agentv 3.10.3 → 3.11.0

package/dist/{chunk-JUQCB3ZW.js → chunk-EZGWZVVK.js}

@@ -301,7 +301,7 @@ var require_dist = __commonJS({
   }
 });
 
-// ../../packages/core/dist/chunk-VCFYWLFV.js
+// ../../packages/core/dist/chunk-AVTN5AB7.js
 import { constants } from "node:fs";
 import { access, readFile } from "node:fs/promises";
 import path from "node:path";
@@ -419,7 +419,7 @@ __export(external_exports2, {
   void: () => voidType
 });
 
-// ../../packages/core/dist/chunk-VCFYWLFV.js
+// ../../packages/core/dist/chunk-AVTN5AB7.js
 import { readFile as readFile2 } from "node:fs/promises";
 import path3 from "node:path";
 import fg from "fast-glob";
@@ -1363,7 +1363,7 @@ function normalizeCopilotLogFormat(value) {
 }
 function resolvePiCodingAgentConfig(target, env, evalFilePath) {
   const executableSource = target.executable ?? target.command ?? target.binary;
-  const providerSource = target.pi_provider ?? target.piProvider ?? target.llm_provider;
+  const subproviderSource = target.subprovider;
   const modelSource = target.model ?? target.pi_model ?? target.piModel;
   const apiKeySource = target.api_key ?? target.apiKey;
   const toolsSource = target.tools ?? target.pi_tools ?? target.piTools;
@@ -1379,10 +1379,15 @@ function resolvePiCodingAgentConfig(target, env, evalFilePath) {
     allowLiteral: true,
     optionalEnv: true
   }) ?? "pi";
-  const provider = resolveOptionalString(providerSource, env, `${target.name} pi provider`, {
-    allowLiteral: true,
-    optionalEnv: true
-  });
+  const subprovider = resolveOptionalString(
+    subproviderSource,
+    env,
+    `${target.name} pi subprovider`,
+    {
+      allowLiteral: true,
+      optionalEnv: true
+    }
+  );
   const model = resolveOptionalString(modelSource, env, `${target.name} pi model`, {
     allowLiteral: true,
     optionalEnv: true
@@ -1430,7 +1435,7 @@ function resolvePiCodingAgentConfig(target, env, evalFilePath) {
   const systemPrompt = typeof systemPromptSource === "string" && systemPromptSource.trim().length > 0 ? systemPromptSource.trim() : void 0;
   return {
     executable,
-    provider,
+    subprovider,
     model,
     apiKey,
     tools,
@@ -1445,15 +1450,15 @@ function resolvePiCodingAgentConfig(target, env, evalFilePath) {
   };
 }
 function resolvePiAgentSdkConfig(target, env) {
-  const providerSource = target.pi_provider ?? target.piProvider ?? target.llm_provider;
+  const subproviderSource = target.subprovider;
   const modelSource = target.model ?? target.pi_model ?? target.piModel;
   const apiKeySource = target.api_key ?? target.apiKey;
   const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
   const systemPromptSource = target.system_prompt ?? target.systemPrompt;
-  const provider = resolveOptionalString(
-    providerSource,
+  const subprovider = resolveOptionalString(
+    subproviderSource,
     env,
-    `${target.name} pi-agent-sdk provider`,
+    `${target.name} pi-agent-sdk subprovider`,
     {
       allowLiteral: true,
       optionalEnv: true
@@ -1470,7 +1475,7 @@ function resolvePiAgentSdkConfig(target, env) {
   const timeoutMs = resolveTimeoutMs(timeoutSource, `${target.name} pi-agent-sdk timeout`);
   const systemPrompt = typeof systemPromptSource === "string" && systemPromptSource.trim().length > 0 ? systemPromptSource.trim() : void 0;
   return {
-    provider,
+    subprovider,
     model,
     apiKey,
     timeoutMs,
@@ -2039,7 +2044,7 @@ import path8 from "node:path";
 import { parse as parse3 } from "yaml";
 import { createOpenAI } from "@ai-sdk/openai";
 
-// ../../node_modules/.bun/@openrouter+ai-sdk-provider@2.3.1+3ab978b6804fd9e7/node_modules/@openrouter/ai-sdk-provider/dist/index.mjs
+// ../../node_modules/.bun/@openrouter+ai-sdk-provider@2.3.3+3ab978b6804fd9e7/node_modules/@openrouter/ai-sdk-provider/dist/index.mjs
 var __defProp = Object.defineProperty;
 var __defProps = Object.defineProperties;
 var __getOwnPropDescs = Object.getOwnPropertyDescriptors;
@@ -4202,11 +4207,13 @@ function isDefinedOrNotNull(value) {
 var ReasoningFormat = /* @__PURE__ */ ((ReasoningFormat2) => {
   ReasoningFormat2["Unknown"] = "unknown";
   ReasoningFormat2["OpenAIResponsesV1"] = "openai-responses-v1";
+  ReasoningFormat2["AzureOpenAIResponsesV1"] = "azure-openai-responses-v1";
   ReasoningFormat2["XAIResponsesV1"] = "xai-responses-v1";
   ReasoningFormat2["AnthropicClaudeV1"] = "anthropic-claude-v1";
   ReasoningFormat2["GoogleGeminiV1"] = "google-gemini-v1";
   return ReasoningFormat2;
 })(ReasoningFormat || {});
+var DEFAULT_REASONING_FORMAT = "anthropic-claude-v1";
 var CommonReasoningDetailSchema = external_exports.object({
   id: external_exports.string().nullish(),
   format: external_exports.enum(ReasoningFormat).nullish(),
@@ -4360,7 +4367,11 @@ var OpenRouterProviderMetadataSchema = external_exports.object({
 }).catchall(external_exports.any());
 var OpenRouterProviderOptionsSchema = external_exports.object({
   openrouter: external_exports.object({
-    reasoning_details: external_exports.array(ReasoningDetailUnionSchema).optional(),
+    // Use ReasoningDetailArraySchema (with unknown fallback) instead of
+    // z.array(ReasoningDetailUnionSchema) so that a single malformed entry
+    // (e.g., a future format not yet in the enum) is individually dropped
+    // rather than causing the entire array to fail parsing.
