agent-duelist 0.3.0 → 0.3.1

package/README.md

@@ -161,9 +161,11 @@ export default defineArena({
 
 ### Available packs
 
-| Pack | Tasks | Description |
-|------|-------|-------------|
-| `structured-output` | 6 | Zod schema stress test — flat objects, nesting, arrays, enums, empty arrays, and adversarial input |
+| Pack | Tasks | Scorers | Description |
+|------|-------|---------|-------------|
+| `structured-output` | 6 | correctness, schema-correctness, latency, cost | Zod schema stress test — flat objects, nesting, arrays, enums, empty arrays, and adversarial input |
+| `tool-calling` | 4 | tool-usage, latency, cost | Function invocation accuracy — single calls, complex params, tool selection, and parallel calls |
+| `reasoning` | 5 | correctness, latency, cost | Logic, math, and multi-step thinking — arithmetic, deduction, data interpretation, critical path, and business rules |
 
 Packs work with both `run` and `ci` commands:
 
@@ -291,7 +293,7 @@ Scorers turn raw model outputs into **numeric scores** (0–1) with optional det
 | `correctness` | Exact match against `expected` (deep-equal, key-order independent for objects) |
 | `schema-correctness` | Validates output against the task's Zod `schema` via `safeParse()` |
 | `fuzzy-similarity` | Jaccard token-overlap similarity between output and `expected` |
-| `tool-usage` | Whether the model invoked the expected tool(s) during a tool-calling task |
+| `tool-usage` | Tool calling accuracy — checks tool selection and argument correctness (1.0 exact match, 0.5 right tool / wrong args, 0.0 wrong tool) |
 | `llm-judge-correctness` | LLM-as-judge — calls a judge model to score accuracy, completeness, and conciseness |
 
 Configure them in your arena:
@@ -637,7 +639,7 @@ With cost summary, flakiness warnings, and pass/fail verdict.
 - 5 provider types: OpenAI, Azure OpenAI, Anthropic, Google Gemini, and any OpenAI-compatible gateway
 - 7 built-in scorers including LLM-as-judge, tool-usage, schema validation, and fuzzy similarity
 - Tool-calling support with local handlers for agent task benchmarking
-- **Task packs**: built-in benchmark suites (`structured-output`) — run with `--pack`, no config writing needed
+- **Task packs**: built-in benchmark suites (`structured-output`, `tool-calling`, `reasoning`) — run with `--pack`, no config writing needed
 - Quality-first medal ranking: output quality decides medals, efficiency only breaks ties
 - Fair head-to-head benchmarking with parallel provider execution
 - 4 reporters: console (tables + medals + sparklines), JSON, HTML (sortable, self-contained), and Markdown (PR comments)
@@ -649,7 +651,7 @@ With cost summary, flakiness warnings, and pass/fail verdict.
 
 **Planned** (subject to community feedback):
 
-- **More task packs** — reasoning, summarization, tool-calling, and multi-turn conversation packs
+- **More task packs** — summarization, multi-turn conversation, and code generation packs
 - **Agent workflows** — multi-step tool chains, multi-hop reasoning, and agent traces
 - **More export formats** — CSV
 - **Plugin system** — first-class support for user-defined providers and scorers

