAC;IAEN,OAAO,IAAI,CAAC;AACd,CAAC;AAED,8EAA8E;AAC9E,uCAAuC;AACvC,8EAA8E;AAE9E,MAAM,EAAE,eAAe,EAAE,GAAG,MAAM,MAAM,CAAC,WAAW,CAAC,CAAC;AAEtD,8EAA8E;AAC9E,QAAQ;AACR,8EAA8E;AAE9E,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;IAC/B,MAAM,OAAO,GAAG,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC;IAEnC,UAAU,CAAC,GAAG,EAAE;QACd,EAAE,CAAC,aAAa,EAAE,CAAC;QACnB,OAAO,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC;QACxC,OAAO,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;QACrC,OAAO,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;QACrC,OAAO,OAAO,CAAC,GAAG,CAAC,eAAe,CAAC;QACnC,OAAO,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC;IAChC,CAAC,CAAC,CAAC;IAEH,SAAS,CAAC,GAAG,EAAE;QACb,OAAO,CAAC,GAAG,GAAG,EAAE,GAAG,OAAO,EAAE,CAAC;IAC/B,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,iBAAiB;IACjB,4EAA4E;IAE5E,EAAE,CAAC,8CAA8C,EAAE,GAAG,EAAE;QACtD,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IAC7C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gDAAgD,EAAE,GAAG,EAAE;QACxD,OAAO,CAAC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;QAC7B,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,sDAAsD,EAAE,GAAG,EAAE;QAC9D,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACxD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,qEAAqE,EAAE,GAAG,EAAE;QAC7E,OAAO,CAAC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;QAC7B,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wDAAwD,EAAE,GAAG,EAAE;QAChE,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,QAAQ,CAAC;QAC5C,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;QACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4BAA4B,EAAE,GAAG,EAAE;QACpC,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,WAAW,CAAC;QAC/C,MAAM,CAAC,GAAG,EAAE,CAAC,eAAe,EAAE,CAAC,CAAC,OAAO,CAAC,2CAA2C,CAAC,CAAC;IACvF,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,kBAAkB;IAClB,4EAA4E;IAE5E,EAAE,CAAC,iDAAiD,EAAE,GAAG,EAAE;QACzD,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,QAAQ,CAAC;QAC5C,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,WAAW,EAAE,CAAC,CAAC;QAC1D,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACxD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8CAA8C,EAAE,GAAG,EAAE;QACtD,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC3C,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,wBAAwB,EAAE,CAAC,CAAC;QAC3F,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,wBAAwB,CAAC,CAAC;IACtD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gDAAgD,EAAE,GAAG,EAAE;QACxD,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAC1E,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4CAA4C,EAAE,GAAG,EAAE;QACpD,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,aAAa,EAAE,CAAC,CAAC;QAC7E,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;IAC3C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gEAAgE,EAAE,GAAG,EAAE;QACxE,OAAO,CAAC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;QAC7B,8CAA8C;QAC9C,MAAM,MAAM,GAAG,eAAe,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,CAAC,CAAC;QAC3D,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IAC7C,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,qBAAqB;IACrB,4EAA4E;IAE5E,QAAQ,CAAC,oBAAoB,EAAE,GAAG,EAAE;QAClC,UAAU,CAAC,GAAG,EAAE;YACd,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,WAAW,CAAC;YAC/C,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,UAAU,CAAC;QAC7C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kDAAkD,EAAE,KAAK,IAAI,EAAE;YAChE,UAAU,CAAC,iBAAiB,CAAC;gBAC3B,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,oBAAoB,EAAE,CAAC;gBACvD,KAAK,EAAE,EAAE,YAAY,EAAE,EAAE,EAAE,aAAa,EAAE,GAAG,EAAE;aAChD,CAAC,CAAC;YAEH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,eAAe,EAAE,aAAa,CAAC,CAAC;YAErE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;YAC/C,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,sBAAsB,CAAC,CAAC,CAAC,CAAC;YACpD,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACpC,MAAM,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACtC,MAAM,CAAC,UAAU,CAAC,CAAC,oBAAoB,EAAE,CAAC;QAC5C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,6CAA6C,EAAE,KAAK,IAAI,EAAE;YAC3D,UAAU,CAAC,iBAAiB,CAAC;gBAC3B,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;aACxC,CAAC,CAAC;YAEH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAEpC,MAAM,CAAC,UAAU,CAAC,CAAC,oBAAoB,CACrC,MAAM,CAAC,gBAAgB,CAAC,EAAE,KAAK,EAAE,0BAA0B,EAAE,CAAC,EAC9D,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;YACxD,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,wBAAwB,CAAC;YACzD,UAAU,CAAC,iBAAiB,CAAC;gBAC3B,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;aACxC,CAAC,CAAC;YAEH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAEpC,MAAM,CAAC,UAAU,CAAC,CAAC,oBAAoB,CACrC,MAAM,CAAC,gBAAgB,CAAC,EAAE,KAAK,EAAE,wBAAwB,EAAE,CAAC,EAC5D,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,mCAAmC,EAAE,KAAK,IAAI,EAAE;YACjD,UAAU,CAAC,iBAAiB,CAAC,IAAI,KAAK,CAAC,oBAAoB,CAAC,CAAC,CAAC;YAE9D,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CACzD,oBAAoB,CACrB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;YACxD,UAAU,CAAC,iBAAiB,CAAC;gBAC3B,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;aACxC,CAAC,CAAC;YAEH,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,QAAQ,CAAC,kBAAkB,EAAE,gBAAgB,CAAC,CAAC;YAE5D,MAAM,CAAC,UAAU,CAAC,CAAC,oBAAoB,CACrC,MAAM,CAAC,gBAAgB,CAAC;gBACtB,MAAM,EAAE,kBAAkB;gBAC1B,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,gBAAgB,EAAE,CAAC;gBACvD,UAAU,EAAE,IAAI;aACjB,CAAC,EACF,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;YAClD,OAAO,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;YACrC,MAAM,CAAC,GAAG,EAAE,CAAC,eAAe,EAAE,CAAC,CAAC,OAAO,CAAC,8BAA8B,CAAC,CAAC;QAC1E,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,sBAAsB;IACtB,4EAA4E;IAE5E,QAAQ,CAAC,qBAAqB,EAAE,GAAG,EAAE;QACnC,UAAU,CAAC,GAAG,EAAE;YACd,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,YAAY,CAAC;QAClD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,+DAA+D,EAAE,KAAK,IAAI,EAAE;YAC7E,SAAS,CAAC,eAAe,CAAC,cAAc,CAAC,gBAAgB,CAAC,CAAC,CAAC;YAE5D,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