+    reasoning_details: ReasoningDetailArraySchema.optional(),
     annotations: external_exports.array(FileAnnotationSchema).optional()
   }).optional()
 }).optional();
@@ -4758,8 +4769,24 @@ function convertToOpenRouterChatMessages(prompt) {
         const candidateReasoningDetails = messageReasoningDetails && Array.isArray(messageReasoningDetails) && messageReasoningDetails.length > 0 ? messageReasoningDetails : findFirstReasoningDetails(content);
         let finalReasoningDetails;
         if (candidateReasoningDetails && candidateReasoningDetails.length > 0) {
+          const validDetails = candidateReasoningDetails.filter((detail) => {
+            var _a173;
+            if (detail.type !== "reasoning.text") {
+              return true;
+            }
+            const format = (_a173 = detail.format) != null ? _a173 : DEFAULT_REASONING_FORMAT;
+            if (format !== "anthropic-claude-v1") {
+              return true;
+            }
+            return !!detail.signature;
+          });
+          if (validDetails.length < candidateReasoningDetails.length) {
+            console.warn(
+              "[openrouter] Some reasoning_details entries were removed because they were missing signatures. See https://github.com/OpenRouterTeam/ai-sdk-provider/issues/423 for more details."
+            );
+          }
           const uniqueDetails = [];
-          for (const detail of candidateReasoningDetails) {
+          for (const detail of validDetails) {
             if (reasoningDetailsTracker.upsert(detail)) {
               uniqueDetails.push(detail);
             }
@@ -4808,20 +4835,135 @@ function getToolResultContent(input) {
       return input.output.value;
     case "json":
     case "error-json":
-    case "content":
       return JSON.stringify(input.output.value);
+    case "content":
+      return mapToolResultContentParts(input.output.value);
     case "execution-denied":
       return (_a163 = input.output.reason) != null ? _a163 : "Tool execution denied";
   }
 }
+function mapToolResultContentParts(parts) {
+  return parts.map((part) => {
+    var _a163, _b162, _c;
+    switch (part.type) {
+      case "text":
+        return { type: "text", text: part.text };
+      case "image-data":
+        return {
+          type: "image_url",
+          image_url: {
+            url: buildFileDataUrl({
+              data: part.data,
+              mediaType: part.mediaType,
+              defaultMediaType: "image/jpeg"
+            })
+          }
+        };
+      case "image-url":
+        return {
+          type: "image_url",
+          image_url: { url: part.url }
+        };
+      case "file-data": {
+        const dataUrl = buildFileDataUrl({
+          data: part.data,
+          mediaType: part.mediaType,
+          defaultMediaType: "application/octet-stream"
+        });
+        if ((_a163 = part.mediaType) == null ? void 0 : _a163.startsWith("image/")) {
+          return {
+            type: "image_url",
+            image_url: { url: dataUrl }
+          };
+        }
+        if ((_b162 = part.mediaType) == null ? void 0 : _b162.startsWith("audio/")) {
+          const rawFormat = part.mediaType.replace("audio/", "");
+          const format = MIME_TO_FORMAT[rawFormat];
+          if (format !== void 0) {
+            return {
+              type: "input_audio",
+              input_audio: {
+                data: getBase64FromDataUrl(dataUrl),
+                format
+              }
+            };
+          }
+        }
+        return {
+          type: "file",
+          file: {
+            filename: (_c = part.filename) != null ? _c : "",
+            file_data: dataUrl
+          }
+        };
+      }
+      case "file-url": {
+        if (looksLikeImageUrl(part.url)) {
+          return {
+            type: "image_url",
+            image_url: { url: part.url }
+          };
+        }
+        return {
+          type: "file",
+          file: {
+            filename: filenameFromUrl(part.url),
+            file_data: part.url
+          }
+        };
+      }
+      case "file-id":
+      case "image-file-id":
+      case "custom":
+        return { type: "text", text: JSON.stringify(part) };
+      default: {
+        const _exhaustiveCheck = part;
+        return { type: "text", text: JSON.stringify(_exhaustiveCheck) };
+      }
+    }
+  });
+}
+var IMAGE_EXTENSIONS = /* @__PURE__ */ new Set([
+  "jpg",
+  "jpeg",
+  "png",
+  "gif",
+  "webp",
+  "svg",
+  "bmp",
+  "ico",
+  "tif",
+  "tiff",
+  "avif"
+]);
+function looksLikeImageUrl(url) {