package/dist/cli.js

@@ -498,7 +498,7 @@ function Pe(e5, t3, s5 = Q.DEFAULT) {
     return p5(), n3.getToken() !== 2 ? k3(7, [2], []) : b3(), true;
   }
   i2(M2, "parseObject");
-  function z3() {
+  function z5() {
     D3(), b3();
     let w4 = true, j2 = false;
     for (; n3.getToken() !== 4 && n3.getToken() !== 17; ) {
@@ -509,11 +509,11 @@ function Pe(e5, t3, s5 = Q.DEFAULT) {
     }
     return L3(), w4 || o7.pop(), n3.getToken() !== 4 ? k3(8, [4], []) : b3(), true;
   }
-  i2(z3, "parseArray");
+  i2(z5, "parseArray");
   function U2() {
     switch (n3.getToken()) {
       case 3:
-        return z3();
+        return z5();
       case 1:
         return M2();
       case 10:
@@ -746,7 +746,7 @@ var init_dist2 = __esm({
     }, "interpolateConfigDir");
     Me = ["outDir", "declarationDir", "outFile", "rootDir", "baseUrl", "tsBuildInfoFile"];
     ze = i2((e5) => {
-      var t3, s5, n3, o7, l3, u5, a7, r3, g2, v4, d5, _4, p5, D3, L3, T3, F4, x, c3, y3, A3, b3, k3, R5, W, V2, M2, z3, U2, w4, j2, S2, $2;
+      var t3, s5, n3, o7, l3, u5, a7, r3, g2, v4, d5, _4, p5, D3, L3, T3, F4, x, c3, y3, A3, b3, k3, R5, W, V2, M2, z5, U2, w4, j2, S2, $2;
       if (e5.strict) {
         const f6 = ["noImplicitAny", "noImplicitThis", "strictNullChecks", "strictFunctionTypes", "strictBindCallApply", "strictPropertyInitialization", "strictBuiltinIteratorReturn", "alwaysStrict", "useUnknownInCatchVariables"];
         for (const B2 of f6) e5[B2] === void 0 && (e5[B2] = true);
@@ -767,7 +767,7 @@ var init_dist2 = __esm({
         let f6 = e5.moduleResolution.toLowerCase();
         f6 === "node" && (f6 = "node10"), e5.moduleResolution = f6, (f6 === "node16" || f6 === "nodenext" || f6 === "bundler") && ((R5 = e5.resolvePackageJsonExports) != null || (e5.resolvePackageJsonExports = true), (W = e5.resolvePackageJsonImports) != null || (e5.resolvePackageJsonImports = true)), f6 === "bundler" && ((V2 = e5.allowSyntheticDefaultImports) != null || (e5.allowSyntheticDefaultImports = true), (M2 = e5.resolveJsonModule) != null || (e5.resolveJsonModule = true));
       }
-      e5.jsx && (e5.jsx = e5.jsx.toLowerCase()), e5.moduleDetection && (e5.moduleDetection = e5.moduleDetection.toLowerCase()), e5.importsNotUsedAsValues && (e5.importsNotUsedAsValues = e5.importsNotUsedAsValues.toLowerCase()), e5.newLine && (e5.newLine = e5.newLine.toLowerCase()), e5.esModuleInterop && ((z3 = e5.allowSyntheticDefaultImports) != null || (e5.allowSyntheticDefaultImports = true)), e5.verbatimModuleSyntax && ((U2 = e5.isolatedModules) != null || (e5.isolatedModules = true), (w4 = e5.preserveConstEnums) != null || (e5.preserveConstEnums = true)), e5.isolatedModules && ((j2 = e5.preserveConstEnums) != null || (e5.preserveConstEnums = true)), e5.rewriteRelativeImportExtensions && ((S2 = e5.allowImportingTsExtensions) != null || (e5.allowImportingTsExtensions = true)), e5.lib && (e5.lib = e5.lib.map((f6) => f6.toLowerCase())), e5.checkJs && (($2 = e5.allowJs) != null || (e5.allowJs = true));
+      e5.jsx && (e5.jsx = e5.jsx.toLowerCase()), e5.moduleDetection && (e5.moduleDetection = e5.moduleDetection.toLowerCase()), e5.importsNotUsedAsValues && (e5.importsNotUsedAsValues = e5.importsNotUsedAsValues.toLowerCase()), e5.newLine && (e5.newLine = e5.newLine.toLowerCase()), e5.esModuleInterop && ((z5 = e5.allowSyntheticDefaultImports) != null || (e5.allowSyntheticDefaultImports = true)), e5.verbatimModuleSyntax && ((U2 = e5.isolatedModules) != null || (e5.isolatedModules = true), (w4 = e5.preserveConstEnums) != null || (e5.preserveConstEnums = true)), e5.isolatedModules && ((j2 = e5.preserveConstEnums) != null || (e5.preserveConstEnums = true)), e5.rewriteRelativeImportExtensions && ((S2 = e5.allowImportingTsExtensions) != null || (e5.allowImportingTsExtensions = true)), e5.lib && (e5.lib = e5.lib.map((f6) => f6.toLowerCase())), e5.checkJs && (($2 = e5.allowJs) != null || (e5.allowJs = true));
