@agentv/core 4.6.0 → 4.6.1

package/dist/evaluation/validation/index.js

@@ -9,7 +9,7 @@ import {
   isEvaluatorKind,
   loadCasesFromFile,
   resolveFileReference
-} from "../../chunk-AIQ5FO4G.js";
+} from "../../chunk-ZK4GG7PR.js";
 
 // src/evaluation/validation/file-type.ts
 import { readFile } from "node:fs/promises";

package/dist/index.cjs

@@ -1561,6 +1561,7 @@ __export(index_exports, {
   readTranscriptFile: () => readTranscriptFile,
   removeProject: () => removeProject,
   resolveAndCreateProvider: () => resolveAndCreateProvider,
+  resolveDelegatedTargetDefinition: () => resolveDelegatedTargetDefinition,
   resolveFileReference: () => resolveFileReference3,
   resolveTargetDefinition: () => resolveTargetDefinition,
   resolveWorkspaceTemplate: () => resolveWorkspaceTemplate,
@@ -7169,15 +7170,16 @@ var CliProvider = class {
       outputFilePath
     );
     const renderedCommand = renderTemplate(this.config.command, templateValues);
+    const effectiveCwd = requests[0]?.cwd ?? this.config.cwd;
     if (this.verbose) {
       console.log(
-        `[cli-provider:${this.targetName}] (batch size=${requests.length}) cwd=${this.config.cwd ?? ""} command=${renderedCommand}`
+        `[cli-provider:${this.targetName}] (batch size=${requests.length}) cwd=${effectiveCwd ?? ""} command=${renderedCommand}`
       );
     }
     try {
       const startTime = Date.now();
       const result = await this.runCommand(renderedCommand, {
-        cwd: this.config.cwd,
+        cwd: effectiveCwd,
         env: process.env,
         timeoutMs: this.config.timeoutMs,
         signal: controller.signal
@@ -7210,7 +7212,7 @@ var CliProvider = class {
               command: renderedCommand,
               stderr: result.stderr,
               exitCode: result.exitCode ?? 0,
-              cwd: this.config.cwd,
+              cwd: effectiveCwd,
               outputFile: outputFilePath
             }
           };
@@ -7228,7 +7230,7 @@ var CliProvider = class {
               command: renderedCommand,
               stderr: result.stderr,
               exitCode: result.exitCode ?? 0,
-              cwd: this.config.cwd,
+              cwd: effectiveCwd,
               outputFile: outputFilePath,
               error: errorMessage
             }
@@ -7243,7 +7245,7 @@ var CliProvider = class {
             command: renderedCommand,
             stderr: result.stderr,
             exitCode: result.exitCode ?? 0,
-            cwd: this.config.cwd,
+            cwd: effectiveCwd,
             outputFile: outputFilePath,
             recordId: evalCaseId
           }
@@ -9267,6 +9269,60 @@ function subscribeToPiLogEntries(listener) {
   };
 }
 
+// src/evaluation/providers/pi-provider-aliases.ts
+init_cjs_shims();
+var SUBPROVIDER_ALIASES = {
+  azure: "azure-openai-responses"
+};
+var SUBPROVIDER_ALIASES_WITH_BASE_URL = {
+  // Azure v1 endpoints are OpenAI-compatible; use the standard client
+  // to avoid AzureOpenAI adding api-version query params.