+  var _a163;
+  try {
+    const pathname = new URL(url).pathname;
+    const ext = (_a163 = pathname.split(".").pop()) == null ? void 0 : _a163.toLowerCase();
+    return ext !== void 0 && IMAGE_EXTENSIONS.has(ext);
+  } catch (e) {
+    return false;
+  }
+}
+function filenameFromUrl(url) {
+  try {
+    const pathname = new URL(url).pathname;
+    const last = pathname.split("/").pop();
+    return (last == null ? void 0 : last.includes(".")) ? last : "";
+  } catch (e) {
+    return "";
+  }
+}
 function findFirstReasoningDetails(content) {
-  var _a163, _b162, _c;
+  var _a163, _b162, _c, _d;
   for (const part of content) {
     if (part.type === "tool-call") {
-      const openrouter2 = (_a163 = part.providerOptions) == null ? void 0 : _a163.openrouter;
-      const details = openrouter2 == null ? void 0 : openrouter2.reasoning_details;
-      if (Array.isArray(details) && details.length > 0) {
-        return details;
+      const parsed = OpenRouterProviderOptionsSchema.safeParse(
+        part.providerOptions
+      );
+      if (parsed.success && ((_b162 = (_a163 = parsed.data) == null ? void 0 : _a163.openrouter) == null ? void 0 : _b162.reasoning_details) && parsed.data.openrouter.reasoning_details.length > 0) {
+        return parsed.data.openrouter.reasoning_details;
       }
     }
   }
@@ -4830,7 +4972,7 @@ function findFirstReasoningDetails(content) {
       const parsed = OpenRouterProviderOptionsSchema.safeParse(
         part.providerOptions
       );
-      if (parsed.success && ((_c = (_b162 = parsed.data) == null ? void 0 : _b162.openrouter) == null ? void 0 : _c.reasoning_details) && parsed.data.openrouter.reasoning_details.length > 0) {
+      if (parsed.success && ((_d = (_c = parsed.data) == null ? void 0 : _c.openrouter) == null ? void 0 : _d.reasoning_details) && parsed.data.openrouter.reasoning_details.length > 0) {
         return parsed.data.openrouter.reasoning_details;
       }
     }
@@ -6490,7 +6632,7 @@ function withUserAgentSuffix22(headers, ...userAgentSuffixParts) {
     "user-agent": userAgent
   });
 }
-var VERSION2 = false ? "0.0.0-test" : "2.3.1";
+var VERSION2 = false ? "0.0.0-test" : "2.3.3";
 function createOpenRouter(options = {}) {
   var _a163, _b162, _c;
   const baseURL = (_b162 = withoutTrailingSlash2((_a163 = options.baseURL) != null ? _a163 : options.baseUrl)) != null ? _b162 : "https://openrouter.ai/api/v1";
@@ -14227,6 +14369,62 @@ function mergeExecutionMetrics(computed, metrics) {
     endTime: metrics.endTime ?? computed.endTime
   };
 }
+function flattenInputMessages(messages) {
+  return messages.flatMap((message) => extractContentSegments(message.content));
+}
+function collectResolvedInputFilePaths(messages) {
+  const filePaths = [];
+  for (const message of messages) {
+    if (!Array.isArray(message.content)) {
+      continue;
+    }
+    for (const segment of message.content) {
+      if (isJsonObject(segment) && segment.type === "file" && typeof segment.resolvedPath === "string") {
+        filePaths.push(segment.resolvedPath);
+      }
+    }
+  }
+  return filePaths;
+}
+function extractContentSegments(content) {
+  if (typeof content === "string") {
+    return content.trim().length > 0 ? [{ type: "text", value: content }] : [];
+  }
+  if (isJsonObject(content)) {
+    const rendered = JSON.stringify(content, null, 2);
+    return rendered.trim().length > 0 ? [{ type: "text", value: rendered }] : [];
+  }
+  if (!Array.isArray(content)) {
+    return [];
+  }
+  const segments = [];
+  for (const segment of content) {
+    if (!isJsonObject(segment)) {
+      continue;
+    }
+    segments.push(cloneJsonObject(segment));
+  }
+  return segments;
+}
+function cloneJsonObject(source) {
+  const entries = Object.entries(source).map(([key, value]) => [key, cloneJsonValue(value)]);
+  return Object.fromEntries(entries);
+}
+function cloneJsonValue(value) {
+  if (value === null) {
+    return null;
+  }
+  if (typeof value === "string" || typeof value === "number" || typeof value === "boolean") {
+    return value;
+  }
+  if (Array.isArray(value)) {
+    return value.map((item) => cloneJsonValue(item));
+  }
+  if (typeof value === "object") {
+    return cloneJsonObject(value);
+  }
+  return value;
+}
 var ANSI_RED = "\x1B[31m";
 var ANSI_RESET2 = "\x1B[0m";
 function logError(msg) {
@@ -14292,7 +14490,6 @@ function parseAgentSkillsEvals(parsed, source = "evals.json", baseDir) {
       id: String(id),
       question: prompt,
       input: [{ role: "user", content: prompt }],
-      input_segments: [{ type: "text", value: prompt }],