     }, "normalizeCompilerOptions");
     pe = i2((e5, t3 = /* @__PURE__ */ new Map()) => {
       const s5 = m3.resolve(e5), n3 = ve(s5, t3), o7 = m3.dirname(s5), { compilerOptions: l3 } = n3;
@@ -3345,14 +3345,14 @@ function fn(s5, e5 = "@") {
               case 32:
                 break;
               case 101: {
-                if (!(d5[400] | 0) && z3(h4) | 0 && !(A3(t3 + 4 | 0, 16, 10) | 0) && ($2(), (b3[804] | 0) == 0)) {
+                if (!(d5[400] | 0) && z5(h4) | 0 && !(A3(t3 + 4 | 0, 16, 10) | 0) && ($2(), (b3[804] | 0) == 0)) {
                   l3 = 9;
                   break e;
                 } else l3 = 17;
                 break;
               }
               case 105: {
-                z3(h4) | 0 && !(A3(t3 + 4 | 0, 26, 10) | 0) && W(), l3 = 17;
+                z5(h4) | 0 && !(A3(t3 + 4 | 0, 26, 10) | 0) && W(), l3 = 17;
                 break;
               }
               case 59: {
@@ -3400,15 +3400,15 @@ function fn(s5, e5 = "@") {
                   case 32:
                     break;
                   case 101: {
-                    !(d5[400] | 0) && z3(a7) | 0 && !(A3(t3 + 4 | 0, 16, 10) | 0) && $2(), l3 = 91;
+                    !(d5[400] | 0) && z5(a7) | 0 && !(A3(t3 + 4 | 0, 16, 10) | 0) && $2(), l3 = 91;
                     break;
                   }
                   case 105: {
-                    z3(a7) | 0 && !(A3(t3 + 4 | 0, 26, 10) | 0) && W(), l3 = 91;
+                    z5(a7) | 0 && !(A3(t3 + 4 | 0, 26, 10) | 0) && W(), l3 = 91;
                     break;
                   }
                   case 99: {
-                    z3(a7) | 0 && !(A3(t3 + 4 | 0, 36, 8) | 0) && P3(d5[t3 + 12 >> 1] | 0) | 0 && (b3[806] = 1), l3 = 91;
+                    z5(a7) | 0 && !(A3(t3 + 4 | 0, 36, 8) | 0) && P3(d5[t3 + 12 >> 1] | 0) | 0 && (b3[806] = 1), l3 = 91;
                     break;
                   }
                   case 40: {
@@ -4437,10 +4437,10 @@ function fn(s5, e5 = "@") {
         return t3 = t3 | 0, (d5[t3 >> 1] | 0) == 46 && (d5[t3 + -2 >> 1] | 0) == 46 ? t3 = (d5[t3 + -4 >> 1] | 0) == 46 : t3 = 0, t3 | 0;
       }
       f2(de3, "G");
-      function z3(t3) {
+      function z5(t3) {
         return t3 = t3 | 0, (r3[3] | 0) == (t3 | 0) ? t3 = 1 : t3 = Oe2(t3 + -2 | 0) | 0, t3 | 0;
       }
-      f2(z3, "H");
+      f2(z5, "H");
       function vt() {
         var t3 = 0;
         return t3 = r3[(r3[62] | 0) + 12 >> 2] | 0, t3 ? t3 = t3 - (r3[3] | 0) >> 1 : t3 = -1, t3 | 0;
@@ -5821,7 +5821,7 @@ import p4 from "path";
 import { fileURLToPath as O4 } from "url";
 import se3, { writeSync as te2 } from "fs";
 import { inspect as oe3 } from "util";
-var K3, o4, R4, D2, me3, N2, j, pe2, y2, C3, de2, E4, ge2, Q4, M, _3, S, A2, T2, Pe3, I4, F3, v3, J3, P2, je3, be2, xe3, k2, $, ye3, Ee, B, G3, _e3, Se3, b2, X3, w2, ve3, z2, we2, Me3, Te3, Fe3, H2, $e3;
+var K3, o4, R4, D2, me3, N2, j, pe2, y2, C3, de2, E4, ge2, Q4, M, _3, S, A2, T2, Pe3, I4, F3, v3, J3, P2, je3, be2, xe3, k2, $, ye3, Ee, B, G3, _e3, Se3, b2, X3, w2, ve3, z4, we2, Me3, Te3, Fe3, H2, $e3;
 var init_register_CFH5oNdT = __esm({
   "node_modules/tsx/dist/register-CFH5oNdT.mjs"() {
     "use strict";
@@ -5995,11 +5995,11 @@ var init_register_CFH5oNdT = __esm({
         throw t3;
       }
     }, "createTsExtensionResolver");
-    z2 = "at cjsPreparseModuleExports (node:internal";
+    z4 = "at cjsPreparseModuleExports (node:internal";
     we2 = o4((s5) => {
       const e5 = s5.stack.split(`
 `).slice(1);
-      return e5[1].includes(z2) || e5[2].includes(z2);
+      return e5[1].includes(z4) || e5[2].includes(z4);
     }, "isFromCjsLexer");
     Me3 = o4((s5, e5) => {
       const a7 = s5.split("?"), n3 = new URLSearchParams(a7[1]);
@@ -7540,6 +7540,27 @@ var costScorer = ({ result }, providerId) => {
   };
 };
 