+  azure: "openai-responses"
+};
+var ENV_KEY_MAP = {
+  google: "GEMINI_API_KEY",
+  gemini: "GEMINI_API_KEY",
+  anthropic: "ANTHROPIC_API_KEY",
+  openai: "OPENAI_API_KEY",
+  groq: "GROQ_API_KEY",
+  xai: "XAI_API_KEY",
+  openrouter: "OPENROUTER_API_KEY",
+  azure: "AZURE_OPENAI_API_KEY"
+};
+var ENV_BASE_URL_MAP = {
+  openai: "OPENAI_BASE_URL",
+  azure: "AZURE_OPENAI_BASE_URL",
+  openrouter: "OPENROUTER_BASE_URL"
+};
+function resolveSubprovider(name, hasBaseUrl = false) {
+  const lower = name.toLowerCase();
+  if (hasBaseUrl) {
+    const alias = SUBPROVIDER_ALIASES_WITH_BASE_URL[lower];
+    if (alias) return alias;
+  }
+  return SUBPROVIDER_ALIASES[lower] ?? name;
+}
+function resolveCliProvider(name) {
+  const lower = name.toLowerCase();
+  if (lower === "azure") return "azure-openai-responses";
+  return name;
+}
+function resolveEnvKeyName(provider, hasBaseUrl = false) {
+  const lower = provider.toLowerCase();
+  if (hasBaseUrl && lower === "azure") return "OPENAI_API_KEY";
+  return ENV_KEY_MAP[lower];
+}
+function resolveEnvBaseUrlName(provider, hasBaseUrl = false) {
+  const lower = provider.toLowerCase();
+  if (hasBaseUrl && lower === "azure") return "OPENAI_BASE_URL";
+  return ENV_BASE_URL_MAP[lower];
+}
+function extractAzureResourceName(baseUrl) {
+  const urlMatch = baseUrl.match(/^https?:\/\/([^./]+)/);
+  if (urlMatch) return urlMatch[1];
+  return baseUrl;
+}
+
 // src/evaluation/providers/pi-utils.ts
 init_cjs_shims();
 function extractPiTextContent(content) {
@@ -9426,12 +9482,12 @@ var PiCliProvider = class {
   buildPiArgs(prompt, inputFiles) {
     const args = [];
     if (this.config.subprovider) {
-      args.push("--provider", this.config.subprovider);
+      args.push("--provider", resolveCliProvider(this.config.subprovider));
     }
     if (this.config.model) {
       args.push("--model", this.config.model);
     }
-    if (this.config.apiKey) {
+    if (this.config.apiKey && this.config.subprovider?.toLowerCase() !== "azure") {
       args.push("--api-key", this.config.apiKey);
     }
     args.push("--mode", "json");
@@ -9483,35 +9539,35 @@ ${prompt}` : prompt;
   }
   buildEnv() {
     const env = { ...process.env };
-    if (this.config.apiKey) {
-      const provider = this.config.subprovider?.toLowerCase() ?? "google";
-      const ENV_KEY_MAP = {
-        google: "GEMINI_API_KEY",
-        gemini: "GEMINI_API_KEY",
-        anthropic: "ANTHROPIC_API_KEY",
-        openai: "OPENAI_API_KEY",
-        groq: "GROQ_API_KEY",
-        xai: "XAI_API_KEY",
-        openrouter: "OPENROUTER_API_KEY"
-      };
-      const envKey = ENV_KEY_MAP[provider];
-      if (envKey) {
-        env[envKey] = this.config.apiKey;
+    const provider = this.config.subprovider?.toLowerCase() ?? "google";
+    if (provider === "azure") {
+      if (this.config.apiKey) {
+        env.AZURE_OPENAI_API_KEY = this.config.apiKey;
+      }
+      if (this.config.baseUrl) {
+        env.AZURE_OPENAI_RESOURCE_NAME = extractAzureResourceName(this.config.baseUrl);
+      }
+    } else {
+      if (this.config.apiKey) {
+        const envKey = resolveEnvKeyName(provider);
+        if (envKey) {
+          env[envKey] = this.config.apiKey;
+        }
       }
     }
     if (this.config.subprovider) {
-      const provider = this.config.subprovider.toLowerCase();
+      const resolvedProvider = resolveCliProvider(this.config.subprovider);
       const PROVIDER_OWN_PREFIXES = {
         openrouter: ["OPENROUTER_"],
         anthropic: ["ANTHROPIC_"],
         openai: ["OPENAI_"],
-        azure: ["AZURE_OPENAI_"],
+        "azure-openai-responses": ["AZURE_OPENAI_"],
         google: ["GEMINI_", "GOOGLE_GENERATIVE_AI_"],
         gemini: ["GEMINI_", "GOOGLE_GENERATIVE_AI_"],
         groq: ["GROQ_"],
         xai: ["XAI_"]
       };
-      const ownPrefixes = PROVIDER_OWN_PREFIXES[provider] ?? [];
+      const ownPrefixes = PROVIDER_OWN_PREFIXES[resolvedProvider] ?? [];
       const allOtherPrefixes = Object.entries(PROVIDER_OWN_PREFIXES).filter(([key]) => key !== provider).flatMap(([, prefixes]) => prefixes);
       for (const key of Object.keys(env)) {