       expected_output: evalCase.expected_output ? [{ role: "assistant", content: evalCase.expected_output }] : [],
       reference_answer: evalCase.expected_output,
       file_paths: filePaths,
@@ -14414,7 +14611,7 @@ async function loadConfig(evalFilePath, repoRoot) {
     }
     try {
       const rawConfig = await readFile22(configPath, "utf8");
-      const parsed = parse(rawConfig);
+      const parsed = interpolateEnv(parse(rawConfig), process.env);
       if (!isJsonObject(parsed)) {
         logWarning(`Invalid .agentv/config.yaml format at ${configPath}`);
         continue;
@@ -14632,6 +14829,27 @@ function parseExecutionDefaults(raw, configPath) {
   } else if (otelFile !== void 0) {
     logWarning(`Invalid execution.otel_file in ${configPath}, expected non-empty string`);
   }
+  if (typeof obj.export_otel === "boolean") {
+    result.export_otel = obj.export_otel;
+  } else if (obj.export_otel !== void 0) {
+    logWarning(`Invalid execution.export_otel in ${configPath}, expected boolean`);
+  }
+  const otelBackend = obj.otel_backend;
+  if (typeof otelBackend === "string" && otelBackend.trim().length > 0) {
+    result.otel_backend = otelBackend.trim();
+  } else if (otelBackend !== void 0) {
+    logWarning(`Invalid execution.otel_backend in ${configPath}, expected non-empty string`);
+  }
+  if (typeof obj.otel_capture_content === "boolean") {
+    result.otel_capture_content = obj.otel_capture_content;
+  } else if (obj.otel_capture_content !== void 0) {
+    logWarning(`Invalid execution.otel_capture_content in ${configPath}, expected boolean`);
+  }
+  if (typeof obj.otel_group_turns === "boolean") {
+    result.otel_group_turns = obj.otel_group_turns;
+  } else if (obj.otel_group_turns !== void 0) {
+    logWarning(`Invalid execution.otel_group_turns in ${configPath}, expected boolean`);
+  }
   if (typeof obj.pool_workspaces === "boolean") {
     result.pool_workspaces = obj.pool_workspaces;
   } else if (obj.pool_workspaces !== void 0) {
@@ -16076,27 +16294,28 @@ var ANSI_YELLOW4 = "\x1B[33m";
 var ANSI_RESET5 = "\x1B[0m";
 async function processMessages(options) {
   const { messages, searchRoots, repoRootPath, textParts, messageType, verbose } = options;
-  const segments = [];
+  const processedMessages = [];
   for (const message of messages) {
     const content = message.content;
     if (typeof content === "string") {
-      segments.push({ type: "text", value: content });
       if (textParts) {
         textParts.push(content);
       }
+      processedMessages.push({ ...message, content });
       continue;
     }
     if (isJsonObject(content)) {
       const rendered = JSON.stringify(content, null, 2);
-      segments.push({ type: "text", value: rendered });
       if (textParts) {
         textParts.push(rendered);
       }
+      processedMessages.push({ ...message, content: cloneJsonObject(content) });
       continue;
     }
     if (!Array.isArray(content)) {
       continue;
     }
+    const processedContent = [];
     for (const rawSegment of content) {
       if (!isJsonObject(rawSegment)) {
         continue;
@@ -16119,8 +16338,8 @@ async function processMessages(options) {
         }
         try {
           const fileContent = (await readFile4(resolvedPath, "utf8")).replace(/\r\n/g, "\n");
-          segments.push({
-            type: "file",
+          processedContent.push({
+            ...cloneJsonObject(rawSegment),
             path: displayPath,
             text: fileContent,
             resolvedPath: path5.resolve(resolvedPath)
@@ -16137,37 +16356,19 @@ async function processMessages(options) {
         continue;
       }
       const clonedSegment = cloneJsonObject(rawSegment);
-      segments.push(clonedSegment);
+      processedContent.push(clonedSegment);
       const inlineValue = clonedSegment.value;
       if (typeof inlineValue === "string" && textParts) {
         textParts.push(inlineValue);
       }
     }
+    processedMessages.push({ ...message, content: processedContent });
   }
-  return segments;
+  return processedMessages;
 }
 function asString3(value) {
   return typeof value === "string" ? value : void 0;
 }
-function cloneJsonObject(source) {
-  const entries = Object.entries(source).map(([key, value]) => [key, cloneJsonValue(value)]);
-  return Object.fromEntries(entries);
-}
-function cloneJsonValue(value) {
-  if (value === null) {
-    return null;
-  }
-  if (typeof value === "string" || typeof value === "number" || typeof value === "boolean") {
-    return value;
-  }
-  if (Array.isArray(value)) {