,eAAe,EAAE,aAAa,CAAC,CAAC;YAErE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;YACzC,MAAM,CAAC,SAAS,CAAC,CAAC,oBAAoB,CACpC,QAAQ,EACR,CAAC,IAAI,EAAE,SAAS,EAAE,QAAQ,CAAC,EAC3B,MAAM,CAAC,gBAAgB,CAAC,EAAE,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC,CAC7D,CAAC;YACF,oCAAoC;YACpC,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC;YAC7C,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,oBAAoB,CAAC,8BAA8B,CAAC,CAAC;QAC9E,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0BAA0B,EAAE,GAAG,EAAE;YAClC,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;QAC7C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,4CAA4C,EAAE,KAAK,IAAI,EAAE;YAC1D,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,MAAM,CAAC;YACvC,SAAS,CAAC,eAAe,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,CAAC;YAElD,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;YACzC,MAAM,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAEpC,MAAM,CAAC,SAAS,CAAC,CAAC,oBAAoB,CACpC,QAAQ,EACR,CAAC,IAAI,EAAE,SAAS,EAAE,MAAM,CAAC,EACzB,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,gDAAgD,EAAE,KAAK,IAAI,EAAE;YAC9D,MAAM,IAAI,GAAG,IAAI,YAAY,EAAS,CAAC;YACvC,IAAI,CAAC,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;YACjC,IAAI,CAAC,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;YACjC,IAAI,CAAC,KAAK,GAAG,EAAE,GAAG,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC;YAC9C,IAAI,CAAC,IAAI,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;YACpB,SAAS,CAAC,eAAe,CAAC,IAAI,CAAC,CAAC;YAEhC,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,OAAO,GAAG,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAE9C,oBAAoB;YACpB,UAAU,CAAC,GAAG,EAAE;gBACd,MAAM,GAAG,GAAG,IAAI,KAAK,CAAC,qBAAqB,CAA0B,CAAC;gBACtE,GAAG,CAAC,IAAI,GAAG,QAAQ,CAAC;gBACpB,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC;YAC1B,CAAC,EAAE,CAAC,CAAC,CAAC;YAEN,MAAM,MAAM,CAAC,OAAO,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,sBAAsB,CAAC,CAAC;QAChE,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;YAC5D,SAAS,CAAC,eAAe,CAAC,cAAc,CAAC,EAAE,EAAE,CAAC,EAAE,sBAAsB,CAAC,CAAC,CAAC;YAEzE,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CACzD,qDAAqD,CACtD,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,4DAA4D,EAAE,GAAG,EAAE;YACpE,OAAO,CAAC,GAAG,CAAC,UAAU,GAAG,GAAG,CAAC;YAC7B,MAAM,CAAC,GAAG,EAAE,CAAC,eAAe,EAAE,CAAC,CAAC,OAAO,CACrC,6DAA6D,CAC9D,CAAC;QACJ,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,4EAA4E;IAC5E,kBAAkB;IAClB,4EAA4E;IAE5E,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;QAC/B,UAAU,CAAC,GAAG,EAAE;YACd,OAAO,CAAC,GAAG,CAAC,oBAAoB,GAAG,QAAQ,CAAC;QAC9C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,gCAAgC,EAAE,GAAG,EAAE;YACxC,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QAC3C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;YAClD,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,aAAa,CAAC;YAC9C,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QAC3C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,4CAA4C,EAAE,KAAK,IAAI,EAAE;YAC1D,MAAM,SAAS,GAAG,EAAE,CAAC,KAAK,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC,iBAAiB,CAC/D,IAAI,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,cAAc,EAAE,CAAC,EAAE,EAAE,MAAM,EAAE,GAAG,EAAE,CAAC,CAC5E,CAAC;YAEF,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,eAAe,EAAE,aAAa,CAAC,CAAC;YAErE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;YACzC,MAAM,CAAC,SAAS,CAAC,CAAC,oBAAoB,CACpC,qCAAqC,EACrC,MAAM,CAAC,gBAAgB,CAAC;gBACtB,MAAM,EAAE,MAAM;gBACd,IAAI,EAAE,MAAM,CAAC,gBAAgB,CAAC,uBAAuB,CAAC;aACvD,CAAC,CACH,CAAC;YAEF,SAAS,CAAC,WAAW,EAAE,CAAC;QAC1B,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,2CAA2C,EAAE,KAAK,IAAI,EAAE;YACzD,OAAO,CAAC,GAAG,CAAC,eAAe,GAAG,yBAAyB,CAAC;YAExD,MAAM,SAAS,GAAG,EAAE,CAAC,KAAK,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC,iBAAiB,CAC/D,IAAI,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,EAAE,EAAE,MAAM,EAAE,GAAG,EAAE,CAAC,CAClE,CAAC;YAEF,MAAM,MAAM,GAAG,eAAe,EAAE,CAAC;YACjC,MAAM,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;YAEpC,MAAM,CAAC,SAAS,CAAC,CAAC,oBAAoB,CACpC,sCAAsC,EACtC,MAAM,CAAC,QAAQ,EAAE,CAClB,CAAC;YAEF,SAAS,CAAC,WAAW,EAAE,CAAC;QAC1B,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
@@ -25,7 +25,7 @@ ${p.prompt}
25
25
 
26
26
  Would this user prompt trigger this skill?`;
27
27
  try {
28
- const response = await client.generate(ACTIVATION_SYSTEM_PROMPT, userPrompt);
28
+ const { text: response } = await client.generate(ACTIVATION_SYSTEM_PROMPT, userPrompt);
29
29
  const jsonMatch = response.match(/```(?:json)?\s*([\s\S]*?)```/) || [null, response];
30
30
  const json = JSON.parse(jsonMatch[1].trim());
31
31
  const activate = !!json.activate;
@@ -1 +1 @@
1
- {"version":3,"file":"activation-tester.js","sourceRoot":"","sources":["../../src/eval/activation-tester.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,4EAA4E;AAC5E,8EAA8E;AA8B9E,MAAM,wBAAwB,GAAG;;;;;;;;;;;;EAY/B,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,gBAAwB,EACxB,OAA2B,EAC3B,MAAiB,EACjB,QAA6C;IAE7C,MAAM,OAAO,GAAuB,EAAE,CAAC;IAEvC,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,UAAU,GAAG;EACrB,gBAAgB;;;EAGhB,CAAC,CAAC,MAAM;;2CAEiC,CAAC;QAExC,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,wBAAwB,EAAE,UAAU,CAAC,CAAC;YAC7E,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,8BAA8B,CAAC,IAAI,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;YACrF,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAE,CAAC,IAAI,EAAE,CAAC,CAAC;YAE9C,MAAM,QAAQ,GAAG,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC;YACjC,MAAM,UAAU,GAAG,CAAC,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,UAAU,CAAC;gBACpE,CAAC,CAAE,IAAI,CAAC,UAAwC;gBAChD,CAAC,CAAC,KAAK,CAAC;YAEV,MAAM,cAAc,GAAG,cAAc,CAAC,CAAC,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;YAE5D,MAAM,MAAM,GAAqB;gBAC/B,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,QAAQ,EAAE,CAAC,CAAC,QAAQ;gBACpB,QAAQ;gBACR,UAAU;gBACV,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC;gBACvC,cAAc;aACf,CAAC;YACF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACrB,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC;QACrB,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAqB;gBAC/B,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,QAAQ,EAAE,CAAC,CAAC,QAAQ;gBACpB,QAAQ,EAAE,KAAK;gBACf,UAAU,EAAE,KAAK;gBACjB,SAAS,EAAE,UAAU,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE;gBACvE,cAAc,EAAE,CAAC,CAAC,QAAQ,KAAK,iBAAiB,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI;aAC/D,CAAC;YACF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACrB,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC;QACrB,CAAC;IACH,CAAC;IAED,OAAO,cAAc,CAAC,OAAO,CAAC,CAAC;AACjC,CAAC;AAED,SAAS,cAAc,CACrB,QAAmD,EACnD,MAAe;IAEf,IAAI,QAAQ,KAAK,iBAAiB,IAAI,MAAM;QAAE,OAAO,IAAI,CAAC;IAC1D,IAAI,QAAQ,KAAK,iBAAiB,IAAI,CAAC,MAAM;QAAE,OAAO,IAAI,CAAC;IAC3D,IAAI,QAAQ,KAAK,qBAAqB,IAAI,CAAC,MAAM;QAAE,OAAO,IAAI,CAAC;IAC/D,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,cAAc,CAAC,OAA2B;IACjD,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACnE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACnE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACnE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACnE,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC;IAE7B,OAAO;QACL,OAAO;QACP,SAAS,EAAE,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QAC3C,MAAM,EAAE,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QACxC,WAAW,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QAC9C,KAAK;QACL,EAAE;QACF,EAAE;QACF,EAAE;QACF,EAAE;KACH,CAAC;AACJ,CAAC"}
1
+ {"version":3,"file":"activation-tester.js","sourceRoot":"","sources":["../../src/eval/activation-tester.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,4EAA4E;AAC5E,8EAA8E;AA8B9E,MAAM,wBAAwB,GAAG;;;;;;;;;;;;EAY/B,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,gBAAwB,EACxB,OAA2B,EAC3B,MAAiB,EACjB,QAA6C;IAE7C,MAAM,OAAO,GAAuB,EAAE,CAAC;IAEvC,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,UAAU,GAAG;EACrB,gBAAgB;;;EAGhB,CAAC,CAAC,MAAM;;2CAEiC,CAAC;QAExC,IAAI,CAAC;YACH,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,wBAAwB,EAAE,UAAU,CAAC,CAAC;YACvF,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,8BAA8B,CAAC,IAAI,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;YACrF,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAE,CAAC,IAAI,EAAE,CAAC,CAAC;YAE9C,MAAM,QAAQ,GAAG,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC;YACjC,MAAM,UAAU,GAAG,CAAC,MAAM,EAAE,QAAQ,EAAE,KAAK,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,UAAU,CAAC;gBACpE,CAAC,CAAE,IAAI,CAAC,UAAwC;gBAChD,CAAC,CAAC,KAAK,CAAC;YAEV,MAAM,cAAc,GAAG,cAAc,CAAC,CAAC,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;YAE5D,MAAM,MAAM,GAAqB;gBAC/B,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,QAAQ,EAAE,CAAC,CAAC,QAAQ;gBACpB,QAAQ;gBACR,UAAU;gBACV,SAAS,EAAE,MAAM,CAAC,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC;gBACvC,cAAc;aACf,CAAC;YACF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACrB,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC;QACrB,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,MAAM,GAAqB;gBAC/B,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,QAAQ,EAAE,CAAC,CAAC,QAAQ;gBACpB,QAAQ,EAAE,KAAK;gBACf,UAAU,EAAE,KAAK;gBACjB,SAAS,EAAE,UAAU,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE;gBACvE,cAAc,EAAE,CAAC,CAAC,QAAQ,KAAK,iBAAiB,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI;aAC/D,CAAC;YACF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACrB,QAAQ,EAAE,CAAC,MAAM,CAAC,CAAC;QACrB,CAAC;IACH,CAAC;IAED,OAAO,cAAc,CAAC,OAAO,CAAC,CAAC;AACjC,CAAC;AAED,SAAS,cAAc,CACrB,QAAmD,EACnD,MAAe;IAEf,IAAI,QAAQ,KAAK,iBAAiB,IAAI,MAAM;QAAE,OAAO,IAAI,CAAC;IAC1D,IAAI,QAAQ,KAAK,iBAAiB,IAAI,CAAC,MAAM;QAAE,OAAO,IAAI,CAAC;IAC3D,IAAI,QAAQ,KAAK,qBAAqB,IAAI,CAAC,MAAM;QAAE,OAAO,IAAI,CAAC;IAC/D,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,cAAc,CAAC,OAA2B;IACjD,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACnE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACnE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACnE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;IACnE,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC;IAE7B,OAAO;QACL,OAAO;QACP,SAAS,EAAE,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QAC3C,MAAM,EAAE,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;QACxC,WAAW,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QAC9C,KAAK;QACL,EAAE;QACF,EAAE;QACF,EAAE;QACF,EAAE;KACH,CAAC;AACJ,CAAC"}
@@ -10,6 +10,8 @@ export interface BenchmarkCase {
10
10
  status: "pass" | "fail" | "error";
11
11
  error_message: string | null;
12
12
  pass_rate: number;
13
+ durationMs?: number;
14
+ tokens?: number | null;
13
15
  assertions: BenchmarkAssertionResult[];
14
16
  }
15
17
  export interface BenchmarkResult {
@@ -1 +1 @@
1
- {"version":3,"file":"benchmark.js","sourceRoot":"","sources":["../../src/eval/benchmark.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,4BAA4B;AAC5B,8EAA8E;AAE9E,OAAO,EAAE,aAAa,EAAE,YAAY,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AA0BjC,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,QAAgB,EAChB,MAAuB;IAEvB,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IACzC,SAAS,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACzC,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,gBAAgB,CAAC,CAAC;IAClD,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;AACpE,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,QAAgB;IAEhB,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,gBAAgB,CAAC,CAAC;IAC3D,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;QAAE,OAAO,IAAI,CAAC;IAEvC,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;QAC5C,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAoB,CAAC;IAC5C,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC"}
1