         if (allOtherPrefixes.some((prefix) => key.startsWith(prefix)) && !ownPrefixes.some((prefix) => key.startsWith(prefix))) {
@@ -9802,6 +9858,24 @@ function extractMessages(events) {
       }
     }
   }
+  if (messages) {
+    for (let i = messages.length - 1; i >= 0; i--) {
+      if (messages[i].role === "assistant" && !messages[i].content) {
+        for (let j = events.length - 1; j >= 0; j--) {
+          const evt = events[j];
+          if (!evt || evt.type !== "message_end") continue;
+          const msg = evt.message;
+          if (msg?.role !== "assistant") continue;
+          const text = extractPiTextContent(msg.content);
+          if (text) {
+            messages[i] = { ...messages[i], content: text };
+            break;
+          }
+        }
+        break;
+      }
+    }
+  }
   const eventToolCalls = extractToolCallsFromEvents(events);
   if (eventToolCalls.length > 0) {
     injectEventToolCalls(messages, eventToolCalls);
@@ -9986,17 +10060,43 @@ function formatTimeoutSuffix3(timeoutMs) {
   if (!timeoutMs || timeoutMs <= 0) return "";
   return ` after ${Math.ceil(timeoutMs / 1e3)}s`;
 }
+function resolveWindowsCmd(executable) {
+  if (process.platform !== "win32") return [executable, []];
+  const lower = executable.toLowerCase();
+  if (lower.endsWith(".js") || lower.endsWith(".exe")) return [executable, []];
+  let fullPath;
+  try {
+    fullPath = (0, import_node_child_process4.execSync)(`where ${executable}`, { encoding: "utf-8" }).trim().split(/\r?\n/)[0].trim();
+  } catch {
+    return [executable, []];
+  }
+  const cmdPath = fullPath.endsWith(".cmd") ? fullPath : `${fullPath}.cmd`;
+  try {
+    const content = (0, import_node_fs9.readFileSync)(cmdPath, "utf-8");
+    const match = content.match(/"?%_prog%"?\s+"([^"]+\.js)"/);
+    if (match) {
+      const dp0 = import_node_path21.default.dirname(import_node_path21.default.resolve(cmdPath));
+      const scriptPath = match[1].replace(/%dp0%[/\\]?/gi, `${dp0}${import_node_path21.default.sep}`);
+      try {
+        (0, import_node_fs9.accessSync)(scriptPath);
+        return ["node", [scriptPath]];
+      } catch {
+      }
+    }
+  } catch {
+  }
+  return [executable, []];
+}
 async function defaultPiRunner(options) {
   return await new Promise((resolve, reject) => {
     const parts = options.executable.split(/\s+/);
-    const executable = parts[0];
-    const executableArgs = parts.slice(1);
+    const [resolvedExe, prefixArgs] = resolveWindowsCmd(parts[0]);
+    const executableArgs = [...prefixArgs, ...parts.slice(1)];
     const allArgs = [...executableArgs, ...options.args];
-    const child = (0, import_node_child_process4.spawn)(executable, allArgs, {
+    const child = (0, import_node_child_process4.spawn)(resolvedExe, allArgs, {
       cwd: options.cwd,
       env: options.env,
-      stdio: ["pipe", "pipe", "pipe"],
-      shell: false
+      stdio: ["pipe", "pipe", "pipe"]
     });
     let stdout = "";
     let stderr = "";
@@ -10143,7 +10243,9 @@ async function loadSdkModules() {
     codingTools: piSdk.codingTools,
     toolMap,
     SessionManager: piSdk.SessionManager,
-    getModel: piAi.getModel
+    getModel: piAi.getModel,
+    // biome-ignore lint/suspicious/noExplicitAny: registerBuiltInApiProviders exists at runtime but not in type defs
+    registerBuiltInApiProviders: piAi.registerBuiltInApiProviders
   };
 }
 var PiCodingAgentProvider = class {
@@ -10165,17 +10267,31 @@ var PiCodingAgentProvider = class {
     const startTime = (/* @__PURE__ */ new Date()).toISOString();
     const startMs = Date.now();
     const sdk = await loadSdkModules();
+    sdk.registerBuiltInApiProviders();
     const logger = await this.createStreamLogger(request).catch(() => void 0);
     try {
       const cwd = this.resolveCwd(request.cwd);
-      const providerName = this.config.subprovider ?? "google";
+      const rawProvider = this.config.subprovider ?? "google";
+      const hasBaseUrl = !!this.config.baseUrl;
+      const providerName = resolveSubprovider(rawProvider, hasBaseUrl);
       const modelId = this.config.model ?? "gemini-2.5-flash";
-      this.setApiKeyEnv(providerName);