-    return value.map((item) => cloneJsonValue(item));
-  }
-  if (typeof value === "object") {
-    return cloneJsonObject(value);
-  }
-  return value;
-}
 function logWarning3(message, details) {
   if (details && details.length > 0) {
     const detailBlock = details.join("\n");
@@ -16412,10 +16613,10 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
         );
       }
     }
-    const inputMessages = resolveInputMessages(evalcase);
+    const rawInputMessages = resolveInputMessages(evalcase);
     const expectedMessages = resolveExpectedMessages(evalcase) ?? [];
     const hasEvaluationSpec = !!outcome || expectedMessages.length > 0 || evalcase.assert !== void 0;
-    if (!id || !hasEvaluationSpec || !inputMessages || inputMessages.length === 0) {
+    if (!id || !hasEvaluationSpec || !rawInputMessages || rawInputMessages.length === 0) {
       logError2(
         `Skipping incomplete test at line ${lineNumber}: ${id ?? "unknown"}. Missing required fields: id, input, and at least one of criteria/expected_output/assert`
       );
@@ -16423,8 +16624,8 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
     }
     const hasExpectedMessages = expectedMessages.length > 0;
     const inputTextParts = [];
-    const inputSegments = await processMessages({
-      messages: inputMessages,
+    const inputMessages = await processMessages({
+      messages: rawInputMessages,
       searchRoots,
       repoRootPath,
       textParts: inputTextParts,
@@ -16470,19 +16671,13 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
       }
     }
     warnUnconsumedCriteria(outcome, evaluators, id ?? "unknown");
-    const userFilePaths = [];
-    for (const segment of inputSegments) {
-      if (segment.type === "file" && typeof segment.resolvedPath === "string") {
-        userFilePaths.push(segment.resolvedPath);
-      }
-    }
+    const userFilePaths = collectResolvedInputFilePaths(inputMessages);
     const testCase = {
       id,
       eval_set: evalSetName,
       conversation_id: conversationId,
       question,
       input: inputMessages,
-      input_segments: inputSegments,
       expected_output: outputSegments,
       reference_answer: referenceAnswer,
       file_paths: userFilePaths,
@@ -16543,50 +16738,9 @@ function parseMetadata(suite) {
   });
 }
 async function buildPromptInputs(testCase, mode = "lm") {
-  const segmentsByMessage = [];
-  const fileContentsByPath = /* @__PURE__ */ new Map();
-  for (const segment of testCase.input_segments) {
-    if (segment.type === "file" && typeof segment.path === "string" && typeof segment.text === "string") {
-      fileContentsByPath.set(segment.path, segment.text);
-    }
-  }
-  for (const message of testCase.input) {
-    const messageSegments = [];
-    if (typeof message.content === "string") {
-      if (message.content.trim().length > 0) {
-        messageSegments.push({ type: "text", value: message.content });
-      }
-    } else if (Array.isArray(message.content)) {
-      for (const segment of message.content) {
-        if (typeof segment === "string") {
-          if (segment.trim().length > 0) {
-            messageSegments.push({ type: "text", value: segment });
-          }
-        } else if (isJsonObject(segment)) {
-          const type = asString5(segment.type);
-          if (type === "file") {
-            const value = asString5(segment.value);
-            if (!value) continue;
-            const fileText = fileContentsByPath.get(value);
-            if (fileText !== void 0) {
-              messageSegments.push({ type: "file", text: fileText, path: value });
-            }
-          } else if (type === "text") {
-            const textValue = asString5(segment.value);
-            if (textValue && textValue.trim().length > 0) {
-              messageSegments.push({ type: "text", value: textValue });
-            }
-          }
-        }
-      }
-    } else if (isJsonObject(message.content)) {
-      const rendered = JSON.stringify(message.content, null, 2);
-      if (rendered.trim().length > 0) {
-        messageSegments.push({ type: "text", value: rendered });
-      }
-    }
-    segmentsByMessage.push(messageSegments);
-  }
+  const segmentsByMessage = testCase.input.map(
+    (message) => extractContentSegments(message.content)
+  );
   const useRoleMarkers = needsRoleMarkers(testCase.input, segmentsByMessage);
   let question;
   if (useRoleMarkers) {
@@ -16614,7 +16768,7 @@ ${messageContent}`);
     question = messageParts.join("\n\n");
   } else {
     const questionParts = [];
