@agentv/core 2.0.1 → 2.0.2

package/dist/index.cjs

@@ -67,8 +67,6 @@ __export(index_exports, {
   loadEvalCases: () => loadEvalCases,
   mergeExecutionMetrics: () => mergeExecutionMetrics,
   normalizeLineEndings: () => normalizeLineEndings,
-  parseCodeJudgePayload: () => parseCodeJudgePayload,
-  readCodeJudgePayload: () => readCodeJudgePayload,
   readJsonFile: () => readJsonFile,
   readTargetDefinitions: () => readTargetDefinitions,
   readTestSuiteMetadata: () => readTestSuiteMetadata,
@@ -4272,6 +4270,167 @@ var MockProvider = class {
   }
 };
 
+// src/evaluation/providers/pi-agent-sdk.ts
+var piAgentModule = null;
+var piAiModule = null;
+async function loadPiModules() {
+  if (!piAgentModule || !piAiModule) {
+    try {
+      [piAgentModule, piAiModule] = await Promise.all([
+        import("@mariozechner/pi-agent"),
+        import("@mariozechner/pi-ai")
+      ]);
+    } catch (error) {
+      throw new Error(
+        `Failed to load pi-agent-sdk dependencies. Please install them:
+  npm install @mariozechner/pi-agent @mariozechner/pi-ai
+
+Original error: ${error instanceof Error ? error.message : String(error)}`
+      );
+    }
+  }
+  return {
+    Agent: piAgentModule.Agent,
+    ProviderTransport: piAgentModule.ProviderTransport,
+    getModel: piAiModule.getModel,
+    getEnvApiKey: piAiModule.getEnvApiKey
+  };
+}
+var PiAgentSdkProvider = class {
+  id;
+  kind = "pi-agent-sdk";
+  targetName;
+  supportsBatch = false;
+  config;
+  constructor(targetName, config) {
+    this.id = `pi-agent-sdk:${targetName}`;
+    this.targetName = targetName;
+    this.config = config;
+  }
+  async invoke(request) {
+    if (request.signal?.aborted) {
+      throw new Error("Pi agent SDK request was aborted before execution");
+    }
+    const { Agent, ProviderTransport, getModel, getEnvApiKey } = await loadPiModules();
+    const startTime = Date.now();
+    const providerName = this.config.provider ?? "anthropic";
+    const modelId = this.config.model ?? "claude-sonnet-4-20250514";
+    const model = getModel(providerName, modelId);
+    const systemPrompt = this.config.systemPrompt ?? "Answer directly and concisely.";
+    const transport = new ProviderTransport({
+      getApiKey: async (provider) => {
+        return this.config.apiKey ?? getEnvApiKey(provider) ?? void 0;
+      }
+    });
+    const agent = new Agent({
+      initialState: {
+        systemPrompt,
+        model,
+        tools: [],
+        // No tools for simple Q&A
+        messages: []
+      },
+      transport
+    });
+    const outputMessages = [];
+    let finalAssistantContent = "";
+    const unsubscribe = agent.subscribe((event) => {
+      if (event.type === "message_end") {
+        const msg = event.message;
+        if (msg.role === "assistant") {
+          const content = extractTextContent2(msg.content);
+          if (content) {
+            finalAssistantContent = content;
+          }
+        }
+      }
+    });
+    try {
+      const timeoutMs = this.config.timeoutMs ?? 12e4;
+      const timeoutPromise = new Promise((_, reject) => {
+        setTimeout(
+          () => reject(new Error(`Pi agent SDK timed out after ${timeoutMs}ms`)),
+          timeoutMs
+        );
+      });
+      await Promise.race([agent.prompt(request.question), timeoutPromise]);
+      await agent.waitForIdle();
+      const agentMessages = agent.state.messages;
+      for (const msg of agentMessages) {
+        outputMessages.push(convertAgentMessage(msg));
+      }
+      const durationMs = Date.now() - startTime;
+      return {
+        raw: {