+// src/utils/deep-equal.ts
+function deepEqual(expected, actual) {
+  if (expected === actual) return true;
+  if (typeof expected === "string" && typeof actual === "string") {
+    return expected.trim().toLowerCase() === actual.trim().toLowerCase();
+  }
+  if (typeof expected !== typeof actual) return false;
+  if (expected === null || actual === null) return expected === actual;
+  if (Array.isArray(expected) && Array.isArray(actual)) {
+    if (expected.length !== actual.length) return false;
+    return expected.every((val, i7) => deepEqual(val, actual[i7]));
+  }
+  if (typeof expected === "object" && typeof actual === "object") {
+    const objExpected = expected;
+    const objActual = actual;
+    const keysExpected = Object.keys(objExpected);
+    return keysExpected.every((key) => key in objActual && deepEqual(objExpected[key], objActual[key]));
+  }
+  return expected === actual;
+}
+
 // src/scorers/correctness.ts
 var correctnessScorer = ({ task, result }) => {
   if (task.expected === void 0) {
@@ -7563,25 +7584,6 @@ function normalizeOutput(expected, actual) {
   }
   return actual;
 }
-function deepEqual(expected, actual) {
-  if (expected === actual) return true;
-  if (typeof expected === "string" && typeof actual === "string") {
-    return expected.trim().toLowerCase() === actual.trim().toLowerCase();
-  }
-  if (typeof expected !== typeof actual) return false;
-  if (expected === null || actual === null) return expected === actual;
-  if (Array.isArray(expected) && Array.isArray(actual)) {
-    if (expected.length !== actual.length) return false;
-    return expected.every((val, i7) => deepEqual(val, actual[i7]));
-  }
-  if (typeof expected === "object" && typeof actual === "object") {
-    const objExpected = expected;
-    const objActual = actual;
-    const keysExpected = Object.keys(objExpected);
-    return keysExpected.every((key) => key in objActual && deepEqual(objExpected[key], objActual[key]));
-  }
-  return expected === actual;
-}
 
 // src/scorers/schema-correctness.ts
 var schemaCorrectnessScorer = ({ task, result }) => {
@@ -7790,15 +7792,54 @@ function parseJudgeResponse(response, model) {
 
