@agentv/eval 0.2.0 → 0.2.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/index.cjs +2 -0
- package/dist/index.cjs.map +1 -1
- package/dist/index.d.cts +8 -0
- package/dist/index.d.ts +8 -0
- package/dist/index.js +2 -0
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
package/dist/index.cjs
CHANGED
|
@@ -57,6 +57,8 @@ var ToolCallSchema = import_zod.z.object({
|
|
|
57
57
|
});
|
|
58
58
|
var OutputMessageSchema = import_zod.z.object({
|
|
59
59
|
role: import_zod.z.enum(["assistant", "user", "system", "tool"]),
|
|
60
|
+
// Optional message name (e.g., agent name) used by some providers for multi-agent transcripts.
|
|
61
|
+
name: import_zod.z.string().optional(),
|
|
60
62
|
content: import_zod.z.union([import_zod.z.string(), import_zod.z.record(import_zod.z.unknown()), import_zod.z.array(import_zod.z.record(import_zod.z.unknown()))]).optional(),
|
|
61
63
|
toolCalls: import_zod.z.array(ToolCallSchema).optional(),
|
|
62
64
|
timestamp: import_zod.z.string().optional(),
|
package/dist/index.cjs.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/index.ts","../src/schemas.ts","../src/runtime.ts","../src/case-conversion.ts"],"sourcesContent":["/**\n * AgentV Evaluation SDK\n *\n * Build custom code judges for evaluating AI agent outputs.\n *\n * @example\n * ```typescript\n * #!/usr/bin/env bun\n * import { defineCodeJudge } from '@agentv/eval';\n *\n * export default defineCodeJudge(({ traceSummary, candidateAnswer }) => ({\n * score: traceSummary?.eventCount <= 5 ? 1.0 : 0.5,\n * hits: ['Efficient tool usage'],\n * misses: [],\n * }));\n * ```\n *\n * @packageDocumentation\n */\n\n// Re-export schemas and types\nexport {\n CodeJudgeInputSchema,\n CodeJudgeResultSchema,\n TraceSummarySchema,\n OutputMessageSchema,\n ToolCallSchema,\n TokenUsageSchema,\n type CodeJudgeInput,\n type CodeJudgeResult,\n type TraceSummary,\n type OutputMessage,\n type ToolCall,\n type TokenUsage,\n} from './schemas.js';\n\n// Re-export Zod for typed config support\nexport { z } from 'zod';\n\n// Import runtime\nimport { type CodeJudgeHandler, runCodeJudge } from './runtime.js';\n\nexport type { CodeJudgeHandler };\n\n/**\n * Define a code judge evaluator with automatic stdin/stdout handling.\n *\n * This function:\n * 1. Reads JSON from stdin (snake_case format)\n * 2. Converts to camelCase and validates with Zod\n * 3. Calls your handler with typed input\n * 4. Validates the result and outputs JSON to stdout\n * 5. Handles errors gracefully with proper exit codes\n *\n * @param handler - Function that evaluates the input and returns a result\n *\n * @example\n * ```typescript\n * import { defineCodeJudge } from '@agentv/eval';\n *\n * export default defineCodeJudge(({ traceSummary }) => {\n * if (!traceSummary) {\n * return { score: 0.5, reasoning: 'No trace available' };\n * }\n *\n * const efficient = traceSummary.eventCount <= 10;\n * return {\n * score: efficient ? 1.0 : 0.5,\n * hits: efficient ? ['Efficient execution'] : [],\n * misses: efficient ? [] : ['Too many tool calls'],\n * };\n * });\n * ```\n *\n * @example With typed config\n * ```typescript\n * import { defineCodeJudge, z } from '@agentv/eval';\n *\n * const ConfigSchema = z.object({\n * maxToolCalls: z.number().default(10),\n * });\n *\n * export default defineCodeJudge(({ traceSummary, config }) => {\n * const { maxToolCalls } = ConfigSchema.parse(config ?? {});\n * // Use maxToolCalls...\n * });\n * ```\n */\nexport function defineCodeJudge(handler: CodeJudgeHandler): void {\n // Run immediately when module is loaded\n runCodeJudge(handler);\n}\n","/**\n * Zod schemas for code judge input/output validation.\n * Provides both compile-time types and runtime validation.\n */\nimport { z } from 'zod';\n\n/**\n * Token usage metrics schema.\n */\nexport const TokenUsageSchema = z.object({\n input: z.number(),\n output: z.number(),\n cached: z.number().optional(),\n});\n\n/**\n * Trace summary schema (camelCase for TypeScript ergonomics).\n */\nexport const TraceSummarySchema = z.object({\n eventCount: z.number(),\n toolNames: z.array(z.string()),\n toolCallsByName: z.record(z.string(), z.number()),\n errorCount: z.number(),\n tokenUsage: TokenUsageSchema.optional(),\n costUsd: z.number().optional(),\n durationMs: z.number().optional(),\n toolDurations: z.record(z.string(), z.array(z.number())).optional(),\n});\n\n/**\n * Tool call schema for output messages.\n */\nexport const ToolCallSchema = z.object({\n tool: z.string(),\n input: z.unknown().optional(),\n output: z.unknown().optional(),\n id: z.string().optional(),\n timestamp: z.string().optional(),\n});\n\n/**\n * Output message schema.\n */\nexport const OutputMessageSchema = z.object({\n role: z.enum(['assistant', 'user', 'system', 'tool']),\n content: z.union([z.string(), z.record(z.unknown()), z.array(z.record(z.unknown()))]).optional(),\n toolCalls: z.array(ToolCallSchema).optional(),\n timestamp: z.string().optional(),\n metadata: z.record(z.unknown()).optional(),\n});\n\n/**\n * Test message schema.\n */\nexport const TestMessageSchema = z.object({\n role: z.enum(['system', 'user', 'assistant', 'tool']),\n content: z.union([z.string(), z.record(z.unknown()), z.array(z.record(z.unknown()))]),\n});\n\n/**\n * Code judge input schema (camelCase, converted from snake_case wire format).\n */\nexport const CodeJudgeInputSchema = z.object({\n question: z.string(),\n expectedOutcome: z.string(),\n expectedMessages: z.array(z.record(z.unknown())),\n referenceAnswer: z.string().optional(),\n candidateAnswer: z.string(),\n outputMessages: z.array(OutputMessageSchema).nullable().optional(),\n guidelineFiles: z.array(z.string()),\n inputFiles: z.array(z.string()),\n inputMessages: z.array(TestMessageSchema),\n traceSummary: TraceSummarySchema.nullable().optional(),\n config: z.record(z.unknown()).nullable().optional(),\n});\n\n/**\n * Code judge result schema (validated before output).\n */\nexport const CodeJudgeResultSchema = z.object({\n score: z.number().min(0).max(1),\n hits: z.array(z.string()).optional().default([]),\n misses: z.array(z.string()).optional().default([]),\n reasoning: z.string().optional(),\n});\n\n/**\n * Inferred types from schemas.\n */\nexport type CodeJudgeInput = z.infer<typeof CodeJudgeInputSchema>;\nexport type CodeJudgeResult = z.infer<typeof CodeJudgeResultSchema>;\nexport type TraceSummary = z.infer<typeof TraceSummarySchema>;\nexport type OutputMessage = z.infer<typeof OutputMessageSchema>;\nexport type ToolCall = z.infer<typeof ToolCallSchema>;\nexport type TokenUsage = z.infer<typeof TokenUsageSchema>;\n","/**\n * Runtime for code judge evaluators.\n * Handles stdin parsing, validation, error handling, and output formatting.\n */\nimport { readFileSync } from 'node:fs';\n\nimport { toCamelCaseDeep } from './case-conversion.js';\nimport {\n type CodeJudgeInput,\n CodeJudgeInputSchema,\n type CodeJudgeResult,\n CodeJudgeResultSchema,\n} from './schemas.js';\n\n/**\n * Handler function type for code judges.\n */\nexport type CodeJudgeHandler = (\n input: CodeJudgeInput,\n) => CodeJudgeResult | Promise<CodeJudgeResult>;\n\n/**\n * Read stdin synchronously (works in both Node.js and Bun).\n */\nfunction readStdin(): string {\n return readFileSync(0, 'utf8');\n}\n\n/**\n * Clamp a value to the range [0, 1].