-    for (const segment of testCase.input_segments) {
+    for (const segment of flattenInputMessages(testCase.input)) {
       const formattedContent = formatSegment(segment, mode);
       if (formattedContent) {
         questionParts.push(formattedContent);
@@ -16701,9 +16855,6 @@ function buildChatPromptFromSegments(options) {
   }
   return chatPrompt.length > 0 ? chatPrompt : void 0;
 }
-function asString5(value) {
-  return typeof value === "string" ? value : void 0;
-}
 var ANSI_YELLOW6 = "\x1B[33m";
 var ANSI_RED3 = "\x1B[31m";
 var ANSI_RESET7 = "\x1B[0m";
@@ -16784,7 +16935,7 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
     throw new Error(`Invalid test file format: ${evalFilePath}`);
   }
   const suite = interpolated;
-  const evalSetNameFromSuite = asString6(suite.name)?.trim();
+  const evalSetNameFromSuite = asString5(suite.name)?.trim();
   const fallbackEvalSet = path7.basename(absoluteTestPath).replace(/\.ya?ml$/i, "") || "eval";
   const evalSetName = evalSetNameFromSuite && evalSetNameFromSuite.length > 0 ? evalSetNameFromSuite : fallbackEvalSet;
   const rawTestcases = resolveTests(suite);
@@ -16803,7 +16954,7 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
   const suiteInputMessages = expandInputShorthand(suite.input);
   const suiteInputFiles = suite.input_files;
   const rawGlobalExecution = isJsonObject(suite.execution) ? suite.execution : void 0;
-  const _globalTarget = asString6(rawGlobalExecution?.target) ?? asString6(suite.target);
+  const _globalTarget = asString5(rawGlobalExecution?.target) ?? asString5(suite.target);
   const suiteAssertions = suite.assertions ?? suite.assert;
   if (suite.assert !== void 0 && suite.assertions === void 0) {
     logWarning5("'assert' is deprecated at the suite level. Use 'assertions' instead.");
@@ -16816,17 +16967,17 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
       continue;
     }
     const evalcase = rawEvalcase;
-    const id = asString6(evalcase.id);
+    const id = asString5(evalcase.id);
     if (filterPattern && (!id || !micromatch2.isMatch(id, filterPattern))) {
       continue;
     }
-    const conversationId = asString6(evalcase.conversation_id);
-    let outcome = asString6(evalcase.criteria);
+    const conversationId = asString5(evalcase.conversation_id);
+    let outcome = asString5(evalcase.criteria);
     if (!outcome && evalcase.expected_outcome !== void 0) {
-      outcome = asString6(evalcase.expected_outcome);
+      outcome = asString5(evalcase.expected_outcome);
       if (outcome) {
         logWarning5(
-          `Test '${asString6(evalcase.id) ?? "unknown"}': 'expected_outcome' is deprecated. Use 'criteria' instead.`
+          `Test '${asString5(evalcase.id) ?? "unknown"}': 'expected_outcome' is deprecated. Use 'criteria' instead.`
         );
       }
     }
@@ -16843,10 +16994,9 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
       continue;
     }
     const effectiveSuiteInputMessages = suiteInputMessages && !skipDefaults ? suiteInputMessages : void 0;
-    const inputMessages = effectiveSuiteInputMessages ? [...effectiveSuiteInputMessages, ...testInputMessages] : testInputMessages;
     const hasExpectedMessages = expectedMessages.length > 0;
     const inputTextParts = [];
-    const suiteInputSegments = effectiveSuiteInputMessages ? await processMessages({
+    const suiteResolvedInputMessages = effectiveSuiteInputMessages ? await processMessages({
       messages: effectiveSuiteInputMessages,
       searchRoots,
       repoRootPath,
@@ -16854,7 +17004,7 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
       messageType: "input",
       verbose
     }) : [];
-    const testInputSegments = await processMessages({
+    const testResolvedInputMessages = await processMessages({
       messages: testInputMessages,
       searchRoots,
       repoRootPath,
@@ -16862,7 +17012,7 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
       messageType: "input",
       verbose
     });
-    const inputSegments = [...suiteInputSegments, ...testInputSegments];
+    const inputMessages = [...suiteResolvedInputMessages, ...testResolvedInputMessages];
     const outputSegments = hasExpectedMessages ? await processExpectedMessages({
       messages: expectedMessages,
       searchRoots,
@@ -16900,12 +17050,7 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
       }
     }
     warnUnconsumedCriteria(outcome, evaluators, id ?? "unknown");
-    const userFilePaths = [];
-    for (const segment of inputSegments) {
-      if (segment.type === "file" && typeof segment.resolvedPath === "string") {