 // src/scorers/tool-usage.ts
 var toolUsageScorer = ({ task, result }) => {
-  const expectedToolName = task.tools?.[0]?.name;
-  if (!expectedToolName) {
+  if (!task.tools?.length) {
     return { name: "tool-usage", value: -1, details: { reason: "no tools configured on task" } };
   }
-  const usedTool = result.toolCalls?.some((c3) => c3.name === expectedToolName) ?? false;
+  const calls = result.toolCalls ?? [];
+  const expectedIsObject = task.expected !== void 0 && typeof task.expected === "object" && task.expected !== null && !Array.isArray(task.expected);
+  if (expectedIsObject) {
+    const matchingCall = calls.find((c3) => {
+      const toolDef = task.tools.find((t3) => t3.name === c3.name);
+      if (!toolDef) return false;
+      return deepEqual(task.expected, c3.arguments);
+    });
+    if (matchingCall) {
+      return {
+        name: "tool-usage",
+        value: 1,
+        details: { matchedTool: matchingCall.name, arguments: matchingCall.arguments, toolCalls: calls }
+      };
+    }
+    const expectedKeys = Object.keys(task.expected);
+    const partialMatch = calls.find((c3) => {
+      if (typeof c3.arguments !== "object" || c3.arguments === null) return false;
+      const argKeys = Object.keys(c3.arguments);
+      return expectedKeys.some((k3) => argKeys.includes(k3));
+    });
+    if (partialMatch) {
+      return {
+        name: "tool-usage",
+        value: 0.5,
+        details: {
+          reason: "correct tool but wrong arguments",
+          expected: task.expected,
+          actual: partialMatch.arguments,
+          toolCalls: calls
+        }
+      };
+    }
+    return {
+      name: "tool-usage",
+      value: 0,
+      details: { reason: "no matching tool call", expected: task.expected, toolCalls: calls }
+    };
+  }
+  const expectedToolName = task.tools[0].name;
+  const usedTool = calls.some((c3) => c3.name === expectedToolName);
   return {
     name: "tool-usage",
     value: usedTool ? 1 : 0,
-    details: { expectedToolName, usedTool, toolCalls: result.toolCalls ?? [] }
+    details: { expectedToolName, usedTool, toolCalls: calls }
   };
 };
 
@@ -9706,12 +9747,247 @@ Return as JSON. Use ISO 8601 date format (YYYY-MM-DD).`,
   scorers: ["correctness", "schema-correctness", "latency", "cost"]
 };
 