\n */\nfunction clampScore(value: number): number {\n if (Number.isNaN(value) || !Number.isFinite(value)) {\n return 0;\n }\n return Math.max(0, Math.min(1, value));\n}\n\n/**\n * Format an error for output.\n */\nfunction formatError(error: unknown): string {\n if (error instanceof Error) {\n return error.message;\n }\n return String(error);\n}\n\n/**\n * Run a code judge handler with full stdin/stdout handling.\n * This is the internal implementation called by defineCodeJudge.\n */\nexport async function runCodeJudge(handler: CodeJudgeHandler): Promise<void> {\n try {\n // 1. Read stdin\n const stdin = readStdin();\n\n // 2. Parse JSON\n const rawInput = JSON.parse(stdin) as Record<string, unknown>;\n\n // 3. Convert snake_case to camelCase\n const camelInput = toCamelCaseDeep(rawInput);\n\n // 4. Validate input with Zod\n const input = CodeJudgeInputSchema.parse(camelInput);\n\n // 5. Run handler\n const rawResult = await handler(input);\n\n // 6. Validate and normalize output\n const result = CodeJudgeResultSchema.parse({\n ...rawResult,\n score: clampScore(rawResult.score),\n });\n\n // 7. Output JSON\n console.log(JSON.stringify(result, null, 2));\n } catch (error) {\n // Output failure result\n const errorMessage = formatError(error);\n const errorResult: CodeJudgeResult = {\n score: 0,\n hits: [],\n misses: [errorMessage],\n reasoning: `Evaluation failed: ${errorMessage}`,\n };\n console.log(JSON.stringify(errorResult, null, 2));\n process.exit(1);\n }\n}\n","/**\n * Case conversion utilities for JSON payloads.\n * Converts between snake_case (wire format) and camelCase (TypeScript).\n */\n\nfunction toCamelCase(str: string): string {\n // Don't convert keys that start with uppercase (proper nouns/tool names)\n if (/^[A-Z]/.test(str)) {\n return str;\n }\n return str.replace(/_([a-z0-9])/g, (_, letter) => letter.toUpperCase());\n}\n\n/**\n * Recursively converts all keys in an object from snake_case to camelCase.\n * Used to map wire payloads into TypeScript-friendly shapes.\n *\n * @param obj - The object to convert (can be any JSON-serializable value)\n * @returns A new object with all keys converted to camelCase\n */\nexport function toCamelCaseDeep(obj: unknown): unknown {\n if (obj === null || obj === undefined) {\n return obj;\n }\n\n if (Array.isArray(obj)) {\n return obj.map((item) => toCamelCaseDeep(item));\n }\n\n if (typeof obj === 'object') {\n const result: Record<string, unknown> = {};\n for (const [key, value] of Object.entries(obj)) {\n const camelKey = toCamelCase(key);\n result[camelKey] = toCamelCaseDeep(value);\n }\n return result;\n }\n\n return obj;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACIA,iBAAkB;AAKX,IAAM,mBAAmB,aAAE,OAAO;AAAA,EACvC,OAAO,aAAE,OAAO;AAAA,EAChB,QAAQ,aAAE,OAAO;AAAA,EACjB,QAAQ,aAAE,OAAO,EAAE,SAAS;AAC9B,CAAC;AAKM,IAAM,qBAAqB,aAAE,OAAO;AAAA,EACzC,YAAY,aAAE,OAAO;AAAA,EACrB,WAAW,aAAE,MAAM,aAAE,OAAO,CAAC;AAAA,EAC7B,iBAAiB,aAAE,OAAO,aAAE,OAAO,GAAG,aAAE,OAAO,CAAC;AAAA,EAChD,YAAY,aAAE,OAAO;AAAA,EACrB,YAAY,iBAAiB,SAAS;AAAA,EACtC,SAAS,aAAE,OAAO,EAAE,SAAS;AAAA,EAC7B,YAAY,aAAE,OAAO,EAAE,SAAS;AAAA,EAChC,eAAe,aAAE,OAAO,aAAE,OAAO,GAAG,aAAE,MAAM,aAAE,OAAO,CAAC,CAAC,EAAE,SAAS;AACpE,CAAC;AAKM,IAAM,iBAAiB,aAAE,OAAO;AAAA,EACrC,MAAM,aAAE,OAAO;AAAA,EACf,OAAO,aAAE,QAAQ,EAAE,SAAS;AAAA,EAC5B,QAAQ,aAAE,QAAQ,EAAE,SAAS;AAAA,EAC7B,IAAI,aAAE,OAAO,EAAE,SAAS;AAAA,EACxB,WAAW,aAAE,OAAO,EAAE,SAAS;AACjC,CAAC;AAKM,IAAM,sBAAsB,aAAE,OAAO;AAAA,EAC1C,MAAM,aAAE,KAAK,CAAC,aAAa,QAAQ,UAAU,MAAM,CAAC;AAAA,EACpD,SAAS,aAAE,MAAM,CAAC,aAAE,OAAO,GAAG,aAAE,OAAO,aAAE,QAAQ,CAAC,GAAG,aAAE,MAAM,aAAE,OAAO,aAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,SAAS;AAAA,EAC/F,WAAW,aAAE,MAAM,cAAc,EAAE,SAAS;AAAA,EAC5C,WAAW,aAAE,OAAO,EAAE,SAAS;AAAA,EAC/B,UAAU,aAAE,OAAO,aAAE,QAAQ,CAAC,EAAE,SAAS;AAC3C,CAAC;AAKM,IAAM,oBAAoB,aAAE,OAAO;AAAA,EACxC,MAAM,aAAE,KAAK,CAAC,UAAU,QAAQ,aAAa,MAAM,CAAC;AAAA,EACpD,SAAS,aAAE,MAAM,CAAC,aAAE,OAAO,GAAG,aAAE,OAAO,aAAE,QAAQ,CAAC,GAAG,aAAE,MAAM,aAAE,OAAO,aAAE,QAAQ,CAAC,CAAC,CAAC,CAAC;AACtF,CAAC;AAKM,IAAM,uBAAuB,aAAE,OAAO;AAAA,EAC3C,UAAU,aAAE,OAAO;AAAA,EACnB,iBAAiB,aAAE,OAAO;AAAA,EAC1B,kBAAkB,aAAE,MAAM,aAAE,OAAO,aAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,iBAAiB,aAAE,OAAO,EAAE,SAAS;AAAA,EACrC,iBAAiB,aAAE,OAAO;AAAA,EAC1B,gBAAgB,aAAE,MAAM,mBAAmB,EAAE,SAAS,EAAE,SAAS;AAAA,EACjE,gBAAgB,aAAE,MAAM,aAAE,OAAO,CAAC;AAAA,EAClC,YAAY,aAAE,MAAM,aAAE,OAAO,CAAC;AAAA,EAC9B,eAAe,aAAE,MAAM,iBAAiB;AAAA,EACxC,cAAc,mBAAmB,SAAS,EAAE,SAAS;AAAA,EACrD,QAAQ,aAAE,OAAO,aAAE,QAAQ,CAAC,EAAE,SAAS,EAAE,SAAS;AACpD,CAAC;AAKM,IAAM,wBAAwB,aAAE,OAAO;AAAA,EAC5C,OAAO,aAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC;AAAA,EAC9B,MAAM,aAAE,MAAM,aAAE,OAAO,CAAC,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,QAAQ,aAAE,MAAM,aAAE,OAAO,CAAC,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;AAAA,EACjD,WAAW,aAAE,OAAO,EAAE,SAAS;AACjC,CAAC;;;AD/CD,IAAAA,cAAkB;;;AEjClB,qBAA6B;;;ACC7B,SAAS,YAAY,KAAqB;AAExC,MAAI,SAAS,KAAK,GAAG,GAAG;AACtB,WAAO;AAAA,EACT;AACA,SAAO,IAAI,QAAQ,gBAAgB,CAAC,GAAG,WAAW,OAAO,YAAY,CAAC;AACxE;AASO,SAAS,gBAAgB,KAAuB;AACrD,MAAI,QAAQ,QAAQ,QAAQ,QAAW;AACrC,WAAO;AAAA,EACT;AAEA,MAAI,MAAM,QAAQ,GAAG,GAAG;AACtB,WAAO,IAAI,IAAI,CAAC,SAAS,gBAAgB,IAAI,CAAC;AAAA,EAChD;AAEA,MAAI,OAAO,QAAQ,UAAU;AAC3B,UAAM,SAAkC,CAAC;AACzC,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,GAAG,GAAG;AAC9C,YAAM,WAAW,YAAY,GAAG;AAChC,aAAO,QAAQ,IAAI,gBAAgB,KAAK;AAAA,IAC1C;AACA,WAAO;AAAA,EACT;AAEA,SAAO;AACT;;;ADfA,SAAS,YAAoB;AAC3B,aAAO,6BAAa,GAAG,MAAM;AAC/B;AAKA,SAAS,WAAW,OAAuB;AACzC,MAAI,OAAO,MAAM,KAAK,KAAK,CAAC,OAAO,SAAS,KAAK,GAAG;AAClD,WAAO;AAAA,EACT;AACA,SAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,KAAK,CAAC;AACvC;AAKA,SAAS,YAAY,OAAwB;AAC3C,MAAI,iBAAiB,OAAO;AAC1B,WAAO,MAAM;AAAA,EACf;AACA,SAAO,OAAO,KAAK;AACrB;AAMA,eAAsB,aAAa,SAA0C;AAC3E,MAAI;AAEF,UAAM,QAAQ,UAAU;AAGxB,UAAM,WAAW,KAAK,MAAM,KAAK;AAGjC,UAAM,aAAa,gBAAgB,QAAQ;AAG3C,UAAM,QAAQ,qBAAqB,MAAM,UAAU;AAGnD,UAAM,YAAY,MAAM,QAAQ,KAAK;AAGrC,UAAM,SAAS,sBAAsB,MAAM;AAAA,MACzC,GAAG;AAAA,MACH,OAAO,WAAW,UAAU,KAAK;AAAA,IACnC,CAAC;AAGD,YAAQ,IAAI,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AAAA,EAC7C,SAAS,OAAO;AAEd,UAAM,eAAe,YAAY,KAAK;AACtC,UAAM,cAA+B;AAAA,MACnC,OAAO;AAAA,MACP,MAAM,CAAC;AAAA,MACP,QAAQ,CAAC,YAAY;AAAA,MACrB,WAAW,sBAAsB,YAAY;AAAA,IAC/C;AACA,YAAQ,IAAI,KAAK,UAAU,aAAa,MAAM,CAAC,CAAC;AAChD,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;;;AFDO,SAAS,gBAAgB,SAAiC;AAE/D,eAAa,OAAO;AACtB;","names":["import_zod"]}
|
|
1
|
+