+ {"version":3,"file":"benchmark.js","sourceRoot":"","sources":["../../src/eval/benchmark.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,4BAA4B;AAC5B,8EAA8E;AAE9E,OAAO,EAAE,aAAa,EAAE,YAAY,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AA4BjC,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,QAAgB,EAChB,MAAuB;IAEvB,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IACzC,SAAS,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACzC,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,gBAAgB,CAAC,CAAC;IAClD,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;AACpE,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,QAAgB;IAEhB,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,gBAAgB,CAAC,CAAC;IAC3D,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;QAAE,OAAO,IAAI,CAAC;IAEvC,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;QAC5C,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAoB,CAAC;IAC5C,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC"}
@@ -2,8 +2,10 @@ import type { LlmClient } from "./llm.js";
2
2
  export interface ComparisonOutput {
3
3
  skillOutput: string;
4
4
  skillDurationMs: number;
5
+ skillTokens: number | null;
5
6
  baselineOutput: string;
6
7
  baselineDurationMs: number;
8
+ baselineTokens: number | null;
7
9
  }
8
10
  export interface ComparisonScore {
9
11
  contentScoreA: number;
@@ -16,8 +18,10 @@ export interface ComparisonResult {
16
18
  prompt: string;
17
19
  skillOutput: string;
18
20
  skillDurationMs: number;
21
+ skillTokens: number | null;
19
22
  baselineOutput: string;
20
23
  baselineDurationMs: number;
24
+ baselineTokens: number | null;
21
25
  skillContentScore: number;
22
26
  skillStructureScore: number;
23
27
  baselineContentScore: number;
@@ -22,13 +22,17 @@ export async function generateComparisonOutputs(prompt, skillContent, client) {
22
22
  const skillSystemPrompt = `You are an AI assistant enhanced with the following skill:\n\n${skillContent}`;
23
23
  const baselineSystemPrompt = "You are a helpful AI assistant.";
24
24
  // Run sequentially (claude-cli can only handle one at a time)
25
- const skillStart = Date.now();
26
- const skillOutput = await client.generate(skillSystemPrompt, prompt);
27
- const skillDurationMs = Date.now() - skillStart;
28
- const baselineStart = Date.now();
29
- const baselineOutput = await client.generate(baselineSystemPrompt, prompt);
30
- const baselineDurationMs = Date.now() - baselineStart;
31
- return { skillOutput, skillDurationMs, baselineOutput, baselineDurationMs };
25
+ const skillResult = await client.generate(skillSystemPrompt, prompt);
26
+ const baselineResult = await client.generate(baselineSystemPrompt, prompt);
27
+ const totalTokens = (r) => r.inputTokens != null && r.outputTokens != null ? r.inputTokens + r.outputTokens : null;
28
+ return {
29
+ skillOutput: skillResult.text,
30
+ skillDurationMs: skillResult.durationMs,
31
+ skillTokens: totalTokens(skillResult),
32
+ baselineOutput: baselineResult.text,
33
+ baselineDurationMs: baselineResult.durationMs,
34
+ baselineTokens: totalTokens(baselineResult),
35
+ };
32
36
  }
33
37
  export async function scoreComparison(outputA, outputB, prompt, client) {
34
38
  const userPrompt = `## Original Prompt
@@ -41,7 +45,7 @@ ${outputA}
41
45
  ${outputB}
42
46
 
43
47
  Evaluate both responses.`;
44
- const response = await client.generate(COMPARATOR_SYSTEM_PROMPT, userPrompt);
48
+ const { text: response } = await client.generate(COMPARATOR_SYSTEM_PROMPT, userPrompt);
45
49
  // Parse JSON from response (may be in code fence)
46
50
  const jsonMatch = response.match(/```(?:json)?\s*([\s\S]*?)```/) || [null, response];
47
51
  const json = JSON.parse(jsonMatch[1].trim());
@@ -82,8 +86,10 @@ export async function runComparison(prompt, skillContent, client) {
82
86
  prompt,
83
87
  skillOutput: outputs.skillOutput,
84
88
  skillDurationMs: outputs.skillDurationMs,
89
+ skillTokens: outputs.skillTokens,
85
90
  baselineOutput: outputs.baselineOutput,
86
91
  baselineDurationMs: outputs.baselineDurationMs,
92
+ baselineTokens: outputs.baselineTokens,
87
93
  skillContentScore,
88
94
  skillStructureScore,
89
95
  baselineContentScore,
@@ -1 +1 @@
1
- {"version":3,"file":"comparator.js","sourceRoot":"","sources":["../../src/eval/comparator.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,6DAA6D;AAC7D,8EAA8E;AAgC9E,MAAM,wBAAwB,GAAG;;;;;;;;;;;;;;;;EAgB/B,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC7C,MAAc,EACd,YAAoB,EACpB,MAAiB;IAEjB,MAAM,iBAAiB,GAAG,iEAAiE,YAAY,EAAE,CAAC;IAC1G,MAAM,oBAAoB,GAAG,iCAAiC,CAAC;IAE/D,8DAA8D;IAC9D,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAC9B,MAAM,WAAW,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,iBAAiB,EAAE,MAAM,CAAC,CAAC;IACrE,MAAM,eAAe,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,UAAU,CAAC;IAEhD,MAAM,aAAa,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACjC,MAAM,cAAc,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,oBAAoB,EAAE,MAAM,CAAC,CAAC;IAC3E,MAAM,kBAAkB,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,aAAa,CAAC;IAEtD,OAAO,EAAE,WAAW,EAAE,eAAe,EAAE,cAAc,EAAE,kBAAkB,EAAE,CAAC;AAC9E,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,eAAe,CACnC,OAAe,EACf,OAAe,EACf,MAAc,EACd,MAAiB;IAEjB,MAAM,UAAU,GAAG;EACnB,MAAM;;;EAGN,OAAO;;;EAGP,OAAO;;yBAEgB,CAAC;IAExB,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,wBAAwB,EAAE,UAAU,CAAC,CAAC;IAE7E,kDAAkD;IAClD,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,8BAA8B,CAAC,IAAI,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;IACrF,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAE,CAAC,IAAI,EAAE,CAAC,CAAC;IAE9C,OAAO;QACL,aAAa,EAAE,UAAU,CAAC,IAAI,CAAC,eAAe,CAAC;QAC/C,eAAe,EAAE,UAAU,CAAC,IAAI,CAAC,iBAAiB,CAAC;QACnD,aAAa,EAAE,UAAU,CAAC,IAAI,CAAC,eAAe,CAAC;QAC/C,eAAe,EAAE,UAAU,CAAC,IAAI,CAAC,iBAAiB,CAAC;QACnD,MAAM,EAAE,IAAI,CAAC,MAAM,KAAK,OAAO,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,IAAI,IAAI,CAAC,MAAM,KAAK,KAAK;YAClF,CAAC,CAAC,IAAI,CAAC,MAAM;YACb,CAAC,CAAC,KAAK;KACV,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,MAAc,EACd,YAAoB,EACpB,MAAiB;IAEjB,MAAM,OAAO,GAAG,MAAM,yBAAyB,CAAC,MAAM,EAAE,YAAY,EAAE,MAAM,CAAC,CAAC;IAE9E,uCAAuC;IACvC,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,CAAC;IACrC,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,OAAO,CAAC,cAAc,CAAC;IACxE,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC;IAExE,