+// src/packs/tool-calling.ts
+import { z as z2 } from "zod";
+var toolCallingPack = {
+  name: "tool-calling",
+  label: "Tool Calling",
+  description: "Function invocation accuracy \u2014 single calls, complex params, tool selection, parallel calls, and relevance detection",
+  tasks: [
+    {
+      name: "tc:simple-single-tool",
+      prompt: "What's the current weather in Tokyo?",
+      tools: [{
+        name: "getWeather",
+        description: "Get current weather for a city",
+        parameters: z2.object({
+          city: z2.string(),
+          units: z2.enum(["celsius", "fahrenheit"]).optional()
+        }),
+        handler: async ({ city, units }) => ({
+          city,
+          tempC: 8,
+          condition: "cloudy",
+          units: units ?? "celsius"
+        })
+      }],
+      expected: { city: "Tokyo" }
+    },
+    {
+      name: "tc:complex-params",
+      prompt: "Search for Italian restaurants within 2 miles of downtown Portland that are open now and have at least a 4-star rating.",
+      tools: [{
+        name: "searchRestaurants",
+        description: "Search for restaurants matching criteria",
+        parameters: z2.object({
+          cuisine: z2.string(),
+          location: z2.string(),
+          radiusMiles: z2.number(),
+          minRating: z2.number(),
+          openNow: z2.boolean()
+        }),
+        handler: async (_args) => ({
+          results: [{ name: "Trattoria Roma", rating: 4.5, distance: 1.2 }]
+        })
+      }],
+      expected: {
+        cuisine: "Italian",
+        location: "downtown Portland",
+        radiusMiles: 2,
+        minRating: 4,
+        openNow: true
+      }
+    },
+    {
+      name: "tc:select-from-many",
+      prompt: "Convert 150 USD to Euros.",
+      tools: [
+        {
+          name: "getWeather",
+          description: "Get current weather for a city",
+          parameters: z2.object({ city: z2.string() }),
+          handler: async () => ({ tempC: 20 })
+        },
+        {
+          name: "convertCurrency",
+          description: "Convert an amount between currencies",
+          parameters: z2.object({
+            amount: z2.number(),
+            from: z2.string(),
+            to: z2.string()
+          }),
+          handler: async ({ amount, from, to }) => ({
+            amount,
+            from,
+            to,
+            result: 138.75,
+            rate: 0.925
+          })
+        },
+        {
+          name: "translateText",
+          description: "Translate text between languages",
+          parameters: z2.object({ text: z2.string(), targetLang: z2.string() }),
+          handler: async () => ({ translated: "" })
+        },
+        {
+          name: "calculateTip",
+          description: "Calculate tip amount for a bill",
+          parameters: z2.object({ billAmount: z2.number(), tipPercent: z2.number() }),
+          handler: async () => ({ tip: 0 })
+        }
+      ],
+      expected: { amount: 150, from: "USD", to: "EUR" }
+    },
+    {
+      name: "tc:parallel-calls",
+      prompt: "I'm planning a trip. What's the weather like in both Paris and London right now?",
+      tools: [{
+        name: "getWeather",
+        description: "Get current weather for a city",
+        parameters: z2.object({ city: z2.string() }),
+        handler: async ({ city }) => {
+          const data = {
+            Paris: { tempC: 12, condition: "partly cloudy" },
+            London: { tempC: 9, condition: "rainy" }
+          };
+          return data[city] ?? { tempC: 15, condition: "unknown" };
+        }
+      }],
+      expected: "weather data for Paris and London"
+    }
+  ],
+  scorers: ["tool-usage", "latency", "cost"]
+};
+
+// src/packs/reasoning.ts
+import { z as z3 } from "zod";
+var reasoningPack = {
+  name: "reasoning",
+  label: "Reasoning",
+  description: "Logic, math, and multi-step thinking \u2014 arithmetic, deduction, data interpretation, critical path, and business rules",
+  tasks: [
+    {
+      name: "rs:saas-mrr-calc",
+      prompt: `A SaaS company charges $49/month for the basic plan and $149/month for pro.
+In Q1 they had 200 basic subscribers and 85 pro subscribers.
+In Q2, 15% of basic users upgraded to pro and they gained 40 new basic subscribers.
+No one churned. What is the Q2 monthly recurring revenue (MRR)?
+Return as JSON with your reasoning and the final MRR number.`,
+      expected: { mrr: 27425 },
+      schema: z3.object({