{"version":3,"sources":["../src/index.ts","../src/schemas.ts","../src/runtime.ts","../src/case-conversion.ts"],"sourcesContent":["/**\n * AgentV Evaluation SDK\n *\n * Build custom code judges for evaluating AI agent outputs.\n *\n * @example\n * ```typescript\n * #!/usr/bin/env bun\n * import { defineCodeJudge } from '@agentv/eval';\n *\n * export default defineCodeJudge(({ traceSummary, candidateAnswer }) => ({\n * score: traceSummary?.eventCount <= 5 ? 1.0 : 0.5,\n * hits: ['Efficient tool usage'],\n * misses: [],\n * }));\n * ```\n *\n * @packageDocumentation\n */\n\n// Re-export schemas and types\nexport {\n CodeJudgeInputSchema,\n CodeJudgeResultSchema,\n TraceSummarySchema,\n OutputMessageSchema,\n ToolCallSchema,\n TokenUsageSchema,\n type CodeJudgeInput,\n type CodeJudgeResult,\n type TraceSummary,\n type OutputMessage,\n type ToolCall,\n type TokenUsage,\n} from './schemas.js';\n\n// Re-export Zod for typed config support\nexport { z } from 'zod';\n\n// Import runtime\nimport { type CodeJudgeHandler, runCodeJudge } from './runtime.js';\n\nexport type { CodeJudgeHandler };\n\n/**\n * Define a code judge evaluator with automatic stdin/stdout handling.\n *\n * This function:\n * 1. Reads JSON from stdin (snake_case format)\n * 2. Converts to camelCase and validates with Zod\n * 3. Calls your handler with typed input\n * 4. Validates the result and outputs JSON to stdout\n * 5. Handles errors gracefully with proper exit codes\n *\n * @param handler - Function that evaluates the input and returns a result\n *\n * @example\n * ```typescript\n * import { defineCodeJudge } from '@agentv/eval';\n *\n * export default defineCodeJudge(({ traceSummary }) => {\n * if (!traceSummary) {\n * return { score: 0.5, reasoning: 'No trace available' };\n * }\n *\n * const efficient = traceSummary.eventCount <= 10;\n * return {\n * score: efficient ? 1.0 : 0.5,\n * hits: efficient ? ['Efficient execution'] : [],\n * misses: efficient ? [] : ['Too many tool calls'],\n * };\n * });\n * ```\n *\n * @example With typed config\n * ```typescript\n * import { defineCodeJudge, z } from '@agentv/eval';\n *\n * const ConfigSchema = z.object({\n * maxToolCalls: z.number().default(10),\n * });\n *\n * export default defineCodeJudge(({ traceSummary, config }) => {\n * const { maxToolCalls } = ConfigSchema.parse(config ?? {});\n * // Use maxToolCalls...\n * });\n * ```\n */\nexport function defineCodeJudge(handler: CodeJudgeHandler): void {\n // Run immediately when module is loaded\n runCodeJudge(handler);\n}\n","/**\n * Zod schemas for code judge input/output validation.\n * Provides both compile-time types and runtime validation.\n */\nimport { z } from 'zod';\n\n/**\n * Token usage metrics schema.\n */\nexport const TokenUsageSchema = z.object({\n input: z.number(),\n output: z.number(),\n cached: z.number().optional(),\n});\n\n/**\n * Trace summary schema (camelCase for TypeScript ergonomics).\n */\nexport const TraceSummarySchema = z.object({\n eventCount: z.number(),\n toolNames: z.array(z.string()),\n toolCallsByName: z.record(z.string(), z.number()),\n errorCount: z.number(),\n tokenUsage: TokenUsageSchema.optional(),\n costUsd: z.number().optional(),\n durationMs: z.number().optional(),\n toolDurations: z.record(z.string(), z.array(z.number())).optional(),\n});\n\n/**\n * Tool call schema for output messages.\n */\nexport const ToolCallSchema = z.object({\n tool: z.string(),\n input: z.unknown().optional(),\n output: z.unknown().optional(),\n id: z.string().optional(),\n timestamp: z.string().optional(),\n});\n\n/**\n * Output message schema.\n */\nexport const OutputMessageSchema = z.object({\n role: z.enum(['assistant', 'user', 'system', 'tool']),\n // Optional message name (e.g., agent name) used by some providers for multi-agent transcripts.\n name: z.string().optional(),\n content: z.union([z.string(), z.record(z.unknown()), z.array(z.record(z.unknown()))]).optional(),\n toolCalls: z.array(ToolCallSchema).optional(),\n timestamp: z.string().optional(),\n metadata: z.record(z.unknown()).optional(),\n});\n\n/**\n * Test message schema.\n */\nexport const TestMessageSchema = z.object({\n role: z.enum(['system', 'user', 'assistant', 'tool']),\n content: z.union([z.string(), z.record(z.unknown()), z.array(z.record(z.unknown()))]),\n});\n\n/**\n * Code judge input schema (camelCase, converted from snake_case wire format).\n */\nexport const CodeJudgeInputSchema = z.object({\n question: z.string(),\n expectedOutcome: z.string(),\n expectedMessages: z.array(z.record(z.unknown())),\n referenceAnswer: z.string().optional(),\n candidateAnswer: z.string(),\n outputMessages: z.array(OutputMessageSchema).nullable().optional(),\n guidelineFiles: z.array(z.string()),\n inputFiles: z.array(z.string()),\n inputMessages: z.array(TestMessageSchema),\n traceSummary: TraceSummarySchema.nullable().optional(),\n config: z.record(z.unknown()).nullable().optional(),\n});\n\n/**\n * Code judge result schema (validated before output).\n */\nexport const CodeJudgeResultSchema = z.object({\n score: z.number().min(0).max(1),\n hits: z.array(z.string()).optional().default([]),\n misses: z.array(z.string()).optional().default([]),\n reasoning: z.string().optional(),\n});\n\n/**\n * Inferred types from schemas.\n */\nexport type CodeJudgeInput = z.infer<typeof CodeJudgeInputSchema>;\nexport type CodeJudgeResult = z.infer<typeof CodeJudgeResultSchema>;\nexport type TraceSummary = z.infer<typeof TraceSummarySchema>;\nexport type OutputMessage = z.infer<typeof OutputMessageSchema>;\nexport type ToolCall = z.infer<typeof ToolCallSchema>;\nexport type TokenUsage = z.infer<typeof TokenUsageSchema>;\n","/**\n * Runtime for code judge evaluators.\n * Handles stdin parsing, validation, error handling, and output formatting.\n */\nimport { readFileSync } from 'node:fs';\n\nimport { toCamelCaseDeep } from './case-conversion.js';\nimport {\n type CodeJudgeInput,\n CodeJudgeInputSchema,\n type CodeJudgeResult,\n CodeJudgeResultSchema,\n} from './schemas.js';\n\n/**\n * Handler function type for code judges.\n */\nexport type CodeJudgeHandler = (\n input: CodeJudgeInput,\n) => CodeJudgeResult | Promise<CodeJudgeResult>;\n\n/**\n * Read stdin synchronously (works in both Node.js and Bun).\n */\nfunction readStdin(): string {\n return readFileSync(0, 'utf8');\n}\n\n/**\n * Clamp a value to the range [0, 1].\n */\nfunction clampScore(value: number): number {\n if (Number.isNaN(value) || !Number.isFinite(value)) {\n return 0;\n }\n return Math.max(0, Math.min(1, value));\n}\n\n/**\n * Format an error for output.\n */\nfunction formatError(error: unknown): string {\n if (error instanceof Error) {\n return error.message;\n }\n return String(error);\n}\n\n/**\n * Run a code judge handler with full stdin/stdout handling.\n * This is the internal implementation called by defineCodeJudge.\n */\nexport async function runCodeJudge(handler: CodeJudgeHandler): Promise<void> {\n try {\n // 1. Read stdin\n const stdin = readStdin();\n\n // 2. Parse JSON\n const rawInput = JSON.parse(stdin) as Record<string, unknown>;\n\n // 3. Convert snake_case to camelCase\n const camelInput = toCamelCaseDeep(rawInput);\n\n // 4. Validate input with Zod\n const input = CodeJudgeInputSchema.parse(camelInput);\n\n // 5. Run handler\n const rawResult = await handler(input);\n\n // 6. Validate and normalize output\n const result = CodeJudgeResultSchema.parse({\n ...rawResult,\n score: clampScore(rawResult.score),\n });\n\n // 7. Output JSON\n console.log(JSON.stringify(result, null, 2));\n } catch (error) {\n // Output failure result\n const errorMessage = formatError(error);\n const errorResult: CodeJudgeResult = {\n score: 0,\n hits: [],\n misses: [errorMessage],\n reasoning: `Evaluation failed: ${errorMessage}`,\n };\n console.log(JSON.stringify(errorResult, null, 2));\n process.exit(1);\n }\n}\n","/**\n * Case conversion utilities for JSON payloads.\n * Converts between snake_case (wire format) and camelCase (TypeScript).\n */\n\nfunction toCamelCase(str: string): string {\n // Don't convert keys that start with uppercase (proper nouns/tool names)\n if (/^[A-Z]/.test(str)) {\n return str;\n }\n return str.replace(/_([a-z0-9])/g, (_, letter) => letter.toUpperCase());\n}\n\n/**\n * Recursively converts all keys in an object from snake_case to camelCase.\n * Used to map wire payloads into TypeScript-friendly shapes.