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;IAEvE,oCAAoC;IACpC,MAAM,iBAAiB,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC;IACjF,MAAM,mBAAmB,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,CAAC,MAAM,CAAC,eAAe,CAAC;IACvF,MAAM,oBAAoB,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC;IACpF,MAAM,sBAAsB,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,CAAC,MAAM,CAAC,eAAe,CAAC;IAE1F,kBAAkB;IAClB,IAAI,MAAoC,CAAC;IACzC,IAAI,MAAM,CAAC,MAAM,KAAK,KAAK,EAAE,CAAC;QAC5B,MAAM,GAAG,KAAK,CAAC;IACjB,CAAC;SAAM,IAAI,MAAM,CAAC,MAAM,KAAK,OAAO,EAAE,CAAC;QACrC,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC;IAC3C,CAAC;SAAM,CAAC;QACN,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO,CAAC;IAC3C,CAAC;IAED,OAAO;QACL,MAAM;QACN,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,eAAe,EAAE,OAAO,CAAC,eAAe;QACxC,cAAc,EAAE,OAAO,CAAC,cAAc;QACtC,kBAAkB,EAAE,OAAO,CAAC,kBAAkB;QAC9C,iBAAiB;QACjB,mBAAmB;QACnB,oBAAoB;QACpB,sBAAsB;QACtB,MAAM;KACP,CAAC;AACJ,CAAC;AAED,SAAS,UAAU,CAAC,GAAY;IAC9B,MAAM,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;IACtB,IAAI,KAAK,CAAC,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IACvB,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACjD,CAAC"}
1
+ {"version":3,"file":"comparator.js","sourceRoot":"","sources":["../../src/eval/comparator.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,6DAA6D;AAC7D,8EAA8E;AAoC9E,MAAM,wBAAwB,GAAG;;;;;;;;;;;;;;;;EAgB/B,CAAC;AAEH,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC7C,MAAc,EACd,YAAoB,EACpB,MAAiB;IAEjB,MAAM,iBAAiB,GAAG,iEAAiE,YAAY,EAAE,CAAC;IAC1G,MAAM,oBAAoB,GAAG,iCAAiC,CAAC;IAE/D,8DAA8D;IAC9D,MAAM,WAAW,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,iBAAiB,EAAE,MAAM,CAAC,CAAC;IACrE,MAAM,cAAc,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,oBAAoB,EAAE,MAAM,CAAC,CAAC;IAE3E,MAAM,WAAW,GAAG,CAAC,CAAqB,EAAE,EAAE,CAC5C,CAAC,CAAC,WAAW,IAAI,IAAI,IAAI,CAAC,CAAC,YAAY,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC;IAE1F,OAAO;QACL,WAAW,EAAE,WAAW,CAAC,IAAI;QAC7B,eAAe,EAAE,WAAW,CAAC,UAAU;QACvC,WAAW,EAAE,WAAW,CAAC,WAAW,CAAC;QACrC,cAAc,EAAE,cAAc,CAAC,IAAI;QACnC,kBAAkB,EAAE,cAAc,CAAC,UAAU;QAC7C,cAAc,EAAE,WAAW,CAAC,cAAc,CAAC;KAC5C,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,eAAe,CACnC,OAAe,EACf,OAAe,EACf,MAAc,EACd,MAAiB;IAEjB,MAAM,UAAU,GAAG;EACnB,MAAM;;;EAGN,OAAO;;;EAGP,OAAO;;yBAEgB,CAAC;IAExB,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,wBAAwB,EAAE,UAAU,CAAC,CAAC;IAEvF,kDAAkD;IAClD,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,8BAA8B,CAAC,IAAI,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;IACrF,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAE,CAAC,IAAI,EAAE,CAAC,CAAC;IAE9C,OAAO;QACL,aAAa,EAAE,UAAU,CAAC,IAAI,CAAC,eAAe,CAAC;QAC/C,eAAe,EAAE,UAAU,CAAC,IAAI,CAAC,iBAAiB,CAAC;QACnD,aAAa,EAAE,UAAU,CAAC,IAAI,CAAC,eAAe,CAAC;QAC/C,eAAe,EAAE,UAAU,CAAC,IAAI,CAAC,iBAAiB,CAAC;QACnD,MAAM,EAAE,IAAI,CAAC,MAAM,KAAK,OAAO,IAAI,IAAI,CAAC,MAAM,KAAK,QAAQ,IAAI,IAAI,CAAC,MAAM,KAAK,KAAK;YAClF,CAAC,CAAC,IAAI,CAAC,MAAM;YACb,CAAC,CAAC,KAAK;KACV,CAAC;AACJ,CAAC;AAED,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,MAAc,EACd,YAAoB,EACpB,MAAiB;IAEjB,MAAM,OAAO,GAAG,MAAM,yBAAyB,CAAC,MAAM,EAAE,YAAY,EAAE,MAAM,CAAC,CAAC;IAE9E,uCAAuC;IACvC,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,CAAC;IACrC,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,OAAO,CAAC,cAAc,CAAC;IACxE,MAAM,OAAO,GAAG,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC;IAExE,MAAM,MAAM,GAAG,MAAM,eAAe,CAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;IAEvE,oCAAoC;IACpC,MAAM,iBAAiB,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC;IACjF,MAAM,mBAAmB,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,CAAC,MAAM,CAAC,eAAe,CAAC;IACvF,MAAM,oBAAoB,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,aAAa,CAAC;IACpF,MAAM,sBAAsB,GAAG,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,CAAC,MAAM,CAAC,eAAe,CAAC;IAE1F,kBAAkB;IAClB,IAAI,MAAoC,CAAC;IACzC,IAAI,MAAM,CAAC,MAAM,KAAK,KAAK,EAAE,CAAC;QAC5B,MAAM,GAAG,KAAK,CAAC;IACjB,CAAC;SAAM,IAAI,MAAM,CAAC,MAAM,KAAK,OAAO,EAAE,CAAC;QACrC,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC;IAC3C,CAAC;SAAM,CAAC;QACN,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,OAAO,CAAC;IAC3C,CAAC;IAED,OAAO;QACL,MAAM;QACN,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,eAAe,EAAE,OAAO,CAAC,eAAe;QACxC,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,cAAc,EAAE,OAAO,CAAC,cAAc;QACtC,kBAAkB,EAAE,OAAO,CAAC,kBAAkB;QAC9C,cAAc,EAAE,OAAO,CAAC,cAAc;QACtC,iBAAiB;QACjB,mBAAmB;QACnB,oBAAoB;QACpB,sBAAsB;QACtB,MAAM;KACP,CAAC;AACJ,CAAC;AAED,SAAS,UAAU,CAAC,GAAY;IAC9B,MAAM,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;IACtB,IAAI,KAAK,CAAC,CAAC,CAAC;QAAE,OAAO,CAAC,CAAC;IACvB,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACjD,CAAC"}
@@ -10,7 +10,7 @@ ${output}
10
10
  ${assertion.text}
11
11
 
12
12
  Does the LLM output satisfy this assertion? Respond with JSON only: { "pass": boolean, "reasoning": "..." }`;
13
- const raw = await client.generate(JUDGE_SYSTEM, userPrompt);
13
+ const { text: raw } = await client.generate(JUDGE_SYSTEM, userPrompt);
14
14
  const parsed = parseJudgeResponse(raw);
15
15
  return {
16
16
  id: assertion.id,
@@ -1 +1 @@
1
- {"version":3,"file":"judge.js","sourceRoot":"","sources":["../../src/eval/judge.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,4BAA4B;AAC5B,8EAA8E;AAY9E,MAAM,YAAY,GAAG,qNAAqN,CAAC;AAE3O,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,MAAc,EACd,SAAoB,EACpB,MAAiB;IAEjB,MAAM,UAAU,GAAG;EACnB,MAAM;;;EAGN,SAAS,CAAC,IAAI;;4GAE4F,CAAC;IAE3G,MAAM,GAAG,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,YAAY,EAAE,UAAU,CAAC,CAAC;IAE5D,MAAM,MAAM,GAAG,kBAAkB,CAAC,GAAG,CAAC,CAAC;IAEvC,OAAO;QACL,EAAE,EAAE,SAAS,CAAC,EAAE;QAChB,IAAI,EAAE,SAAS,CAAC,IAAI;QACpB,IAAI,EAAE,MAAM,CAAC,IAAI;QACjB,SAAS,EAAE,MAAM,CAAC,SAAS;KAC5B,CAAC;AACJ,CAAC;AAED,SAAS,kBAAkB,CAAC,GAAW;IACrC,4CAA4C;IAC5C,MAAM,UAAU,GAAG,GAAG,CAAC,KAAK,CAAC,kCAAkC,CAAC,CAAC;IACjE,MAAM,OAAO,GAAG,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;IAEjD,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QACnC,IAAI,OAAO,MAAM,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YACrC,MAAM,IAAI,KAAK,CAAC,oBAAoB,CAAC,CAAC;QACxC,CAAC;QACD,OAAO;YACL,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,SAAS,EAAE,OAAO,MAAM,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE;SACxE,CAAC;IACJ,CAAC;IAAC,MAAM,CAAC;QACP,MAAM,IAAI,KAAK,CACb,sEAAsE,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAC1F,CAAC;IACJ,CAAC;AACH,CAAC"}