+        reasoning: z3.string().optional(),
+        mrr: z3.number()
+      })
+    },
+    {
+      name: "rs:logical-deduction",
+      prompt: `Five developers \u2014 Alice, Bob, Carol, Dave, and Eve \u2014 each use a different
+primary language: Rust, TypeScript, Python, Go, and Java. Given:
+1. Alice does not use Python, Java, or Go.
+2. Bob uses TypeScript.
+3. Carol uses neither Rust nor Go.
+4. Dave does not use Java.
+5. Eve uses neither Rust, Go, nor Java.
+What language does each developer use? Return as JSON.`,
+      expected: {
+        Alice: "Rust",
+        Bob: "TypeScript",
+        Carol: "Java",
+        Dave: "Go",
+        Eve: "Python"
+      },
+      schema: z3.object({
+        Alice: z3.string(),
+        Bob: z3.string(),
+        Carol: z3.string(),
+        Dave: z3.string(),
+        Eve: z3.string()
+      })
+    },
+    {
+      name: "rs:data-interpretation",
+      prompt: `Given this quarterly revenue data:
+| Quarter | Revenue | Growth |
+|---------|---------|--------|
+| Q1 2025 | $2.1M   | -      |
+| Q2 2025 | $2.4M   | 14.3%  |
+| Q3 2025 | $2.2M   | -8.3%  |
+| Q4 2025 | $2.8M   | 27.3%  |
+
+Which quarter had the highest absolute revenue increase compared to the previous
+quarter? What was the full-year total revenue in millions? Return as JSON.`,
+      expected: {
+        highestGrowthQuarter: "Q4 2025",
+        absoluteIncrease: 0.6,
+        fullYearRevenue: 9.5
+      },
+      schema: z3.object({
+        highestGrowthQuarter: z3.string(),
+        absoluteIncrease: z3.number(),
+        fullYearRevenue: z3.number()
+      })
+    },
+    {
+      name: "rs:critical-path",
+      prompt: `A deployment pipeline has these stages with dependencies:
+- Build (3 min, no dependency)
+- Unit tests (5 min, depends on Build)
+- Integration tests (8 min, depends on Build)
+- Security scan (4 min, depends on Build)
+- Staging deploy (2 min, depends on Unit tests AND Integration tests AND Security scan)
+- Smoke tests (3 min, depends on Staging deploy)
+
+Assuming stages run in parallel where possible, what is the total pipeline
+duration in minutes? Which stages are on the critical path? Return as JSON.`,
+      expected: {
+        totalMinutes: 16,
+        criticalPath: ["Build", "Integration tests", "Staging deploy", "Smoke tests"]
+      },
+      schema: z3.object({
+        totalMinutes: z3.number(),
+        criticalPath: z3.array(z3.string())
+      })
+    },
+    {
+      name: "rs:pricing-rules",
+      prompt: `Apply these pricing rules to each customer and return the final price:
+Rules:
+- Base price: $100
+- Enterprise customers (>100 seats): 30% discount
+- Annual billing: additional 15% off the discounted price
+- Non-profit organizations: flat $50 regardless of other rules
+
+Customers:
+A: 50 seats, monthly billing, for-profit
+B: 200 seats, annual billing, for-profit
+C: 75 seats, annual billing, non-profit
+D: 150 seats, monthly billing, for-profit
+
+Return as a JSON array with customer id and finalPrice.`,
+      expected: [
+        { id: "A", finalPrice: 100 },
+        { id: "B", finalPrice: 59.5 },
+        { id: "C", finalPrice: 50 },
+        { id: "D", finalPrice: 70 }
+      ],
+      schema: z3.array(z3.object({
+        id: z3.string(),
+        finalPrice: z3.number()
+      }))
+    }
+  ],
+  scorers: ["correctness", "latency", "cost"]
+};
+
 // src/packs/index.ts
 var registry = /* @__PURE__ */ new Map();
 function register(pack) {
   registry.set(pack.name, pack);
 }
 register(structuredOutputPack);
+register(toolCallingPack);
+register(reasoningPack);
 function loadPack(name) {
   const pack = registry.get(name);
   if (!pack) {
@@ -9893,11 +10169,15 @@ function printPackList() {
     console.log("No packs available.");
     return;
   }
+  const nameWidth = Math.max(...packs.map((p5) => p5.name.length)) + 2;
   console.log("Available task packs:\n");
   for (const p5 of packs) {
-    console.log(`  ${p5.name.padEnd(24)} ${p5.description} (${p5.taskCount} tasks)`);
+    const tasks = `${p5.taskCount} tasks`;
+    console.log(`  ${p5.name.padEnd(nameWidth)} ${tasks.padEnd(9)} ${p5.description}`);
   }
-  console.log("\nUsage: npx duelist run --pack <name>");
+  console.log(`
+Run:     npx duelist run --pack <name>`);
+  console.log(`Combine: npx duelist run --pack structured-output,tool-calling`);
 }
 async function loadArenaWithPacks(packNames, configOpt) {
   const configPath = resolve(configOpt);