+          messages: agentMessages,
+          systemPrompt,
+          model: this.config.model,
+          provider: this.config.provider
+        },
+        outputMessages,
+        durationMs
+      };
+    } finally {
+      unsubscribe();
+    }
+  }
+};
+function extractTextContent2(content) {
+  if (typeof content === "string") {
+    return content;
+  }
+  if (!Array.isArray(content)) {
+    return void 0;
+  }
+  const textParts = [];
+  for (const part of content) {
+    if (!part || typeof part !== "object") {
+      continue;
+    }
+    const p = part;
+    if (p.type === "text" && typeof p.text === "string") {
+      textParts.push(p.text);
+    }
+  }
+  return textParts.length > 0 ? textParts.join("\n") : void 0;
+}
+function convertAgentMessage(message) {
+  if (!message || typeof message !== "object") {
+    return { role: "unknown", content: String(message) };
+  }
+  const msg = message;
+  const role = typeof msg.role === "string" ? msg.role : "unknown";
+  const content = extractTextContent2(msg.content);
+  const toolCalls = extractToolCalls2(msg.content);
+  const timestamp = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
+  return {
+    role,
+    content,
+    toolCalls: toolCalls.length > 0 ? toolCalls : void 0,
+    timestamp
+  };
+}
+function extractToolCalls2(content) {
+  if (!Array.isArray(content)) {
+    return [];
+  }
+  const toolCalls = [];
+  for (const part of content) {
+    if (!part || typeof part !== "object") {
+      continue;
+    }
+    const p = part;
+    if (p.type === "tool_use" && typeof p.name === "string") {
+      toolCalls.push({
+        tool: p.name,
+        input: p.input,
+        id: typeof p.id === "string" ? p.id : void 0
+      });
+    }
+  }
+  return toolCalls;
+}
+
 // src/evaluation/providers/pi-coding-agent.ts
 var import_node_child_process4 = require("child_process");
 var import_node_crypto3 = require("crypto");
@@ -4787,8 +4946,8 @@ function convertPiMessage(message) {
   if (typeof role !== "string") {
     return void 0;
   }
-  const content = extractTextContent2(msg.content);
-  const toolCalls = extractToolCalls2(msg.content);
+  const content = extractTextContent3(msg.content);
+  const toolCalls = extractToolCalls3(msg.content);
   const timestamp = typeof msg.timestamp === "number" ? new Date(msg.timestamp).toISOString() : typeof msg.timestamp === "string" ? msg.timestamp : void 0;
   const metadata = {};
   if (msg.api) metadata.api = msg.api;
@@ -4804,7 +4963,7 @@ function convertPiMessage(message) {
     metadata: Object.keys(metadata).length > 0 ? metadata : void 0
   };
 }
-function extractTextContent2(content) {
+function extractTextContent3(content) {
   if (typeof content === "string") {
     return content;
   }
@@ -4823,7 +4982,7 @@ function extractTextContent2(content) {
   }
   return textParts.length > 0 ? textParts.join("\n") : void 0;
 }
-function extractToolCalls2(content) {
+function extractToolCalls3(content) {
   if (!Array.isArray(content)) {
     return [];
   }
@@ -5227,6 +5386,15 @@ function resolveTargetDefinition(definition, env = process.env, evalFilePath) {
         providerBatching,
         config: resolvePiCodingAgentConfig(parsed, env)
       };
+    case "pi-agent-sdk":
+      return {
+        kind: "pi-agent-sdk",
+        name: parsed.name,
+        judgeTarget: parsed.judge_target,
+        workers: parsed.workers,
+        providerBatching,
+        config: resolvePiAgentSdkConfig(parsed, env)
+      };
     case "claude-code":
       return {
         kind: "claude-code",
@@ -5448,6 +5616,39 @@ function resolvePiCodingAgentConfig(target, env) {