-        userFilePaths.push(segment.resolvedPath);
-      }
-    }
+    const userFilePaths = collectResolvedInputFilePaths(inputMessages);
     const caseWorkspace = await resolveWorkspaceConfig(evalcase.workspace, evalFileDir);
     const mergedWorkspace = mergeWorkspaceConfigs(suiteWorkspace, caseWorkspace);
     const metadata = isJsonObject(evalcase.metadata) ? evalcase.metadata : void 0;
@@ -16916,7 +17061,6 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
       conversation_id: conversationId,
       question,
       input: inputMessages,
-      input_segments: inputSegments,
       expected_output: outputSegments,
       reference_answer: referenceAnswer,
       file_paths: userFilePaths,
@@ -17125,7 +17269,7 @@ function mergeWorkspaceConfigs(suiteLevel, caseLevel) {
     path: caseLevel.path ?? suiteLevel.path
   };
 }
-function asString6(value) {
+function asString5(value) {
   return typeof value === "string" ? value : void 0;
 }
 function logWarning5(message, details) {
@@ -20792,7 +20936,7 @@ var PiAgentSdkProvider = class {
     const { Agent, getModel, getEnvApiKey } = await loadPiModules();
     const startTimeIso = (/* @__PURE__ */ new Date()).toISOString();
     const startMs = Date.now();
-    const providerName = this.config.provider ?? "anthropic";
+    const providerName = this.config.subprovider ?? "anthropic";
     const modelId = this.config.model ?? "claude-sonnet-4-20250514";
     const model = getModel(providerName, modelId);
     const systemPrompt = this.config.systemPrompt ?? "Answer directly and concisely.";
@@ -20904,7 +21048,7 @@ var PiAgentSdkProvider = class {
           messages: agentMessages,
           systemPrompt,
           model: this.config.model,
-          provider: this.config.provider
+          subprovider: this.config.subprovider
         },
         output,
         tokenUsage,
@@ -21128,8 +21272,8 @@ var PiCodingAgentProvider = class {
   }
   buildPiArgs(prompt, inputFiles, _captureFileChanges) {
     const args = [];
-    if (this.config.provider) {
-      args.push("--provider", this.config.provider);
+    if (this.config.subprovider) {
+      args.push("--provider", this.config.subprovider);
     }
     if (this.config.model) {
       args.push("--model", this.config.model);
@@ -21187,7 +21331,7 @@ ${prompt}` : prompt;
   buildEnv() {
     const env = { ...process.env };
     if (this.config.apiKey) {
-      const provider = this.config.provider?.toLowerCase() ?? "google";
+      const provider = this.config.subprovider?.toLowerCase() ?? "google";
       switch (provider) {
         case "google":
         case "gemini":
@@ -23921,7 +24065,8 @@ var freeformEvaluationSchema = external_exports2.object({
       passed: external_exports2.boolean().describe("Whether this aspect was satisfied"),
       evidence: external_exports2.string().describe("Concise evidence (1-2 sentences)").optional()
     })
-  ).describe("Per-aspect evaluation results \u2014 one entry per aspect checked").optional()
+  ).describe("Per-aspect evaluation results \u2014 one entry per aspect checked").optional(),
+  details: external_exports2.record(external_exports2.unknown()).describe("Optional structured metadata for domain-specific metrics").optional()
 });
 var rubricCheckResultSchema = external_exports2.object({
   id: external_exports2.string().describe("The ID of the rubric item being checked"),
@@ -23983,7 +24128,7 @@ var LlmGraderEvaluator = class {
   async evaluateFreeform(context2, graderProvider) {
     const formattedQuestion = context2.promptInputs.question && context2.promptInputs.question.trim().length > 0 ? context2.promptInputs.question : context2.evalCase.question;
     const variables = {
-      [TEMPLATE_VARIABLES.INPUT]: JSON.stringify(context2.evalCase.input_segments, null, 2),
+      [TEMPLATE_VARIABLES.INPUT]: JSON.stringify(context2.evalCase.input, null, 2),
       [TEMPLATE_VARIABLES.EXPECTED_OUTPUT]: JSON.stringify(
         context2.evalCase.expected_output,
         null,
@@ -24026,6 +24171,7 @@ ${context2.fileChanges}`;
         expectedAspectCount: Math.max(assertions.length, 1),
         evaluatorRawRequest,
         graderTarget: graderProvider.targetName,
+        details: data.details,
         tokenUsage
       };
     } catch (e) {
@@ -24445,7 +24591,7 @@ ${outputSchema2}`;
         expectedAspectCount: Math.max(assertions.length, 1),
         evaluatorRawRequest,
         graderTarget,
-        details
+        details: data.details && Object.keys(data.details).length > 0 ? { ...details, ...data.details } : details
       };