\n *\n * @param obj - The object to convert (can be any JSON-serializable value)\n * @returns A new object with all keys converted to camelCase\n */\nexport function toCamelCaseDeep(obj: unknown): unknown {\n if (obj === null || obj === undefined) {\n return obj;\n }\n\n if (Array.isArray(obj)) {\n return obj.map((item) => toCamelCaseDeep(item));\n }\n\n if (typeof obj === 'object') {\n const result: Record<string, unknown> = {};\n for (const [key, value] of Object.entries(obj)) {\n const camelKey = toCamelCase(key);\n result[camelKey] = toCamelCaseDeep(value);\n }\n return result;\n }\n\n return obj;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACIA,iBAAkB;AAKX,IAAM,mBAAmB,aAAE,OAAO;AAAA,EACvC,OAAO,aAAE,OAAO;AAAA,EAChB,QAAQ,aAAE,OAAO;AAAA,EACjB,QAAQ,aAAE,OAAO,EAAE,SAAS;AAC9B,CAAC;AAKM,IAAM,qBAAqB,aAAE,OAAO;AAAA,EACzC,YAAY,aAAE,OAAO;AAAA,EACrB,WAAW,aAAE,MAAM,aAAE,OAAO,CAAC;AAAA,EAC7B,iBAAiB,aAAE,OAAO,aAAE,OAAO,GAAG,aAAE,OAAO,CAAC;AAAA,EAChD,YAAY,aAAE,OAAO;AAAA,EACrB,YAAY,iBAAiB,SAAS;AAAA,EACtC,SAAS,aAAE,OAAO,EAAE,SAAS;AAAA,EAC7B,YAAY,aAAE,OAAO,EAAE,SAAS;AAAA,EAChC,eAAe,aAAE,OAAO,aAAE,OAAO,GAAG,aAAE,MAAM,aAAE,OAAO,CAAC,CAAC,EAAE,SAAS;AACpE,CAAC;AAKM,IAAM,iBAAiB,aAAE,OAAO;AAAA,EACrC,MAAM,aAAE,OAAO;AAAA,EACf,OAAO,aAAE,QAAQ,EAAE,SAAS;AAAA,EAC5B,QAAQ,aAAE,QAAQ,EAAE,SAAS;AAAA,EAC7B,IAAI,aAAE,OAAO,EAAE,SAAS;AAAA,EACxB,WAAW,aAAE,OAAO,EAAE,SAAS;AACjC,CAAC;AAKM,IAAM,sBAAsB,aAAE,OAAO;AAAA,EAC1C,MAAM,aAAE,KAAK,CAAC,aAAa,QAAQ,UAAU,MAAM,CAAC;AAAA;AAAA,EAEpD,MAAM,aAAE,OAAO,EAAE,SAAS;AAAA,EAC1B,SAAS,aAAE,MAAM,CAAC,aAAE,OAAO,GAAG,aAAE,OAAO,aAAE,QAAQ,CAAC,GAAG,aAAE,MAAM,aAAE,OAAO,aAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,SAAS;AAAA,EAC/F,WAAW,aAAE,MAAM,cAAc,EAAE,SAAS;AAAA,EAC5C,WAAW,aAAE,OAAO,EAAE,SAAS;AAAA,EAC/B,UAAU,aAAE,OAAO,aAAE,QAAQ,CAAC,EAAE,SAAS;AAC3C,CAAC;AAKM,IAAM,oBAAoB,aAAE,OAAO;AAAA,EACxC,MAAM,aAAE,KAAK,CAAC,UAAU,QAAQ,aAAa,MAAM,CAAC;AAAA,EACpD,SAAS,aAAE,MAAM,CAAC,aAAE,OAAO,GAAG,aAAE,OAAO,aAAE,QAAQ,CAAC,GAAG,aAAE,MAAM,aAAE,OAAO,aAAE,QAAQ,CAAC,CAAC,CAAC,CAAC;AACtF,CAAC;AAKM,IAAM,uBAAuB,aAAE,OAAO;AAAA,EAC3C,UAAU,aAAE,OAAO;AAAA,EACnB,iBAAiB,aAAE,OAAO;AAAA,EAC1B,kBAAkB,aAAE,MAAM,aAAE,OAAO,aAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,iBAAiB,aAAE,OAAO,EAAE,SAAS;AAAA,EACrC,iBAAiB,aAAE,OAAO;AAAA,EAC1B,gBAAgB,aAAE,MAAM,mBAAmB,EAAE,SAAS,EAAE,SAAS;AAAA,EACjE,gBAAgB,aAAE,MAAM,aAAE,OAAO,CAAC;AAAA,EAClC,YAAY,aAAE,MAAM,aAAE,OAAO,CAAC;AAAA,EAC9B,eAAe,aAAE,MAAM,iBAAiB;AAAA,EACxC,cAAc,mBAAmB,SAAS,EAAE,SAAS;AAAA,EACrD,QAAQ,aAAE,OAAO,aAAE,QAAQ,CAAC,EAAE,SAAS,EAAE,SAAS;AACpD,CAAC;AAKM,IAAM,wBAAwB,aAAE,OAAO;AAAA,EAC5C,OAAO,aAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC;AAAA,EAC9B,MAAM,aAAE,MAAM,aAAE,OAAO,CAAC,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,QAAQ,aAAE,MAAM,aAAE,OAAO,CAAC,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;AAAA,EACjD,WAAW,aAAE,OAAO,EAAE,SAAS;AACjC,CAAC;;;ADjDD,IAAAA,cAAkB;;;AEjClB,qBAA6B;;;ACC7B,SAAS,YAAY,KAAqB;AAExC,MAAI,SAAS,KAAK,GAAG,GAAG;AACtB,WAAO;AAAA,EACT;AACA,SAAO,IAAI,QAAQ,gBAAgB,CAAC,GAAG,WAAW,OAAO,YAAY,CAAC;AACxE;AASO,SAAS,gBAAgB,KAAuB;AACrD,MAAI,QAAQ,QAAQ,QAAQ,QAAW;AACrC,WAAO;AAAA,EACT;AAEA,MAAI,MAAM,QAAQ,GAAG,GAAG;AACtB,WAAO,IAAI,IAAI,CAAC,SAAS,gBAAgB,IAAI,CAAC;AAAA,EAChD;AAEA,MAAI,OAAO,QAAQ,UAAU;AAC3B,UAAM,SAAkC,CAAC;AACzC,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,GAAG,GAAG;AAC9C,YAAM,WAAW,YAAY,GAAG;AAChC,aAAO,QAAQ,IAAI,gBAAgB,KAAK;AAAA,IAC1C;AACA,WAAO;AAAA,EACT;AAEA,SAAO;AACT;;;ADfA,SAAS,YAAoB;AAC3B,aAAO,6BAAa,GAAG,MAAM;AAC/B;AAKA,SAAS,WAAW,OAAuB;AACzC,MAAI,OAAO,MAAM,KAAK,KAAK,CAAC,OAAO,SAAS,KAAK,GAAG;AAClD,WAAO;AAAA,EACT;AACA,SAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,KAAK,CAAC;AACvC;AAKA,SAAS,YAAY,OAAwB;AAC3C,MAAI,iBAAiB,OAAO;AAC1B,WAAO,MAAM;AAAA,EACf;AACA,SAAO,OAAO,KAAK;AACrB;AAMA,eAAsB,aAAa,SAA0C;AAC3E,MAAI;AAEF,UAAM,QAAQ,UAAU;AAGxB,UAAM,WAAW,KAAK,MAAM,KAAK;AAGjC,UAAM,aAAa,gBAAgB,QAAQ;AAG3C,UAAM,QAAQ,qBAAqB,MAAM,UAAU;AAGnD,UAAM,YAAY,MAAM,QAAQ,KAAK;AAGrC,UAAM,SAAS,sBAAsB,MAAM;AAAA,MACzC,GAAG;AAAA,MACH,OAAO,WAAW,UAAU,KAAK;AAAA,IACnC,CAAC;AAGD,YAAQ,IAAI,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AAAA,EAC7C,SAAS,OAAO;AAEd,UAAM,eAAe,YAAY,KAAK;AACtC,UAAM,cAA+B;AAAA,MACnC,OAAO;AAAA,MACP,MAAM,CAAC;AAAA,MACP,QAAQ,CAAC,YAAY;AAAA,MACrB,WAAW,sBAAsB,YAAY;AAAA,IAC/C;AACA,YAAQ,IAAI,KAAK,UAAU,aAAa,MAAM,CAAC,CAAC;AAChD,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;;;AFDO,SAAS,gBAAgB,SAAiC;AAE/D,eAAa,OAAO;AACtB;","names":["import_zod"]}
|
package/dist/index.d.cts
CHANGED
|
@@ -100,6 +100,7 @@ declare const ToolCallSchema: z.ZodObject<{
|
|
|
100
100
|
*/
|
|
101
101
|
declare const OutputMessageSchema: z.ZodObject<{
|
|
102
102
|
role: z.ZodEnum<["assistant", "user", "system", "tool"]>;
|
|
103
|
+
name: z.ZodOptional<z.ZodString>;
|
|
103
104
|
content: z.ZodOptional<z.ZodUnion<[z.ZodString, z.ZodRecord<z.ZodString, z.ZodUnknown>, z.ZodArray<z.ZodRecord<z.ZodString, z.ZodUnknown>, "many">]>>;
|
|
104
105
|
toolCalls: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
105
106
|
tool: z.ZodString;
|
|
@@ -125,6 +126,7 @@ declare const OutputMessageSchema: z.ZodObject<{
|
|
|
125
126
|
}, "strip", z.ZodTypeAny, {
|
|
126
127
|
role: "tool" | "assistant" | "user" | "system";
|
|
127
128
|
timestamp?: string | undefined;
|
|
129
|
+
name?: string | undefined;
|
|
128
130
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
129
131
|
toolCalls?: {
|
|
130
132
|
tool: string;
|
|
@@ -137,6 +139,7 @@ declare const OutputMessageSchema: z.ZodObject<{
|
|
|
137
139
|
}, {
|
|
138
140
|
role: "tool" | "assistant" | "user" | "system";
|
|
139
141
|
timestamp?: string | undefined;
|
|
142
|
+
name?: string | undefined;
|
|
140
143
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
141
144
|
toolCalls?: {
|
|
142
145
|
tool: string;
|
|
@@ -158,6 +161,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
158
161
|
candidateAnswer: z.ZodString;
|
|
159
162
|
outputMessages: z.ZodOptional<z.ZodNullable<z.ZodArray<z.ZodObject<{
|
|
160
163
|
role: z.ZodEnum<["assistant", "user", "system", "tool"]>;
|
|
164
|
+
name: z.ZodOptional<z.ZodString>;
|
|
161
165
|
content: z.ZodOptional<z.ZodUnion<[z.ZodString, z.ZodRecord<z.ZodString, z.ZodUnknown>, z.ZodArray<z.ZodRecord<z.ZodString, z.ZodUnknown>, "many">]>>;
|
|
162
166
|
toolCalls: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
163
167
|
tool: z.ZodString;
|
|
@@ -183,6 +187,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
183
187
|
}, "strip", z.ZodTypeAny, {
|
|
184
188
|
role: "tool" | "assistant" | "user" | "system";
|
|
185
189
|
timestamp?: string | undefined;
|
|
190
|
+
name?: string | undefined;
|
|
186
191
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
187
192
|
toolCalls?: {
|
|
188
193
|
tool: string;
|
|
@@ -195,6 +200,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
195
200
|
}, {
|
|
196
201
|
role: "tool" | "assistant" | "user" | "system";
|
|
197
202
|
timestamp?: string | undefined;
|
|
203
|
+
name?: string | undefined;
|
|
198
204
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
199
205
|
toolCalls?: {
|
|
200
206
|
tool: string;
|
|
@@ -281,6 +287,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
281
287
|
outputMessages?: {
|
|
282
288
|
role: "tool" | "assistant" | "user" | "system";
|
|
283
289
|
timestamp?: string | undefined;
|
|
290
|
+
name?: string | undefined;
|
|
284
291
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
285
292
|
toolCalls?: {
|
|
286
293
|
tool: string;
|
|