     systemPrompt
   };
 }
+function resolvePiAgentSdkConfig(target, env) {
+  const providerSource = target.pi_provider ?? target.piProvider ?? target.llm_provider;
+  const modelSource = target.model ?? target.pi_model ?? target.piModel;
+  const apiKeySource = target.api_key ?? target.apiKey;
+  const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
+  const systemPromptSource = target.system_prompt ?? target.systemPrompt;
+  const provider = resolveOptionalString(
+    providerSource,
+    env,
+    `${target.name} pi-agent-sdk provider`,
+    {
+      allowLiteral: true,
+      optionalEnv: true
+    }
+  );
+  const model = resolveOptionalString(modelSource, env, `${target.name} pi-agent-sdk model`, {
+    allowLiteral: true,
+    optionalEnv: true
+  });
+  const apiKey = resolveOptionalString(apiKeySource, env, `${target.name} pi-agent-sdk api key`, {
+    allowLiteral: false,
+    optionalEnv: true
+  });
+  const timeoutMs = resolveTimeoutMs(timeoutSource, `${target.name} pi-agent-sdk timeout`);
+  const systemPrompt = typeof systemPromptSource === "string" && systemPromptSource.trim().length > 0 ? systemPromptSource.trim() : void 0;
+  return {
+    provider,
+    model,
+    apiKey,
+    timeoutMs,
+    systemPrompt
+  };
+}
 function resolveClaudeCodeConfig(target, env) {
   const executableSource = target.executable ?? target.command ?? target.binary;
   const modelSource = target.model;
@@ -6106,6 +6307,8 @@ function createProvider(target) {
       return new CodexProvider(target.name, target.config);
     case "pi-coding-agent":
       return new PiCodingAgentProvider(target.name, target.config);
+    case "pi-agent-sdk":
+      return new PiAgentSdkProvider(target.name, target.config);
     case "claude-code":
       return new ClaudeCodeProvider(target.name, target.config);
     case "mock":
@@ -6273,12 +6476,6 @@ function toSnakeCase(str) {
   }
   return str.replace(/[A-Z]/g, (letter) => `_${letter.toLowerCase()}`);
 }
-function toCamelCase(str) {
-  if (/^[A-Z]/.test(str)) {
-    return str;
-  }
-  return str.replace(/_([a-z0-9])/g, (_, letter) => letter.toUpperCase());
-}
 function toSnakeCaseDeep(obj) {
   if (obj === null || obj === void 0) {
     return obj;
@@ -6296,23 +6493,6 @@ function toSnakeCaseDeep(obj) {
   }
   return obj;
 }
-function toCamelCaseDeep(obj) {
-  if (obj === null || obj === void 0) {
-    return obj;
-  }
-  if (Array.isArray(obj)) {
-    return obj.map((item) => toCamelCaseDeep(item));
-  }
-  if (typeof obj === "object") {
-    const result = {};
-    for (const [key, value] of Object.entries(obj)) {
-      const camelKey = toCamelCase(key);
-      result[camelKey] = toCamelCaseDeep(value);
-    }
-    return result;
-  }
-  return obj;
-}
 
 // src/evaluation/providers/types.ts
 var AGENT_PROVIDER_KINDS = [
@@ -9084,17 +9264,6 @@ function buildPrompt(expectedOutcome, question, referenceAnswer) {
   return parts.join("\n");
 }
 
-// src/evaluation/code-judge-sdk.ts
-var import_node_fs7 = require("fs");
-function parseCodeJudgePayload(payload) {
-  const parsed = JSON.parse(payload);
-  return toCamelCaseDeep(parsed);
-}
-function readCodeJudgePayload() {
-  const stdin = (0, import_node_fs7.readFileSync)(0, "utf8");
-  return parseCodeJudgePayload(stdin);
-}
-
 // src/index.ts
 function createAgentKernel() {
   return { status: "stub" };
@@ -9138,8 +9307,6 @@ function createAgentKernel() {
   loadEvalCases,
   mergeExecutionMetrics,
   normalizeLineEndings,
-  parseCodeJudgePayload,
-  readCodeJudgePayload,
   readJsonFile,
   readTargetDefinitions,
   readTestSuiteMetadata,