1
+ {"version":3,"file":"judge.js","sourceRoot":"","sources":["../../src/eval/judge.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,4BAA4B;AAC5B,8EAA8E;AAY9E,MAAM,YAAY,GAAG,qNAAqN,CAAC;AAE3O,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,MAAc,EACd,SAAoB,EACpB,MAAiB;IAEjB,MAAM,UAAU,GAAG;EACnB,MAAM;;;EAGN,SAAS,CAAC,IAAI;;4GAE4F,CAAC;IAE3G,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,YAAY,EAAE,UAAU,CAAC,CAAC;IAEtE,MAAM,MAAM,GAAG,kBAAkB,CAAC,GAAG,CAAC,CAAC;IAEvC,OAAO;QACL,EAAE,EAAE,SAAS,CAAC,EAAE;QAChB,IAAI,EAAE,SAAS,CAAC,IAAI;QACpB,IAAI,EAAE,MAAM,CAAC,IAAI;QACjB,SAAS,EAAE,MAAM,CAAC,SAAS;KAC5B,CAAC;AACJ,CAAC;AAED,SAAS,kBAAkB,CAAC,GAAW;IACrC,4CAA4C;IAC5C,MAAM,UAAU,GAAG,GAAG,CAAC,KAAK,CAAC,kCAAkC,CAAC,CAAC;IACjE,MAAM,OAAO,GAAG,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;IAEjD,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QACnC,IAAI,OAAO,MAAM,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YACrC,MAAM,IAAI,KAAK,CAAC,oBAAoB,CAAC,CAAC;QACxC,CAAC;QACD,OAAO;YACL,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,SAAS,EAAE,OAAO,MAAM,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE;SACxE,CAAC;IACJ,CAAC;IAAC,MAAM,CAAC;QACP,MAAM,IAAI,KAAK,CACb,sEAAsE,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAC1F,CAAC;IACJ,CAAC;AACH,CAAC"}
@@ -1,5 +1,11 @@
1
+ export interface GenerateResult {
2
+ text: string;
3
+ durationMs: number;
4
+ inputTokens: number | null;
5
+ outputTokens: number | null;
6
+ }
1
7
  export interface LlmClient {
2
- generate(systemPrompt: string, userPrompt: string): Promise<string>;
8
+ generate(systemPrompt: string, userPrompt: string): Promise<GenerateResult>;
3
9
  readonly model: string;
4
10
  }
5
11
  export type ProviderName = "anthropic" | "claude-cli" | "ollama";
package/dist/eval/llm.js CHANGED
@@ -59,6 +59,7 @@ function createAnthropicClient(modelOverride) {
59
59
  }
60
60
  const controller = new AbortController();
61
61
  const timeout = setTimeout(() => controller.abort(), 120_000);
62
+ const start = Date.now();
62
63
  try {
63
64
  const response = await clientInstance.messages.create({
64
65
  model,
@@ -66,8 +67,15 @@ function createAnthropicClient(modelOverride) {
66
67
  messages: [{ role: "user", content: userPrompt }],
67
68
  max_tokens: 4096,
68
69
  }, { signal: controller.signal });
70
+ const durationMs = Date.now() - start;
69
71
  const textBlock = response.content.find((b) => b.type === "text");
70
- return textBlock && "text" in textBlock ? textBlock.text : "";
72
+ const text = textBlock && "text" in textBlock ? textBlock.text : "";
73
+ return {
74
+ text,
75
+ durationMs,
76
+ inputTokens: response.usage?.input_tokens ?? null,
77
+ outputTokens: response.usage?.output_tokens ?? null,
78
+ };
71
79
  }
72
80
  finally {
73
81
  clearTimeout(timeout);
@@ -92,7 +100,8 @@ function createClaudeCliClient(modelOverride, forced = false) {
92
100
  model: `claude-${model}`,
93
101
  async generate(systemPrompt, userPrompt) {
94
102
  const combinedPrompt = `${systemPrompt}\n\n${userPrompt}`;
95
- return new Promise((resolve, reject) => {
103
+ const start = Date.now();
104
+ const text = await new Promise((resolve, reject) => {
96
105
  // Strip all CLAUDE* env vars so the child process doesn't detect nesting
97
106
  const cleanEnv = {};
98
107
  for (const [k, v] of Object.entries(process.env)) {
@@ -133,6 +142,7 @@ function createClaudeCliClient(modelOverride, forced = false) {
133
142
  // Pipe prompt via stdin — avoids ARG_MAX limits for large SKILL.md files
134
143
  proc.stdin.end(combinedPrompt);
135
144
  });
145
+ return { text, durationMs: Date.now() - start, inputTokens: null, outputTokens: null };
136
146
  },
137
147
  };
138
148
  }
@@ -146,6 +156,7 @@ function createOllamaClient(modelOverride) {
146
156
  model,
147
157
  async generate(systemPrompt, userPrompt) {
148
158
  const fullPrompt = `${systemPrompt}\n\n${userPrompt}`;
159
+ const start = Date.now();
149
160
  const response = await fetch(`${baseUrl}/api/generate`, {
150
161
  method: "POST",
151
162
  headers: { "Content-Type": "application/json" },
@@ -168,7 +179,12 @@ function createOllamaClient(modelOverride) {
168
179
  throw new Error(`Ollama request failed: ${error}`);
169
180
  }
170
181
  const data = (await response.json());
171
- return data.response || "";
182
+ return {
183
+ text: data.response || "",
184
+ durationMs: Date.now() - start,
185
+ inputTokens: data.prompt_eval_count ?? null,
186
+ outputTokens: data.eval_count ?? null,
187
+ };
172
188
  },
173
189
  };
174
190
  }
@@ -1 +1 @@
1
- {"version":3,"file":"llm.js","sourceRoot":"","sources":["../../src/eval/llm.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,gFAAgF;AAChF,EAAE;AACF,uDAAuD;AACvD,wEAAwE;AACxE,8DAA8D;AAC9D,uEAAuE;AACvE,EAAE;AACF,uDAAuD;AACvD,4DAA4D;AAC5D,6CAA6C;AAC7C,2DAA2D;AAC3D,EAAE;AACF,iDAAiD;AACjD,+DAA+D;AAC/D,mEAAmE;AACnE,mDAAmD;AACnD,8EAA8E;AAE9E,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAS3C,SAAS,cAAc;IACrB,IAAI,OAAO,CAAC,GAAG,CAAC,UAAU;QAAE,OAAO,QAAQ,CAAC;IAC5C,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB;QAAE,OAAO,WAAW,CAAC;IACtD,OAAO,YAAY,CAAC;AACtB,CAAC;AAOD,MAAM,UAAU,eAAe,CAAC,SAAwB;IACtD,MAAM,QAAQ,GAAG,CAAC,SAAS,EAAE,QAAQ,IAAI,OAAO,CAAC,GAAG,CAAC,oBAAoB,IAAI,cAAc,EAAE,CAAiB,CAAC;IAC/G,MAAM,aAAa,GAAG,SAAS,EAAE,KAAK,CAAC;IACvC,MAAM,MAAM,GAAG,CAAC,CAAC,SAAS,EAAE,QAAQ,CAAC;IAErC,QAAQ,QAAQ,EAAE,CAAC;QACjB,KAAK,WAAW;YACd,OAAO,qBAAqB,CAAC,aAAa,CAAC,CAAC;QAC9C,KAAK,YAAY;YACf,OAAO,qBAAqB,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC;QACtD,KAAK,QAAQ;YACX,OAAO,kBAAkB,CAAC,aAAa,CAAC,CAAC;QAC3C;YACE,MAAM,IAAI,KAAK,CACb,kCAAkC,QAAQ,gDAAgD,CAC3F,CAAC;IACN,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