@@ -321,6 +328,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
321
328
|
outputMessages?: {
|
|
322
329
|
role: "tool" | "assistant" | "user" | "system";
|
|
323
330
|
timestamp?: string | undefined;
|
|
331
|
+
name?: string | undefined;
|
|
324
332
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
325
333
|
toolCalls?: {
|
|
326
334
|
tool: string;
|
package/dist/index.d.ts
CHANGED
|
@@ -100,6 +100,7 @@ declare const ToolCallSchema: z.ZodObject<{
|
|
|
100
100
|
*/
|
|
101
101
|
declare const OutputMessageSchema: z.ZodObject<{
|
|
102
102
|
role: z.ZodEnum<["assistant", "user", "system", "tool"]>;
|
|
103
|
+
name: z.ZodOptional<z.ZodString>;
|
|
103
104
|
content: z.ZodOptional<z.ZodUnion<[z.ZodString, z.ZodRecord<z.ZodString, z.ZodUnknown>, z.ZodArray<z.ZodRecord<z.ZodString, z.ZodUnknown>, "many">]>>;
|
|
104
105
|
toolCalls: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
105
106
|
tool: z.ZodString;
|
|
@@ -125,6 +126,7 @@ declare const OutputMessageSchema: z.ZodObject<{
|
|
|
125
126
|
}, "strip", z.ZodTypeAny, {
|
|
126
127
|
role: "tool" | "assistant" | "user" | "system";
|
|
127
128
|
timestamp?: string | undefined;
|
|
129
|
+
name?: string | undefined;
|
|
128
130
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
129
131
|
toolCalls?: {
|
|
130
132
|
tool: string;
|
|
@@ -137,6 +139,7 @@ declare const OutputMessageSchema: z.ZodObject<{
|
|
|
137
139
|
}, {
|
|
138
140
|
role: "tool" | "assistant" | "user" | "system";
|
|
139
141
|
timestamp?: string | undefined;
|
|
142
|
+
name?: string | undefined;
|
|
140
143
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
141
144
|
toolCalls?: {
|
|
142
145
|
tool: string;
|
|
@@ -158,6 +161,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
158
161
|
candidateAnswer: z.ZodString;
|
|
159
162
|
outputMessages: z.ZodOptional<z.ZodNullable<z.ZodArray<z.ZodObject<{
|
|
160
163
|
role: z.ZodEnum<["assistant", "user", "system", "tool"]>;
|
|
164
|
+
name: z.ZodOptional<z.ZodString>;
|
|
161
165
|
content: z.ZodOptional<z.ZodUnion<[z.ZodString, z.ZodRecord<z.ZodString, z.ZodUnknown>, z.ZodArray<z.ZodRecord<z.ZodString, z.ZodUnknown>, "many">]>>;
|
|
162
166
|
toolCalls: z.ZodOptional<z.ZodArray<z.ZodObject<{
|
|
163
167
|
tool: z.ZodString;
|
|
@@ -183,6 +187,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
183
187
|
}, "strip", z.ZodTypeAny, {
|
|
184
188
|
role: "tool" | "assistant" | "user" | "system";
|
|
185
189
|
timestamp?: string | undefined;
|
|
190
|
+
name?: string | undefined;
|
|
186
191
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
187
192
|
toolCalls?: {
|
|
188
193
|
tool: string;
|
|
@@ -195,6 +200,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
195
200
|
}, {
|
|
196
201
|
role: "tool" | "assistant" | "user" | "system";
|
|
197
202
|
timestamp?: string | undefined;
|
|
203
|
+
name?: string | undefined;
|
|
198
204
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
199
205
|
toolCalls?: {
|
|
200
206
|
tool: string;
|
|
@@ -281,6 +287,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
281
287
|
outputMessages?: {
|
|
282
288
|
role: "tool" | "assistant" | "user" | "system";
|
|
283
289
|
timestamp?: string | undefined;
|
|
290
|
+
name?: string | undefined;
|
|
284
291
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
285
292
|
toolCalls?: {
|
|
286
293
|
tool: string;
|
|
@@ -321,6 +328,7 @@ declare const CodeJudgeInputSchema: z.ZodObject<{
|
|
|
321
328
|
outputMessages?: {
|
|
322
329
|
role: "tool" | "assistant" | "user" | "system";
|
|
323
330
|
timestamp?: string | undefined;
|
|
331
|
+
name?: string | undefined;
|
|
324
332
|
content?: string | Record<string, unknown> | Record<string, unknown>[] | undefined;
|
|
325
333
|
toolCalls?: {
|
|
326
334
|
tool: string;
|
package/dist/index.js
CHANGED
|
@@ -24,6 +24,8 @@ var ToolCallSchema = z.object({
|
|
|
24
24
|
});
|
|
25
25
|
var OutputMessageSchema = z.object({
|
|
26
26
|
role: z.enum(["assistant", "user", "system", "tool"]),
|
|
27
|
+
// Optional message name (e.g., agent name) used by some providers for multi-agent transcripts.
|
|
28
|
+
name: z.string().optional(),
|
|
27
29
|
content: z.union([z.string(), z.record(z.unknown()), z.array(z.record(z.unknown()))]).optional(),
|
|
28
30
|
toolCalls: z.array(ToolCallSchema).optional(),
|
|
29
31
|
timestamp: z.string().optional(),
|
package/dist/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["../src/schemas.ts","../src/index.ts","../src/runtime.ts","../src/case-conversion.ts"],"sourcesContent":["/**\n * Zod schemas for code judge input/output validation.\n * Provides both compile-time types and runtime validation.\n */\nimport { z } from 'zod';\n\n/**\n * Token usage metrics schema.\n */\nexport const TokenUsageSchema = z.object({\n input: z.number(),\n output: z.number(),\n cached: z.number().optional(),\n});\n\n/**\n * Trace summary schema (camelCase for TypeScript ergonomics).\n */\nexport const TraceSummarySchema = z.object({\n eventCount: z.number(),\n toolNames: z.array(z.string()),\n toolCallsByName: z.record(z.string(), z.number()),\n errorCount: z.number(),\n tokenUsage: TokenUsageSchema.optional(),\n costUsd: z.number().optional(),\n durationMs: z.number().optional(),\n toolDurations: z.record(z.string(), z.array(z.number())).optional(),\n});\n\n/**\n * Tool call schema for output messages.\n */\nexport const ToolCallSchema = z.object({\n tool: z.string(),\n input: z.unknown().optional(),\n output: z.unknown().optional(),\n id: z.string().optional(),\n timestamp: z.string().optional(),\n});\n\n/**\n * Output message schema.\n */\nexport const OutputMessageSchema = z.object({\n role: z.enum(['assistant', 'user', 'system', 'tool']),\n content: z.union([z.string(), z.record(z.unknown()), z.array(z.record(z.unknown()))]).optional(),\n toolCalls: z.array(ToolCallSchema).optional(),\n timestamp: z.string().optional(),\n metadata: z.record(z.unknown()).optional(),\n});\n\n/**\n * Test message schema.\n */\nexport const TestMessageSchema = z.object({\n role: z.enum(['system', 'user', 'assistant', 'tool']),\n content: z.union([z.string(), z.record(z.unknown()), z.array(z.record(z.unknown()))]),\n});\n\n/**\n * Code judge input schema (camelCase, converted from snake_case wire format).\n */\nexport const CodeJudgeInputSchema = z.object({\n question: z.string(),\n expectedOutcome: z.string(),\n expectedMessages: z.array(z.record(z.unknown())),\n referenceAnswer: z.string().optional(),\n candidateAnswer: z.string(),\n outputMessages: z.array(OutputMessageSchema).nullable().optional(),\n guidelineFiles: z.array(z.string()),\n inputFiles: z.array(z.string()),\n inputMessages: z.array(TestMessageSchema),\n traceSummary: TraceSummarySchema.nullable().optional(),\n config: z.record(z.unknown()).nullable().optional(),\n});\n\n/**\n * Code judge result schema (validated before output).\n */\nexport const CodeJudgeResultSchema = z.object({\n score: z.number().min(0).max(1),\n hits: z.array(z.string()).optional().default([]),\n misses: z.array(z.string()).optional().default([]),\n reasoning: z.string().optional(),\n});\n\n/**\n * Inferred types from schemas.\n */\nexport type CodeJudgeInput = z.infer<typeof CodeJudgeInputSchema>;\nexport type CodeJudgeResult = z.infer<typeof CodeJudgeResultSchema>;\nexport type TraceSummary = z.infer<typeof TraceSummarySchema>;\nexport type OutputMessage = z.infer<typeof OutputMessageSchema>;\nexport type ToolCall = z.infer<typeof ToolCallSchema>;\nexport type TokenUsage = z.infer<typeof TokenUsageSchema>;\n","/**\n * AgentV Evaluation SDK\n *\n * Build custom code judges for evaluating AI agent outputs.