@agentv/core 0.7.2 → 0.7.4

package/dist/index.js

@@ -6,8 +6,9 @@ import {
   findGitRoot,
   isAgentProvider,
   readTextFile,
-  resolveFileReference
-} from "./chunk-UQLHF3T7.js";
+  resolveFileReference,
+  resolveTargetDefinition
+} from "./chunk-L6RCDZ4Z.js";
 
 // src/evaluation/types.ts
 var TEST_MESSAGE_ROLE_VALUES = ["system", "user", "assistant", "tool"];
@@ -218,6 +219,7 @@ async function processMessages(options) {
 }
 async function loadEvalCases(evalFilePath, repoRoot, options) {
   const verbose = options?.verbose ?? false;
+  const evalIdFilter = options?.evalId;
   const absoluteTestPath = path.resolve(evalFilePath);
   if (!await fileExists2(absoluteTestPath)) {
     throw new Error(`Test file not found: ${evalFilePath}`);
@@ -249,62 +251,39 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
   const results = [];
   for (const rawEvalcase of rawTestcases) {
     if (!isJsonObject(rawEvalcase)) {
-      logWarning("Skipping invalid test case entry (expected object)");
+      logWarning("Skipping invalid eval case entry (expected object)");
       continue;
     }
     const evalcase = rawEvalcase;
     const id = asString(evalcase.id);
+    if (evalIdFilter && id !== evalIdFilter) {
+      continue;
+    }
     const conversationId = asString(evalcase.conversation_id);
     const outcome = asString(evalcase.outcome);
     const inputMessagesValue = evalcase.input_messages;
     const expectedMessagesValue = evalcase.expected_messages;
     if (!id || !outcome || !Array.isArray(inputMessagesValue)) {
-      logWarning(`Skipping incomplete test case: ${id ?? "unknown"}`);
+      logWarning(`Skipping incomplete eval case: ${id ?? "unknown"}`);
       continue;
     }
     if (!Array.isArray(expectedMessagesValue)) {
-      logWarning(`Test case '${id}' missing expected_messages array`);
+      logWarning(`Eval case '${id}' missing expected_messages array`);
       continue;
     }
     const inputMessages = inputMessagesValue.filter((msg) => isTestMessage(msg));
     const expectedMessages = expectedMessagesValue.filter((msg) => isTestMessage(msg));
-    const assistantMessages = expectedMessages.filter((message) => message.role === "assistant");
-    const userMessages = inputMessages.filter((message) => message.role === "user");
-    const systemMessages = inputMessages.filter((message) => message.role === "system");
-    if (assistantMessages.length === 0) {
-      logWarning(`No assistant message found for test case: ${id}`);
+    if (expectedMessages.length === 0) {
+      logWarning(`No expected message found for eval case: ${id}`);
       continue;
     }
-    if (assistantMessages.length > 1) {
-      logWarning(`Multiple assistant messages found for test case: ${id}, using first`);
-    }
-    if (systemMessages.length > 1) {
-      logWarning(`Multiple system messages found for test case: ${id}, using first`);
-    }
-    let systemMessageContent;
-    if (systemMessages.length > 0) {
-      const content = systemMessages[0]?.content;
-      if (typeof content === "string") {
-        systemMessageContent = content;
-      } else if (Array.isArray(content)) {
-        const textParts = [];
-        for (const segment of content) {
-          if (isJsonObject(segment)) {
-            const value = segment.value;
-            if (typeof value === "string") {
-              textParts.push(value);
-            }
-          }
-        }
-        if (textParts.length > 0) {
-          systemMessageContent = textParts.join("\n\n");
-        }
-      }
+    if (expectedMessages.length > 1) {
+      logWarning(`Multiple expected messages found for eval case: ${id}, using first`);
     }
     const guidelinePaths = [];
     const inputTextParts = [];
     const inputSegments = await processMessages({
-      messages: userMessages,
+      messages: inputMessages,
       searchRoots,
       repoRootPath,
       guidelinePatterns,
@@ -314,7 +293,7 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
       verbose
     });
     const outputSegments = await processMessages({
-      messages: assistantMessages,
+      messages: expectedMessages,
       searchRoots,
       repoRootPath,
       guidelinePatterns,
@@ -322,10 +301,10 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
       verbose
     });
     const codeSnippets = extractCodeBlocks(inputSegments);
-    const assistantContent = assistantMessages[0]?.content;
-    const referenceAnswer = await resolveAssistantContent(assistantContent, searchRoots, verbose);
+    const expectedContent = expectedMessages[0]?.content;
+    const referenceAnswer = await resolveAssistantContent(expectedContent, searchRoots, verbose);
     const question = inputTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
-    const testCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
+    const evalCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
     const evaluators = await parseEvaluators(evalcase, searchRoots, id ?? "unknown");
     const userFilePaths = [];
     for (const segment of inputSegments) {
@@ -344,19 +323,18 @@ Please add '$schema: ${SCHEMA_EVAL_V2}' at the top of the file.`;
       question,
       input_segments: inputSegments,
       output_segments: outputSegments,
-      system_message: systemMessageContent,
       reference_answer: referenceAnswer,
       guideline_paths: guidelinePaths.map((guidelinePath) => path.resolve(guidelinePath)),
       guideline_patterns: guidelinePatterns,
       file_paths: allFilePaths,
       code_snippets: codeSnippets,
       expected_outcome: outcome,
-      evaluator: testCaseEvaluatorKind,
+      evaluator: evalCaseEvaluatorKind,
       evaluators
     };
     if (verbose) {
       console.log(`
-[Test Case: ${id}]`);
+[Eval Case: ${id}]`);
       if (testCase.guideline_paths.length > 0) {
         console.log(`  Guidelines used: ${testCase.guideline_paths.length}`);
         for (const guidelinePath of testCase.guideline_paths) {
@@ -415,7 +393,7 @@ ${body}`);
   }
   const question = questionParts.map((part) => part.trim()).filter((part) => part.length > 0).join("\n\n");
   const guidelines = guidelineContents.map((part) => part.trim()).filter((part) => part.length > 0).join("\n\n");
-  return { question, guidelines, systemMessage: testCase.system_message };
+  return { question, guidelines };
 }
 async function fileExists2(absolutePath) {
   try {
@@ -801,6 +779,8 @@ var GeminiProvider = class {
 