,0BAA0B;AAC1B,8EAA8E;AAC9E,SAAS,qBAAqB,CAAC,aAAsB;IACnD,MAAM,aAAa,GAAG,0BAA0B,CAAC;IAEjD,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;IAC7C,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,IAAI,KAAK,CACb,sRAAsR,CACvR,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,aAAa,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,aAAa,CAAC;IAC9E,IAAI,cAAc,GAAQ,IAAI,CAAC;IAE/B,OAAO;QACL,KAAK;QACL,KAAK,CAAC,QAAQ,CAAC,YAAoB,EAAE,UAAkB;YACrD,IAAI,CAAC,cAAc,EAAE,CAAC;gBACpB,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;gBACjE,cAAc,GAAG,IAAI,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;YAC7C,CAAC;YAED,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;YACzC,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,OAAO,CAAC,CAAC;YAC9D,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,MAAM,cAAc,CAAC,QAAQ,CAAC,MAAM,CACnD;oBACE,KAAK;oBACL,MAAM,EAAE,YAAY;oBACpB,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC;oBACjD,UAAU,EAAE,IAAI;iBACjB,EACD,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,EAAE,CAC9B,CAAC;gBAEF,MAAM,SAAS,GAAG,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC;gBACvE,OAAO,SAAS,IAAI,MAAM,IAAI,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;YAChE,CAAC;oBAAS,CAAC;gBACT,YAAY,CAAC,OAAO,CAAC,CAAC;YACxB,CAAC;QACH,CAAC;KACF,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,4EAA4E;AAC5E,EAAE;AACF,uEAAuE;AACvE,EAAE;AACF,6DAA6D;AAC7D,oFAAoF;AACpF,8EAA8E;AAC9E,SAAS,qBAAqB,CAAC,aAAsB,EAAE,MAAM,GAAG,KAAK;IACnE,IAAI,OAAO,CAAC,GAAG,CAAC,UAAU,IAAI,CAAC,MAAM,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CACb,6JAA6J,CAC9J,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,aAAa,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,QAAQ,CAAC;IAEzE,OAAO;QACL,KAAK,EAAE,UAAU,KAAK,EAAE;QACxB,KAAK,CAAC,QAAQ,CAAC,YAAoB,EAAE,UAAkB;YACrD,MAAM,cAAc,GAAG,GAAG,YAAY,OAAO,UAAU,EAAE,CAAC;YAE1D,OAAO,IAAI,OAAO,CAAS,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;gBAC7C,yEAAyE;gBACzE,MAAM,QAAQ,GAA2B,EAAE,CAAC;gBAC5C,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;oBACjD,IAAI,CAAC,KAAK,SAAS,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC;wBAAE,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;gBAClE,CAAC;gBACD,MAAM,IAAI,GAAG,KAAK,CAAC,QAAQ,EAAE,CAAC,IAAI,EAAE,SAAS,EAAE,KAAK,CAAC,EAAE;oBACrD,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC;oBAC/B,GAAG,EAAE,QAAQ;iBACd,CAAC,CAAC;gBAEH,IAAI,MAAM,GAAG,EAAE,CAAC;gBAChB,IAAI,MAAM,GAAG,EAAE,CAAC;gBAChB,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,GAAG,MAAM,IAAI,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;gBACnE,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,GAAG,MAAM,IAAI,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;gBAEnE,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;oBAC5B,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;oBACrB,MAAM,CAAC,IAAI,KAAK,CAAC,iCAAiC,CAAC,CAAC,CAAC;gBACvD,CAAC,EAAE,OAAO,CAAC,CAAC;gBAEZ,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,GAA0B,EAAE,EAAE;oBAC9C,YAAY,CAAC,KAAK,CAAC,CAAC;oBACpB,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;wBAC1B,MAAM,CAAC,IAAI,KAAK,CACd,qJAAqJ,CACtJ,CAAC,CAAC;oBACL,CAAC;yBAAM,CAAC;wBACN,MAAM,CAAC,IAAI,KAAK,CAAC,sBAAsB,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;oBACzD,CAAC;gBACH,CAAC,CAAC,CAAC;gBAEH,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,EAAE;oBACxB,YAAY,CAAC,KAAK,CAAC,CAAC;oBACpB,IAAI,IAAI,KAAK,CAAC,EAAE,CAAC;wBACf,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;oBACzB,CAAC;yBAAM,CAAC;wBACN,MAAM,MAAM,GAAG,CAAC,MAAM,IAAI,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;wBAChD,MAAM,CAAC,IAAI,KAAK,CAAC,+BAA+B,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;oBACzF,CAAC;gBACH,CAAC,CAAC,CAAC;gBAEH,yEAAyE;gBACzE,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;YACjC,CAAC,CAAC,CAAC;QACL,CAAC;KACF,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,qDAAqD;AACrD,8EAA8E;AAC9E,SAAS,kBAAkB,CAAC,aAAsB;IAChD,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,wBAAwB,CAAC;IACxE,MAAM,KAAK,GAAG,aAAa,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,aAAa,CAAC;IAE9E,OAAO;QACL,KAAK;QACL,KAAK,CAAC,QAAQ,CAAC,YAAoB,EAAE,UAAkB;YACrD,MAAM,UAAU,GAAG,GAAG,YAAY,OAAO,UAAU,EAAE,CAAC;YAEtD,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,eAAe,EAAE;gBACtD,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;gBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACnB,KAAK;oBACL,MAAM,EAAE,UAAU;oBAClB,MAAM,EAAE,KAAK;oBACb,OAAO,EAAE;wBACP,WAAW,EAAE,IAAI;wBACjB,WAAW,EAAE,GAAG;qBACjB;iBACF,CAAC;gBACF,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,OAAO,CAAC;aACrC,CAAC,CAAC;YAEH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBACjB,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;gBACpC,IAAI,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC;oBAC3D,MAAM,IAAI,KAAK,CACb,iBAAiB,KAAK,8CAA8C,KAAK,EAAE,CAC5E,CAAC;gBACJ,CAAC;gBACD,MAAM,IAAI,KAAK,CAAC,0BAA0B,KAAK,EAAE,CAAC,CAAC;YACrD,CAAC;YAED,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAA0B,CAAC;YAC9D,OAAO,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC;QAC7B,CAAC;KACF,CAAC;AACJ,CAAC"}
1
+ {"version":3,"file":"llm.js","sourceRoot":"","sources":["../../src/eval/llm.