\n *\n * @example\n * ```typescript\n * #!/usr/bin/env bun\n * import { defineCodeJudge } from '@agentv/eval';\n *\n * export default defineCodeJudge(({ traceSummary, candidateAnswer }) => ({\n * score: traceSummary?.eventCount <= 5 ? 1.0 : 0.5,\n * hits: ['Efficient tool usage'],\n * misses: [],\n * }));\n * ```\n *\n * @packageDocumentation\n */\n\n// Re-export schemas and types\nexport {\n CodeJudgeInputSchema,\n CodeJudgeResultSchema,\n TraceSummarySchema,\n OutputMessageSchema,\n ToolCallSchema,\n TokenUsageSchema,\n type CodeJudgeInput,\n type CodeJudgeResult,\n type TraceSummary,\n type OutputMessage,\n type ToolCall,\n type TokenUsage,\n} from './schemas.js';\n\n// Re-export Zod for typed config support\nexport { z } from 'zod';\n\n// Import runtime\nimport { type CodeJudgeHandler, runCodeJudge } from './runtime.js';\n\nexport type { CodeJudgeHandler };\n\n/**\n * Define a code judge evaluator with automatic stdin/stdout handling.\n *\n * This function:\n * 1. Reads JSON from stdin (snake_case format)\n * 2. Converts to camelCase and validates with Zod\n * 3. Calls your handler with typed input\n * 4. Validates the result and outputs JSON to stdout\n * 5. Handles errors gracefully with proper exit codes\n *\n * @param handler - Function that evaluates the input and returns a result\n *\n * @example\n * ```typescript\n * import { defineCodeJudge } from '@agentv/eval';\n *\n * export default defineCodeJudge(({ traceSummary }) => {\n * if (!traceSummary) {\n * return { score: 0.5, reasoning: 'No trace available' };\n * }\n *\n * const efficient = traceSummary.eventCount <= 10;\n * return {\n * score: efficient ? 1.0 : 0.5,\n * hits: efficient ? ['Efficient execution'] : [],\n * misses: efficient ? [] : ['Too many tool calls'],\n * };\n * });\n * ```\n *\n * @example With typed config\n * ```typescript\n * import { defineCodeJudge, z } from '@agentv/eval';\n *\n * const ConfigSchema = z.object({\n * maxToolCalls: z.number().default(10),\n * });\n *\n * export default defineCodeJudge(({ traceSummary, config }) => {\n * const { maxToolCalls } = ConfigSchema.parse(config ?? {});\n * // Use maxToolCalls...\n * });\n * ```\n */\nexport function defineCodeJudge(handler: CodeJudgeHandler): void {\n // Run immediately when module is loaded\n runCodeJudge(handler);\n}\n","/**\n * Runtime for code judge evaluators.\n * Handles stdin parsing, validation, error handling, and output formatting.\n */\nimport { readFileSync } from 'node:fs';\n\nimport { toCamelCaseDeep } from './case-conversion.js';\nimport {\n type CodeJudgeInput,\n CodeJudgeInputSchema,\n type CodeJudgeResult,\n CodeJudgeResultSchema,\n} from './schemas.js';\n\n/**\n * Handler function type for code judges.\n */\nexport type CodeJudgeHandler = (\n input: CodeJudgeInput,\n) => CodeJudgeResult | Promise<CodeJudgeResult>;\n\n/**\n * Read stdin synchronously (works in both Node.js and Bun).\n */\nfunction readStdin(): string {\n return readFileSync(0, 'utf8');\n}\n\n/**\n * Clamp a value to the range [0, 1].\n */\nfunction clampScore(value: number): number {\n if (Number.isNaN(value) || !Number.isFinite(value)) {\n return 0;\n }\n return Math.max(0, Math.min(1, value));\n}\n\n/**\n * Format an error for output.\n */\nfunction formatError(error: unknown): string {\n if (error instanceof Error) {\n return error.message;\n }\n return String(error);\n}\n\n/**\n * Run a code judge handler with full stdin/stdout handling.\n * This is the internal implementation called by defineCodeJudge.\n */\nexport async function runCodeJudge(handler: CodeJudgeHandler): Promise<void> {\n try {\n // 1. Read stdin\n const stdin = readStdin();\n\n // 2. Parse JSON\n const rawInput = JSON.parse(stdin) as Record<string, unknown>;\n\n // 3. Convert snake_case to camelCase\n const camelInput = toCamelCaseDeep(rawInput);\n\n // 4. Validate input with Zod\n const input = CodeJudgeInputSchema.parse(camelInput);\n\n // 5. Run handler\n const rawResult = await handler(input);\n\n // 6. Validate and normalize output\n const result = CodeJudgeResultSchema.parse({\n ...rawResult,\n score: clampScore(rawResult.score),\n });\n\n // 7. Output JSON\n console.log(JSON.stringify(result, null, 2));\n } catch (error) {\n // Output failure result\n const errorMessage = formatError(error);\n const errorResult: CodeJudgeResult = {\n score: 0,\n hits: [],\n misses: [errorMessage],\n reasoning: `Evaluation failed: ${errorMessage}`,\n };\n console.log(JSON.stringify(errorResult, null, 2));\n process.exit(1);\n }\n}\n","/**\n * Case conversion utilities for JSON payloads.\n * Converts between snake_case (wire format) and camelCase (TypeScript).\n */\n\nfunction toCamelCase(str: string): string {\n // Don't convert keys that start with uppercase (proper nouns/tool names)\n if (/^[A-Z]/.test(str)) {\n return str;\n }\n return str.replace(/_([a-z0-9])/g, (_, letter) => letter.toUpperCase());\n}\n\n/**\n * Recursively converts all keys in an object from snake_case to camelCase.\n * Used to map wire payloads into TypeScript-friendly shapes.\n *\n * @param obj - The object to convert (can be any JSON-serializable value)\n * @returns A new object with all keys converted to camelCase\n */\nexport function toCamelCaseDeep(obj: unknown): unknown {\n if (obj === null || obj === undefined) {\n return obj;\n }\n\n if (Array.isArray(obj)) {\n return obj.map((item) => toCamelCaseDeep(item));\n }\n\n if (typeof obj === 'object') {\n const result: Record<string, unknown> = {};\n for (const [key, value] of Object.entries(obj)) {\n const camelKey = toCamelCase(key);\n result[camelKey] = toCamelCaseDeep(value);\n }\n return result;\n }\n\n return obj;\n}\n"],"mappings":";AAIA,SAAS,SAAS;AAKX,IAAM,mBAAmB,EAAE,OAAO;AAAA,EACvC,OAAO,EAAE,OAAO;AAAA,EAChB,QAAQ,EAAE,OAAO;AAAA,EACjB,QAAQ,EAAE,OAAO,EAAE,SAAS;AAC9B,CAAC;AAKM,IAAM,qBAAqB,EAAE,OAAO;AAAA,EACzC,YAAY,EAAE,OAAO;AAAA,EACrB,WAAW,EAAE,MAAM,EAAE,OAAO,CAAC;AAAA,EAC7B,iBAAiB,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC;AAAA,EAChD,YAAY,EAAE,OAAO;AAAA,EACrB,YAAY,iBAAiB,SAAS;AAAA,EACtC,SAAS,EAAE,OAAO,EAAE,SAAS;AAAA,EAC7B,YAAY,EAAE,OAAO,EAAE,SAAS;AAAA,EAChC,eAAe,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,EAAE,SAAS;AACpE,CAAC;AAKM,IAAM,iBAAiB,EAAE,OAAO;AAAA,EACrC,MAAM,EAAE,OAAO;AAAA,EACf,OAAO,EAAE,QAAQ,EAAE,SAAS;AAAA,EAC5B,QAAQ,EAAE,QAAQ,EAAE,SAAS;AAAA,EAC7B,IAAI,EAAE,OAAO,EAAE,SAAS;AAAA,EACxB,WAAW,EAAE,OAAO,EAAE,SAAS;AACjC,CAAC;AAKM,IAAM,sBAAsB,EAAE,OAAO;AAAA,EAC1C,MAAM,EAAE,KAAK,CAAC,aAAa,QAAQ,UAAU,MAAM,CAAC;AAAA,EACpD,SAAS,EAAE,MAAM,CAAC,EAAE,OAAO,GAAG,EAAE,OAAO,EAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,SAAS;AAAA,EAC/F,WAAW,EAAE,MAAM,cAAc,EAAE,SAAS;AAAA,EAC5C,WAAW,EAAE,OAAO,EAAE,SAAS;AAAA,EAC/B,UAAU,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,SAAS;AAC3C,CAAC;AAKM,IAAM,oBAAoB,EAAE,OAAO;AAAA,EACxC,MAAM,EAAE,KAAK,CAAC,UAAU,QAAQ,aAAa,MAAM,CAAC;AAAA,EACpD,SAAS,EAAE,MAAM,CAAC,EAAE,OAAO,GAAG,EAAE,OAAO,EAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC;AACtF,CAAC;AAKM,IAAM,uBAAuB,EAAE,OAAO;AAAA,EAC3C,UAAU,EAAE,OAAO;AAAA,EACnB,iBAAiB,EAAE,OAAO;AAAA,EAC1B,kBAAkB,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,iBAAiB,EAAE,OAAO,EAAE,SAAS;AAAA,EACrC,iBAAiB,EAAE