     } catch {
       return {
@@ -24592,7 +24738,8 @@ function buildOutputSchema() {
     '      "passed": <boolean>,',
     '      "evidence": "<concise evidence, 1-2 sentences, optional>"',
     "    }",
-    "  ]",
+    "  ],",
+    '  "details": {<optional object with domain-specific structured metrics>}',
     "}"
   ].join("\n");
 }
@@ -25942,7 +26089,7 @@ function assembleLlmGraderPrompt(input) {
 function assembleFreeform(evalCase, candidate, promptInputs, fileChanges, evaluatorTemplateOverride) {
   const formattedQuestion = promptInputs.question && promptInputs.question.trim().length > 0 ? promptInputs.question : evalCase.question;
   const variables = {
-    [TEMPLATE_VARIABLES.INPUT]: JSON.stringify(evalCase.input_segments, null, 2),
+    [TEMPLATE_VARIABLES.INPUT]: JSON.stringify(evalCase.input, null, 2),
     [TEMPLATE_VARIABLES.EXPECTED_OUTPUT]: JSON.stringify(evalCase.expected_output, null, 2),
     [TEMPLATE_VARIABLES.OUTPUT]: JSON.stringify([], null, 2),
     [TEMPLATE_VARIABLES.CRITERIA]: evalCase.criteria.trim(),
@@ -28156,6 +28303,18 @@ var QUALITY_PASS_THRESHOLD = 0.8;
 function classifyQualityStatus(score) {
   return score >= QUALITY_PASS_THRESHOLD ? "ok" : "quality_failure";
 }
+function buildSkippedEvaluatorError(scores) {
+  const skippedScores = scores?.filter((score) => score.verdict === "skip") ?? [];
+  if (skippedScores.length === 0) {
+    return void 0;
+  }
+  const messages = skippedScores.map((score) => {
+    const label = score.name || score.type;
+    const assertionMessage = score.assertions.find((assertion) => !assertion.passed)?.text ?? "Evaluator skipped";
+    return `${label}: ${assertionMessage}`;
+  });
+  return messages.length === 1 ? messages[0] : `Evaluators skipped: ${messages.join(" | ")}`;
+}
 function usesFileReferencePrompt(provider) {
   return isAgentProvider(provider) || provider.kind === "cli";
 }
@@ -29420,7 +29579,8 @@ async function runEvalCase(options) {
       durationMs: totalDurationMs,
       ...evalRunTokenUsage ? { tokenUsage: evalRunTokenUsage } : {}
     };
-    const executionStatus = providerError ? "execution_error" : classifyQualityStatus(result.score);
+    const skippedEvaluatorError = buildSkippedEvaluatorError(result.scores);
+    const executionStatus = providerError || skippedEvaluatorError ? "execution_error" : classifyQualityStatus(result.score);
     const finalResult = providerError ? {
       ...result,
       evalRun,
@@ -29432,7 +29592,26 @@ async function runEvalCase(options) {
       beforeAllOutput,
       beforeEachOutput,
       afterEachOutput
-    } : { ...result, evalRun, executionStatus, beforeAllOutput, beforeEachOutput, afterEachOutput };
+    } : skippedEvaluatorError ? {
+      ...result,
+      score: 0,
+      evalRun,
+      error: skippedEvaluatorError,
+      executionStatus,
+      failureStage: "evaluator",
+      failureReasonCode: "evaluator_error",
+      executionError: { message: skippedEvaluatorError, stage: "evaluator" },
+      beforeAllOutput,
+      beforeEachOutput,
+      afterEachOutput
+    } : {
+      ...result,
+      evalRun,
+      executionStatus,
+      beforeAllOutput,
+      beforeEachOutput,
+      afterEachOutput
+    };
     const isFailure = !!finalResult.error || finalResult.score < 0.5;
     if (workspacePath && !isSharedWorkspace) {
       if (forceCleanup) {
@@ -30169,11 +30348,6 @@ async function evaluate(config) {
     evalCases = (config.tests ?? []).map((test) => {
       const input = typeof test.input === "string" ? [{ role: "user", content: test.input }] : test.input;
       const question = typeof test.input === "string" ? test.input : test.input.find((m) => m.role === "user")?.content ?? "";
-      const inputSegments = input.map((m) => ({
-        type: "text",
-        value: typeof m.content === "string" ? m.content : JSON.stringify(m.content),
-        messageIndex: 0
-      }));
       const expectedOutputValue = test.expectedOutput ?? test.expected_output;
       const expectedOutput = expectedOutputValue ? [
         { role: "assistant", content: expectedOutputValue }
@@ -30202,7 +30376,6 @@ async function evaluate(config) {
         criteria: test.criteria ?? "",
         question: String(question),
         input,
-        input_segments: inputSegments,
         expected_output: expectedOutput,
         reference_answer: expectedOutputValue,
         file_paths: [],
@@ -31103,4 +31276,4 @@ export {
   OtelStreamingObserver,
   createAgentKernel
 };
-//# sourceMappingURL=chunk-JUQCB3ZW.js.map
+//# sourceMappingURL=chunk-EZGWZVVK.js.map