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,gFAAgF;AAChF,EAAE;AACF,uDAAuD;AACvD,wEAAwE;AACxE,8DAA8D;AAC9D,uEAAuE;AACvE,EAAE;AACF,uDAAuD;AACvD,4DAA4D;AAC5D,6CAA6C;AAC7C,2DAA2D;AAC3D,EAAE;AACF,iDAAiD;AACjD,+DAA+D;AAC/D,mEAAmE;AACnE,mDAAmD;AACnD,8EAA8E;AAE9E,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAgB3C,SAAS,cAAc;IACrB,IAAI,OAAO,CAAC,GAAG,CAAC,UAAU;QAAE,OAAO,QAAQ,CAAC;IAC5C,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB;QAAE,OAAO,WAAW,CAAC;IACtD,OAAO,YAAY,CAAC;AACtB,CAAC;AAOD,MAAM,UAAU,eAAe,CAAC,SAAwB;IACtD,MAAM,QAAQ,GAAG,CAAC,SAAS,EAAE,QAAQ,IAAI,OAAO,CAAC,GAAG,CAAC,oBAAoB,IAAI,cAAc,EAAE,CAAiB,CAAC;IAC/G,MAAM,aAAa,GAAG,SAAS,EAAE,KAAK,CAAC;IACvC,MAAM,MAAM,GAAG,CAAC,CAAC,SAAS,EAAE,QAAQ,CAAC;IAErC,QAAQ,QAAQ,EAAE,CAAC;QACjB,KAAK,WAAW;YACd,OAAO,qBAAqB,CAAC,aAAa,CAAC,CAAC;QAC9C,KAAK,YAAY;YACf,OAAO,qBAAqB,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC;QACtD,KAAK,QAAQ;YACX,OAAO,kBAAkB,CAAC,aAAa,CAAC,CAAC;QAC3C;YACE,MAAM,IAAI,KAAK,CACb,kCAAkC,QAAQ,gDAAgD,CAC3F,CAAC;IACN,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,0BAA0B;AAC1B,8EAA8E;AAC9E,SAAS,qBAAqB,CAAC,aAAsB;IACnD,MAAM,aAAa,GAAG,0BAA0B,CAAC;IAEjD,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;IAC7C,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,MAAM,IAAI,KAAK,CACb,sRAAsR,CACvR,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,aAAa,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,aAAa,CAAC;IAC9E,IAAI,cAAc,GAAQ,IAAI,CAAC;IAE/B,OAAO;QACL,KAAK;QACL,KAAK,CAAC,QAAQ,CAAC,YAAoB,EAAE,UAAkB;YACrD,IAAI,CAAC,cAAc,EAAE,CAAC;gBACpB,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;gBACjE,cAAc,GAAG,IAAI,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;YAC7C,CAAC;YAED,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;YACzC,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,OAAO,CAAC,CAAC;YAC9D,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YACzB,IAAI,CAAC;gBACH,MAAM,QAAQ,GAAG,MAAM,cAAc,CAAC,QAAQ,CAAC,MAAM,CACnD;oBACE,KAAK;oBACL,MAAM,EAAE,YAAY;oBACpB,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC;oBACjD,UAAU,EAAE,IAAI;iBACjB,EACD,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,EAAE,CAC9B,CAAC;gBACF,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC;gBAEtC,MAAM,SAAS,GAAG,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAM,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC;gBACvE,MAAM,IAAI,GAAG,SAAS,IAAI,MAAM,IAAI,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;gBACpE,OAAO;oBACL,IAAI;oBACJ,UAAU;oBACV,WAAW,EAAE,QAAQ,CAAC,KAAK,EAAE,YAAY,IAAI,IAAI;oBACjD,YAAY,EAAE,QAAQ,CAAC,KAAK,EAAE,aAAa,IAAI,IAAI;iBACpD,CAAC;YACJ,CAAC;oBAAS,CAAC;gBACT,YAAY,CAAC,OAAO,CAAC,CAAC;YACxB,CAAC;QACH,CAAC;KACF,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,4EAA4E;AAC5E,EAAE;AACF,uEAAuE;AACvE,EAAE;AACF,6DAA6D;AAC7D,oFAAoF;AACpF,8EAA8E;AAC9E,SAAS,qBAAqB,CAAC,aAAsB,EAAE,MAAM,GAAG,KAAK;IACnE,IAAI,OAAO,CAAC,GAAG,CAAC,UAAU,IAAI,CAAC,MAAM,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CACb,6JAA6J,CAC9J,CAAC;IACJ,CAAC;IAED,MAAM,KAAK,GAAG,aAAa,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,QAAQ,CAAC;IAEzE,OAAO;QACL,KAAK,EAAE,UAAU,KAAK,EAAE;QACxB,KAAK,CAAC,QAAQ,CAAC,YAAoB,EAAE,UAAkB;YACrD,MAAM,cAAc,GAAG,GAAG,YAAY,OAAO,UAAU,EAAE,CAAC;YAC1D,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAEzB,MAAM,IAAI,GAAG,MAAM,IAAI,OAAO,CAAS,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;gBACzD,yEAAyE;gBACzE,MAAM,QAAQ,GAA2B,EAAE,CAAC;gBAC5C,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC;oBACjD,IAAI,CAAC,KAAK,SAAS,IAAI,CAAC,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC;wBAAE,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;gBAClE,CAAC;gBACD,MAAM,IAAI,GAAG,KAAK,CAAC,QAAQ,EAAE,CAAC,IAAI,EAAE,SAAS,EAAE,KAAK,CAAC,EAAE;oBACrD,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC;oBAC/B,GAAG,EAAE,QAAQ;iBACd,CAAC,CAAC;gBAEH,IAAI,MAAM,GAAG,EAAE,CAAC;gBAChB,IAAI,MAAM,GAAG,EAAE,CAAC;gBAChB,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,GAAG,MAAM,IAAI,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;gBACnE,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,CAAS,EAAE,EAAE,GAAG,MAAM,IAAI,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;gBAEnE,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;oBAC5B,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;oBACrB,MAAM,CAAC,IAAI,KAAK,CAAC,iCAAiC,CAAC,CAAC,CAAC;gBACvD,CAAC,EAAE,OAAO,CAAC,CAAC;gBAEZ,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,GAA0B,EAAE,EAAE;oBAC9C,YAAY,CAAC,KAAK,CAAC,CAAC;oBACpB,IAAI,GAAG,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;wBAC1B,MAAM,CAAC,IAAI,KAAK,CACd,qJAAqJ,CACtJ,CAAC,CAAC;oBACL,CAAC;yBAAM,CAAC;wBACN,MAAM,CAAC,IAAI,KAAK,CAAC,sBAAsB,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;oBACzD,CAAC;gBACH,CAAC,CAAC,CAAC;gBAEH,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,EAAE;oBACxB,YAAY,CAAC,KAAK,CAAC,CAAC;oBACpB,IAAI,IAAI,KAAK,CAAC,EAAE,CAAC;wBACf,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;oBACzB,CAAC;yBAAM,CAAC;wBACN,MAAM,MAAM,GAAG,CAAC,MAAM,IAAI,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;wBAChD,MAAM,CAAC,IAAI,KAAK,CAAC,+BAA+B,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;oBACzF,CAAC;gBACH,CAAC,CAAC,CAAC;gBAEH,yEAAyE;gBACzE,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC;YACjC,CAAC,CAAC,CAAC;YAEH,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK,EAAE,WAAW,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;QACzF,CAAC;KACF,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,qDAAqD;AACrD,8EAA8E;AAC9E,SAAS,kBAAkB,CAAC,aAAsB;IAChD,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,wBAAwB,CAAC;IACxE,MAAM,KAAK,GAAG,aAAa,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,aAAa,CAAC;IAE9E,OAAO;QACL,KAAK;QACL,KAAK,CAAC,QAAQ,CAAC,YAAoB,EAAE,UAAkB;YACrD,MAAM,UAAU,GAAG,GAAG,YAAY,OAAO,UAAU,EAAE,CAAC;YACtD,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAEzB,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,eAAe,EAAE;gBACtD,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;gBAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACnB,KAAK;oBACL,MAAM,EAAE,UAAU;oBAClB,MAAM,EAAE,KAAK;oBACb,OAAO,EAAE;wBACP,WAAW,EAAE,IAAI;wBACjB,WAAW,EAAE,GAAG;qBACjB;iBACF,CAAC;gBACF,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,OAAO,CAAC;aACrC,CAAC,CAAC;YAEH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;gBACjB,MAAM,KAAK,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;gBACpC,IAAI,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC;oBAC3D,MAAM,IAAI,KAAK,CACb,iBAAiB,KAAK,8CAA8C,KAAK,EAAE,CAC5E,CAAC;gBACJ,CAAC;gBACD,MAAM,IAAI,KAAK,CAAC,0BAA0B,KAAK,EAAE,CAAC,CAAC;YACrD,CAAC;YAED,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAIlC,CAAC;YACF,OAAO;gBACL,IAAI,EAAE,IAAI,CAAC,QAAQ,IAAI,EAAE;gBACzB,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,KAAK;gBAC9B,WAAW,EAAE,IAAI,CAAC,iBAAiB,IAAI,IAAI;gBAC3C,YAAY,EAAE,IAAI,CAAC,UAAU,IAAI,IAAI;aACtC,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC"}