,OAAO;AAAA,EAC1B,gBAAgB,EAAE,MAAM,mBAAmB,EAAE,SAAS,EAAE,SAAS;AAAA,EACjE,gBAAgB,EAAE,MAAM,EAAE,OAAO,CAAC;AAAA,EAClC,YAAY,EAAE,MAAM,EAAE,OAAO,CAAC;AAAA,EAC9B,eAAe,EAAE,MAAM,iBAAiB;AAAA,EACxC,cAAc,mBAAmB,SAAS,EAAE,SAAS;AAAA,EACrD,QAAQ,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,SAAS,EAAE,SAAS;AACpD,CAAC;AAKM,IAAM,wBAAwB,EAAE,OAAO;AAAA,EAC5C,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC;AAAA,EAC9B,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;AAAA,EACjD,WAAW,EAAE,OAAO,EAAE,SAAS;AACjC,CAAC;;;AC/CD,SAAS,KAAAA,UAAS;;;ACjClB,SAAS,oBAAoB;;;ACC7B,SAAS,YAAY,KAAqB;AAExC,MAAI,SAAS,KAAK,GAAG,GAAG;AACtB,WAAO;AAAA,EACT;AACA,SAAO,IAAI,QAAQ,gBAAgB,CAAC,GAAG,WAAW,OAAO,YAAY,CAAC;AACxE;AASO,SAAS,gBAAgB,KAAuB;AACrD,MAAI,QAAQ,QAAQ,QAAQ,QAAW;AACrC,WAAO;AAAA,EACT;AAEA,MAAI,MAAM,QAAQ,GAAG,GAAG;AACtB,WAAO,IAAI,IAAI,CAAC,SAAS,gBAAgB,IAAI,CAAC;AAAA,EAChD;AAEA,MAAI,OAAO,QAAQ,UAAU;AAC3B,UAAM,SAAkC,CAAC;AACzC,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,GAAG,GAAG;AAC9C,YAAM,WAAW,YAAY,GAAG;AAChC,aAAO,QAAQ,IAAI,gBAAgB,KAAK;AAAA,IAC1C;AACA,WAAO;AAAA,EACT;AAEA,SAAO;AACT;;;ADfA,SAAS,YAAoB;AAC3B,SAAO,aAAa,GAAG,MAAM;AAC/B;AAKA,SAAS,WAAW,OAAuB;AACzC,MAAI,OAAO,MAAM,KAAK,KAAK,CAAC,OAAO,SAAS,KAAK,GAAG;AAClD,WAAO;AAAA,EACT;AACA,SAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,KAAK,CAAC;AACvC;AAKA,SAAS,YAAY,OAAwB;AAC3C,MAAI,iBAAiB,OAAO;AAC1B,WAAO,MAAM;AAAA,EACf;AACA,SAAO,OAAO,KAAK;AACrB;AAMA,eAAsB,aAAa,SAA0C;AAC3E,MAAI;AAEF,UAAM,QAAQ,UAAU;AAGxB,UAAM,WAAW,KAAK,MAAM,KAAK;AAGjC,UAAM,aAAa,gBAAgB,QAAQ;AAG3C,UAAM,QAAQ,qBAAqB,MAAM,UAAU;AAGnD,UAAM,YAAY,MAAM,QAAQ,KAAK;AAGrC,UAAM,SAAS,sBAAsB,MAAM;AAAA,MACzC,GAAG;AAAA,MACH,OAAO,WAAW,UAAU,KAAK;AAAA,IACnC,CAAC;AAGD,YAAQ,IAAI,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AAAA,EAC7C,SAAS,OAAO;AAEd,UAAM,eAAe,YAAY,KAAK;AACtC,UAAM,cAA+B;AAAA,MACnC,OAAO;AAAA,MACP,MAAM,CAAC;AAAA,MACP,QAAQ,CAAC,YAAY;AAAA,MACrB,WAAW,sBAAsB,YAAY;AAAA,IAC/C;AACA,YAAQ,IAAI,KAAK,UAAU,aAAa,MAAM,CAAC,CAAC;AAChD,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;;;ADDO,SAAS,gBAAgB,SAAiC;AAE/D,eAAa,OAAO;AACtB;","names":["z"]}
|
|
1
|
+
{"version":3,"sources":["../src/schemas.ts","../src/index.ts","../src/runtime.ts","../src/case-conversion.ts"],"sourcesContent":["/**\n * Zod schemas for code judge input/output validation.\n * Provides both compile-time types and runtime validation.\n */\nimport { z } from 'zod';\n\n/**\n * Token usage metrics schema.\n */\nexport const TokenUsageSchema = z.object({\n input: z.number(),\n output: z.number(),\n cached: z.number().optional(),\n});\n\n/**\n * Trace summary schema (camelCase for TypeScript ergonomics).\n */\nexport const TraceSummarySchema = z.object({\n eventCount: z.number(),\n toolNames: z.array(z.string()),\n toolCallsByName: z.record(z.string(), z.number()),\n errorCount: z.number(),\n tokenUsage: TokenUsageSchema.optional(),\n costUsd: z.number().optional(),\n durationMs: z.number().optional(),\n toolDurations: z.record(z.string(), z.array(z.number())).optional(),\n});\n\n/**\n * Tool call schema for output messages.\n */\nexport const ToolCallSchema = z.object({\n tool: z.string(),\n input: z.unknown().optional(),\n output: z.unknown().optional(),\n id: z.string().optional(),\n timestamp: z.string().optional(),\n});\n\n/**\n * Output message schema.\n */\nexport const OutputMessageSchema = z.object({\n role: z.enum(['assistant', 'user', 'system', 'tool']),\n // Optional message name (e.g., agent name) used by some providers for multi-agent transcripts.\n name: z.string().optional(),\n content: z.union([z.string(), z.record(z.unknown()), z.array(z.record(z.unknown()))]).optional(),\n toolCalls: z.array(ToolCallSchema).optional(),\n timestamp: z.string().optional(),\n metadata: z.record(z.unknown()).optional(),\n});\n\n/**\n * Test message schema.\n */\nexport const TestMessageSchema = z.object({\n role: z.enum(['system', 'user', 'assistant', 'tool']),\n content: z.union([z.string(), z.record(z.unknown()), z.array(z.record(z.unknown()))]),\n});\n\n/**\n * Code judge input schema (camelCase, converted from snake_case wire format).\n */\nexport const CodeJudgeInputSchema = z.object({\n question: z.string(),\n expectedOutcome: z.string(),\n expectedMessages: z.array(z.record(z.unknown())),\n referenceAnswer: z.string().optional(),\n candidateAnswer: z.string(),\n outputMessages: z.array(OutputMessageSchema).nullable().optional(),\n guidelineFiles: z.array(z.string()),\n inputFiles: z.array(z.string()),\n inputMessages: z.array(TestMessageSchema),\n traceSummary: TraceSummarySchema.nullable().optional(),\n config: z.record(z.unknown()).nullable().optional(),\n});\n\n/**\n * Code judge result schema (validated before output).\n */\nexport const CodeJudgeResultSchema = z.object({\n score: z.number().min(0).max(1),\n hits: z.array(z.string()).optional().default([]),\n misses: z.array(z.string()).optional().default([]),\n reasoning: z.string().optional(),\n});\n\n/**\n * Inferred types from schemas.\n */\nexport type CodeJudgeInput = z.infer<typeof CodeJudgeInputSchema>;\nexport type CodeJudgeResult = z.infer<typeof CodeJudgeResultSchema>;\nexport type TraceSummary = z.infer<typeof TraceSummarySchema>;\nexport type OutputMessage = z.infer<typeof OutputMessageSchema>;\nexport type ToolCall = z.infer<typeof ToolCallSchema>;\nexport type TokenUsage = z.infer<typeof TokenUsageSchema>;\n","/**\n * AgentV Evaluation SDK\n *\n * Build custom code judges for evaluating AI agent outputs.\n *\n * @example\n * ```typescript\n * #!/usr/bin/env bun\n * import { defineCodeJudge } from '@agentv/eval';\n *\n * export default defineCodeJudge(({ traceSummary, candidateAnswer }) => ({\n * score: traceSummary?.eventCount <= 5 ? 1.0 : 0.5,\n * hits: ['Efficient tool usage'],\n * misses: [],\n * }));\n * ```\n *\n * @packageDocumentation\n */\n\n// Re-export schemas and types\nexport {\n CodeJudgeInputSchema,\n CodeJudgeResultSchema,\n TraceSummarySchema,\n OutputMessageSchema,\n ToolCallSchema,\n TokenUsageSchema,\n type CodeJudgeInput,\n type CodeJudgeResult,\n type TraceSummary,\n type OutputMessage,\n type ToolCall,\n type TokenUsage,\n} from './schemas.js';\n\n// Re-export Zod for typed config support\nexport { z } from 'zod';\n\n// Import runtime\nimport { type CodeJudgeHandler, runCodeJudge } from './runtime.js';\n\nexport type { CodeJudgeHandler };\n\n/**\n * Define a code judge evaluator with automatic stdin/stdout handling.\n *\n * This function:\n * 1. Reads JSON from stdin (snake_case format)\n * 2. Converts to camelCase and validates with Zod\n * 3. Calls your handler with typed input\n * 4. Validates the result and outputs JSON to stdout\n * 5. Handles errors gracefully with proper exit codes\n *\n * @param handler - Function that evaluates the input and returns a result\n *\n * @example\n * ```typescript\n * import { defineCodeJudge } from '@agentv/eval';\n *\n * export default defineCodeJudge(({ traceSummary }) => {\n * if (!traceSummary) {\n * return { score: 0.5, reasoning: 'No trace available' };\n * }\n *\n * const efficient = traceSummary.eventCount <= 10;\n * return {\n * score: efficient ? 1.0 : 0.5,\n * hits: efficient ? ['Efficient execution'] : [],\n * misses: efficient ? [] : ['Too many tool calls'],\n * };\n * });\n * ```\n *\n * @example With typed config\n * ```typescript\n * import { defineCodeJudge, z } from '@agentv/eval';\n *\n * const ConfigSchema = z.object({\n * maxToolCalls: z.number().default(10),\n * });\n *\n * export default defineCodeJudge(({ traceSummary, config }) => {\n * const { maxToolCalls } = ConfigSchema.parse(config ?? {});\n * // Use maxToolCalls...