-      const model = sdk.getModel(providerName, modelId);
+      this.setApiKeyEnv(rawProvider, hasBaseUrl);
+      this.setBaseUrlEnv(rawProvider, hasBaseUrl);
+      let model = sdk.getModel(providerName, modelId);
       if (!model) {
-        throw new Error(
-          `pi-coding-agent: getModel('${providerName}', '${modelId}') returned undefined. The model '${modelId}' is not registered for provider '${providerName}' in pi-ai. Check that subprovider and model are correct in your target config.`
-        );
+        const envProvider = providerName.replace(/-responses$/, "");
+        model = {
+          id: modelId,
+          name: modelId,
+          api: providerName,
+          provider: envProvider,
+          baseUrl: this.config.baseUrl ?? "",
+          reasoning: false,
+          input: ["text"],
+          cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+          contextWindow: 128e3,
+          maxTokens: 16384
+        };
       }
       const tools = this.resolveTools(sdk);
       const { session } = await sdk.createAgentSession({
@@ -10328,22 +10444,21 @@ ${fileList}`;
     }
   }
   /** Maps config apiKey to the provider-specific env var the SDK reads. */
-  setApiKeyEnv(providerName) {
+  setApiKeyEnv(providerName, hasBaseUrl = false) {
     if (!this.config.apiKey) return;
-    const ENV_KEY_MAP = {
-      google: "GEMINI_API_KEY",
-      gemini: "GEMINI_API_KEY",
-      anthropic: "ANTHROPIC_API_KEY",
-      openai: "OPENAI_API_KEY",
-      groq: "GROQ_API_KEY",
-      xai: "XAI_API_KEY",
-      openrouter: "OPENROUTER_API_KEY"
-    };
-    const envKey = ENV_KEY_MAP[providerName.toLowerCase()];
+    const envKey = resolveEnvKeyName(providerName, hasBaseUrl);
     if (envKey) {
       process.env[envKey] = this.config.apiKey;
     }
   }
+  /** Maps config baseUrl to the provider-specific env var the SDK reads. */
+  setBaseUrlEnv(providerName, hasBaseUrl = false) {
+    if (!this.config.baseUrl) return;
+    const envKey = resolveEnvBaseUrlName(providerName, hasBaseUrl);
+    if (envKey) {
+      process.env[envKey] = this.config.baseUrl;
+    }
+  }
   resolveCwd(cwdOverride) {
     if (cwdOverride) {
       return import_node_path22.default.resolve(cwdOverride);
@@ -10776,6 +10891,7 @@ var COMMON_TARGET_SETTINGS = [
   "fallback_targets",
   "fallbackTargets"
 ];
+var USE_TARGET_ENV_PATTERN = /^\$\{\{\s*([A-Z0-9_]+)\s*\}\}$/i;
 var BASE_TARGET_SCHEMA = import_zod3.z.object({
   name: import_zod3.z.string().min(1, "target name is required"),
   provider: import_zod3.z.string().optional(),
@@ -10835,6 +10951,52 @@ function resolveRetryConfig(target) {
     retryableStatusCodes
   };
 }
+function resolveDelegatedTargetDefinition(name, definitions, env = process.env) {
+  let definition = definitions.get(name);
+  if (!definition) {
+    return void 0;
+  }
+  const visited = [definition.name];
+  for (let depth = 0; depth < 10; depth++) {
+    const rawUseTarget = typeof definition.use_target === "string" ? definition.use_target.trim() : void 0;
+    if (!rawUseTarget) {
+      return definition;
+    }
+    const envMatch = rawUseTarget.match(USE_TARGET_ENV_PATTERN);
+    const envVarName = envMatch?.[1];
+    const resolvedName = envVarName ? env[envVarName]?.trim() ?? "" : rawUseTarget;
+    if (resolvedName.length === 0) {
+      if (envVarName) {
+        throw new Error(
+          `Target "${definition.name}" uses use_target: \${{ ${envVarName} }}, but ${envVarName} is not set. Set ${envVarName} to the name of a concrete target (for example, "azure") before running the eval.`
+        );
+      }
+      throw new Error(
+        `Target "${definition.name}" has an empty use_target value. Point it at a concrete target name before running the eval.`
+      );
+    }
+    const next = definitions.get(resolvedName);
+    if (!next) {
+      if (envVarName) {
+        throw new Error(
+          `Target "${definition.name}" uses use_target: \${{ ${envVarName} }}, which resolved to "${resolvedName}", but no target named "${resolvedName}" exists.`
+        );
+      }
+      throw new Error(
+        `Target "${definition.name}" uses use_target: "${resolvedName}", but no target named "${resolvedName}" exists.`
+      );
+    }
+    if (visited.includes(next.name)) {
+      const chain = [...visited, next.name].join(" -> ");
+      throw new Error(`Circular use_target reference detected: ${chain}`);
+    }
+    definition = next;
+    visited.push(definition.name);
+  }
+  throw new Error(
+    `Target "${name}" exceeded the maximum use_target resolution depth (10). Check for a delegation loop or overly deep alias chain.`
+  );
+}
 function resolveTargetDefinition(definition, env = process.env, evalFilePath) {
   const parsed = BASE_TARGET_SCHEMA.parse(definition);
   if (parsed.workspace_template !== void 0 || parsed.workspaceTemplate !== void 0) {
@@ -11374,6 +11536,11 @@ function resolvePiCodingAgentConfig(target, env, evalFilePath) {
     allowLiteral: false,
     optionalEnv: true
   });
+  const baseUrlSource = target.base_url ?? target.baseUrl ?? target.endpoint;
+  const baseUrl = resolveOptionalString(baseUrlSource, env, `${target.name} pi base url`, {
+    allowLiteral: true,
+    optionalEnv: true
+  });
   const tools = resolveOptionalString(toolsSource, env, `${target.name} pi tools`, {
     allowLiteral: true,
     optionalEnv: true
@@ -11414,6 +11581,7 @@ function resolvePiCodingAgentConfig(target, env, evalFilePath) {
     subprovider,
     model,
     apiKey,
+    baseUrl,
     tools,
     thinking,
     cwd,
@@ -11455,6 +11623,11 @@ function resolvePiCliConfig(target, env, evalFilePath) {
     allowLiteral: false,
     optionalEnv: true
   });
+  const baseUrlSource = target.base_url ?? target.baseUrl ?? target.endpoint;
+  const baseUrl = resolveOptionalString(baseUrlSource, env, `${target.name} pi-cli base url`, {
+    allowLiteral: true,
+    optionalEnv: true
+  });
   const tools = resolveOptionalString(toolsSource, env, `${target.name} pi-cli tools`, {
     allowLiteral: true,
     optionalEnv: true
@@ -11493,6 +11666,7 @@ function resolvePiCliConfig(target, env, evalFilePath) {
     subprovider,
     model,
     apiKey,
+    baseUrl,
     tools,
     thinking,
     args,
@@ -18859,20 +19033,10 @@ async function runEvaluation(options) {
     if (resolvedTargetsByName.has(name)) {
       return resolvedTargetsByName.get(name);
     }
-    let definition = targetDefinitions.get(name);
+    const definition = resolveDelegatedTargetDefinition(name, targetDefinitions, envLookup);
     if (!definition) {
       return void 0;
     }
-    for (let depth = 0; depth < 5; depth++) {
-      const useTarget = definition.use_target;
-      if (typeof useTarget !== "string" || useTarget.trim().length === 0) break;
-      const envMatch = useTarget.trim().match(/^\$\{\{\s*([A-Z0-9_]+)\s*\}\}$/i);
-      const resolvedName = envMatch ? envLookup[envMatch[1]] ?? "" : useTarget.trim();
-      if (resolvedName.length === 0) break;
-      const next = targetDefinitions.get(resolvedName);
-      if (!next) break;
-      definition = next;
-    }
     const resolved = resolveTargetDefinition(definition, envLookup, evalFilePath);
     resolvedTargetsByName.set(name, resolved);
     return resolved;
@@ -20960,7 +21124,7 @@ async function discoverDefaultTarget(repoRoot) {
   return null;
 }
 async function loadEnvHierarchy(repoRoot, startPath) {
-  const { readFileSync: readFileSync3 } = await import("fs");
+  const { readFileSync: readFileSync4 } = await import("fs");
   const chain = buildDirectoryChain2(startPath, repoRoot);
   const envFiles = [];
   for (const dir of chain) {
@@ -20969,7 +21133,7 @@ async function loadEnvHierarchy(repoRoot, startPath) {
   }
   for (let i = 0; i < envFiles.length; i++) {
     try {
-      const content = readFileSync3(envFiles[i], "utf8");
+      const content = readFileSync4(envFiles[i], "utf8");
       for (const line of content.split("\n")) {
         const trimmed = line.trim();
         if (!trimmed || trimmed.startsWith("#")) continue;
@@ -22259,6 +22423,7 @@ function createAgentKernel() {
   readTranscriptFile,
   removeProject,
   resolveAndCreateProvider,
+  resolveDelegatedTargetDefinition,
   resolveFileReference,
   resolveTargetDefinition,
   resolveWorkspaceTemplate,