\n * });\n * ```\n */\nexport function defineCodeJudge(handler: CodeJudgeHandler): void {\n // Run immediately when module is loaded\n runCodeJudge(handler);\n}\n","/**\n * Runtime for code judge evaluators.\n * Handles stdin parsing, validation, error handling, and output formatting.\n */\nimport { readFileSync } from 'node:fs';\n\nimport { toCamelCaseDeep } from './case-conversion.js';\nimport {\n type CodeJudgeInput,\n CodeJudgeInputSchema,\n type CodeJudgeResult,\n CodeJudgeResultSchema,\n} from './schemas.js';\n\n/**\n * Handler function type for code judges.\n */\nexport type CodeJudgeHandler = (\n input: CodeJudgeInput,\n) => CodeJudgeResult | Promise<CodeJudgeResult>;\n\n/**\n * Read stdin synchronously (works in both Node.js and Bun).\n */\nfunction readStdin(): string {\n return readFileSync(0, 'utf8');\n}\n\n/**\n * Clamp a value to the range [0, 1].\n */\nfunction clampScore(value: number): number {\n if (Number.isNaN(value) || !Number.isFinite(value)) {\n return 0;\n }\n return Math.max(0, Math.min(1, value));\n}\n\n/**\n * Format an error for output.\n */\nfunction formatError(error: unknown): string {\n if (error instanceof Error) {\n return error.message;\n }\n return String(error);\n}\n\n/**\n * Run a code judge handler with full stdin/stdout handling.\n * This is the internal implementation called by defineCodeJudge.\n */\nexport async function runCodeJudge(handler: CodeJudgeHandler): Promise<void> {\n try {\n // 1. Read stdin\n const stdin = readStdin();\n\n // 2. Parse JSON\n const rawInput = JSON.parse(stdin) as Record<string, unknown>;\n\n // 3. Convert snake_case to camelCase\n const camelInput = toCamelCaseDeep(rawInput);\n\n // 4. Validate input with Zod\n const input = CodeJudgeInputSchema.parse(camelInput);\n\n // 5. Run handler\n const rawResult = await handler(input);\n\n // 6. Validate and normalize output\n const result = CodeJudgeResultSchema.parse({\n ...rawResult,\n score: clampScore(rawResult.score),\n });\n\n // 7. Output JSON\n console.log(JSON.stringify(result, null, 2));\n } catch (error) {\n // Output failure result\n const errorMessage = formatError(error);\n const errorResult: CodeJudgeResult = {\n score: 0,\n hits: [],\n misses: [errorMessage],\n reasoning: `Evaluation failed: ${errorMessage}`,\n };\n console.log(JSON.stringify(errorResult, null, 2));\n process.exit(1);\n }\n}\n","/**\n * Case conversion utilities for JSON payloads.\n * Converts between snake_case (wire format) and camelCase (TypeScript).\n */\n\nfunction toCamelCase(str: string): string {\n // Don't convert keys that start with uppercase (proper nouns/tool names)\n if (/^[A-Z]/.test(str)) {\n return str;\n }\n return str.replace(/_([a-z0-9])/g, (_, letter) => letter.toUpperCase());\n}\n\n/**\n * Recursively converts all keys in an object from snake_case to camelCase.\n * Used to map wire payloads into TypeScript-friendly shapes.\n *\n * @param obj - The object to convert (can be any JSON-serializable value)\n * @returns A new object with all keys converted to camelCase\n */\nexport function toCamelCaseDeep(obj: unknown): unknown {\n if (obj === null || obj === undefined) {\n return obj;\n }\n\n if (Array.isArray(obj)) {\n return obj.map((item) => toCamelCaseDeep(item));\n }\n\n if (typeof obj === 'object') {\n const result: Record<string, unknown> = {};\n for (const [key, value] of Object.entries(obj)) {\n const camelKey = toCamelCase(key);\n result[camelKey] = toCamelCaseDeep(value);\n }\n return result;\n }\n\n return obj;\n}\n"],"mappings":";AAIA,SAAS,SAAS;AAKX,IAAM,mBAAmB,EAAE,OAAO;AAAA,EACvC,OAAO,EAAE,OAAO;AAAA,EAChB,QAAQ,EAAE,OAAO;AAAA,EACjB,QAAQ,EAAE,OAAO,EAAE,SAAS;AAC9B,CAAC;AAKM,IAAM,qBAAqB,EAAE,OAAO;AAAA,EACzC,YAAY,EAAE,OAAO;AAAA,EACrB,WAAW,EAAE,MAAM,EAAE,OAAO,CAAC;AAAA,EAC7B,iBAAiB,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,OAAO,CAAC;AAAA,EAChD,YAAY,EAAE,OAAO;AAAA,EACrB,YAAY,iBAAiB,SAAS;AAAA,EACtC,SAAS,EAAE,OAAO,EAAE,SAAS;AAAA,EAC7B,YAAY,EAAE,OAAO,EAAE,SAAS;AAAA,EAChC,eAAe,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,EAAE,SAAS;AACpE,CAAC;AAKM,IAAM,iBAAiB,EAAE,OAAO;AAAA,EACrC,MAAM,EAAE,OAAO;AAAA,EACf,OAAO,EAAE,QAAQ,EAAE,SAAS;AAAA,EAC5B,QAAQ,EAAE,QAAQ,EAAE,SAAS;AAAA,EAC7B,IAAI,EAAE,OAAO,EAAE,SAAS;AAAA,EACxB,WAAW,EAAE,OAAO,EAAE,SAAS;AACjC,CAAC;AAKM,IAAM,sBAAsB,EAAE,OAAO;AAAA,EAC1C,MAAM,EAAE,KAAK,CAAC,aAAa,QAAQ,UAAU,MAAM,CAAC;AAAA;AAAA,EAEpD,MAAM,EAAE,OAAO,EAAE,SAAS;AAAA,EAC1B,SAAS,EAAE,MAAM,CAAC,EAAE,OAAO,GAAG,EAAE,OAAO,EAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,EAAE,SAAS;AAAA,EAC/F,WAAW,EAAE,MAAM,cAAc,EAAE,SAAS;AAAA,EAC5C,WAAW,EAAE,OAAO,EAAE,SAAS;AAAA,EAC/B,UAAU,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,SAAS;AAC3C,CAAC;AAKM,IAAM,oBAAoB,EAAE,OAAO;AAAA,EACxC,MAAM,EAAE,KAAK,CAAC,UAAU,QAAQ,aAAa,MAAM,CAAC;AAAA,EACpD,SAAS,EAAE,MAAM,CAAC,EAAE,OAAO,GAAG,EAAE,OAAO,EAAE,QAAQ,CAAC,GAAG,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC;AACtF,CAAC;AAKM,IAAM,uBAAuB,EAAE,OAAO;AAAA,EAC3C,UAAU,EAAE,OAAO;AAAA,EACnB,iBAAiB,EAAE,OAAO;AAAA,EAC1B,kBAAkB,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,iBAAiB,EAAE,OAAO,EAAE,SAAS;AAAA,EACrC,iBAAiB,EAAE,OAAO;AAAA,EAC1B,gBAAgB,EAAE,MAAM,mBAAmB,EAAE,SAAS,EAAE,SAAS;AAAA,EACjE,gBAAgB,EAAE,MAAM,EAAE,OAAO,CAAC;AAAA,EAClC,YAAY,EAAE,MAAM,EAAE,OAAO,CAAC;AAAA,EAC9B,eAAe,EAAE,MAAM,iBAAiB;AAAA,EACxC,cAAc,mBAAmB,SAAS,EAAE,SAAS;AAAA,EACrD,QAAQ,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,SAAS,EAAE,SAAS;AACpD,CAAC;AAKM,IAAM,wBAAwB,EAAE,OAAO;AAAA,EAC5C,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC;AAAA,EAC9B,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;AAAA,EAC/C,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC;AAAA,EACjD,WAAW,EAAE,OAAO,EAAE,SAAS;AACjC,CAAC;;;ACjDD,SAAS,KAAAA,UAAS;;;ACjClB,SAAS,oBAAoB;;;ACC7B,SAAS,YAAY,KAAqB;AAExC,MAAI,SAAS,KAAK,GAAG,GAAG;AACtB,WAAO;AAAA,EACT;AACA,SAAO,IAAI,QAAQ,gBAAgB,CAAC,GAAG,WAAW,OAAO,YAAY,CAAC;AACxE;AASO,SAAS,gBAAgB,KAAuB;AACrD,MAAI,QAAQ,QAAQ,QAAQ,QAAW;AACrC,WAAO;AAAA,EACT;AAEA,MAAI,MAAM,QAAQ,GAAG,GAAG;AACtB,WAAO,IAAI,IAAI,CAAC,SAAS,gBAAgB,IAAI,CAAC;AAAA,EAChD;AAEA,MAAI,OAAO,QAAQ,UAAU;AAC3B,UAAM,SAAkC,CAAC;AACzC,eAAW,CAAC,KAAK,KAAK,KAAK,OAAO,QAAQ,GAAG,GAAG;AAC9C,YAAM,WAAW,YAAY,GAAG;AAChC,aAAO,QAAQ,IAAI,gBAAgB,KAAK;AAAA,IAC1C;AACA,WAAO;AAAA,EACT;AAEA,SAAO;AACT;;;ADfA,SAAS,YAAoB;AAC3B,SAAO,aAAa,GAAG,MAAM;AAC/B;AAKA,SAAS,WAAW,OAAuB;AACzC,MAAI,OAAO,MAAM,KAAK,KAAK,CAAC,OAAO,SAAS,KAAK,GAAG;AAClD,WAAO;AAAA,EACT;AACA,SAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,KAAK,CAAC;AACvC;AAKA,SAAS,YAAY,OAAwB;AAC3C,MAAI,iBAAiB,OAAO;AAC1B,WAAO,MAAM;AAAA,EACf;AACA,SAAO,OAAO,KAAK;AACrB;AAMA,eAAsB,aAAa,SAA0C;AAC3E,MAAI;AAEF,UAAM,QAAQ,UAAU;AAGxB,UAAM,WAAW,KAAK,MAAM,KAAK;AAGjC,UAAM,aAAa,gBAAgB,QAAQ;AAG3C,UAAM,QAAQ,qBAAqB,MAAM,UAAU;AAGnD,UAAM,YAAY,MAAM,QAAQ,KAAK;AAGrC,UAAM,SAAS,sBAAsB,MAAM;AAAA,MACzC,GAAG;AAAA,MACH,OAAO,WAAW,UAAU,KAAK;AAAA,IACnC,CAAC;AAGD,YAAQ,IAAI,KAAK,UAAU,QAAQ,MAAM,CAAC,CAAC;AAAA,EAC7C,SAAS,OAAO;AAEd,UAAM,eAAe,YAAY,KAAK;AACtC,UAAM,cAA+B;AAAA,MACnC,OAAO;AAAA,MACP,MAAM,CAAC;AAAA,MACP,QAAQ,CAAC,YAAY;AAAA,MACrB,WAAW,sBAAsB,YAAY;AAAA,IAC/C;AACA,YAAQ,IAAI,KAAK,UAAU,aAAa,MAAM,CAAC,CAAC;AAChD,YAAQ,KAAK,CAAC;AAAA,EAChB;AACF;;;ADDO,SAAS,gBAAgB,SAAiC;AAE/D,eAAa,OAAO;AACtB;","names":["z"]}
|