 // src/evaluation/providers/cli.ts
 import { exec as execWithCallback } from "node:child_process";
+import fs from "node:fs/promises";
+import os from "node:os";
 import path2 from "node:path";
 import { promisify } from "node:util";
 var execAsync = promisify(execWithCallback);
@@ -816,6 +796,7 @@ async function defaultCommandRunner(command, options) {
   };
   try {
     const { stdout, stderr } = await execAsync(command, execOptions);
+    console.error(`[CLI DEBUG] SUCCESS - stdout: ${stdout.length} bytes, stderr: ${stderr.length} bytes`);
     return {
       stdout,
       stderr,
@@ -826,6 +807,8 @@ async function defaultCommandRunner(command, options) {
     };
   } catch (error) {
     const execError = error;
+    console.error(`[CLI DEBUG] ERROR - code: ${execError.code}, message: ${execError.message}`);
+    console.error(`[CLI DEBUG] stdout: ${execError.stdout?.length ?? 0} bytes, stderr: ${execError.stderr?.length ?? 0} bytes`);
     return {
       stdout: execError.stdout ?? "",
       stderr: execError.stderr ?? "",
@@ -855,7 +838,8 @@ var CliProvider = class {
       throw new Error("CLI provider request was aborted before execution");
     }
     await this.ensureHealthy(request.signal);
-    const templateValues = buildTemplateValues(request, this.config);
+    const outputFilePath = generateOutputFilePath(request.evalCaseId);
+    const templateValues = buildTemplateValues(request, this.config, outputFilePath);
     const renderedCommand = renderTemplate(this.config.commandTemplate, templateValues);
     const env = this.config.env ? { ...process.env, ...this.config.env } : process.env;
     const result = await this.runCommand(renderedCommand, {
@@ -878,16 +862,30 @@ var CliProvider = class {
       const message = detail ? `${detail} (exit code ${codeText})` : `CLI exited with code ${codeText}`;
       throw new Error(message);
     }
+    const responseText = await this.readAndCleanupOutputFile(outputFilePath);
     return {
-      text: result.stdout,
+      text: responseText,
       raw: {
         command: renderedCommand,
         stderr: result.stderr,
         exitCode: result.exitCode ?? 0,
-        cwd: this.config.cwd
+        cwd: this.config.cwd,
+        outputFile: outputFilePath
       }
     };
   }
+  async readAndCleanupOutputFile(filePath) {
+    try {
+      const content = await fs.readFile(filePath, "utf-8");
+      return content;
+    } catch (error) {
+      const errorMsg = error instanceof Error ? error.message : String(error);
+      throw new Error(`Failed to read output file '${filePath}': ${errorMsg}`);
+    } finally {
+      await fs.unlink(filePath).catch(() => {
+      });
+    }
+  }
   async ensureHealthy(signal) {
     if (!this.config.healthcheck) {
       return;
@@ -928,10 +926,11 @@ var CliProvider = class {
           question: "",
           guidelines: "",
           inputFiles: [],
-          evalCaseId: "",
+          evalCaseId: "healthcheck",
           attempt: 0
         },
-        this.config
+        this.config,
+        generateOutputFilePath("healthcheck")
       )
     );
     const env = this.config.env ? { ...process.env, ...this.config.env } : process.env;
@@ -949,14 +948,15 @@ var CliProvider = class {
     }
   }
 };
-function buildTemplateValues(request, config) {
+function buildTemplateValues(request, config, outputFilePath) {
   const inputFiles = normalizeInputFiles(request.inputFiles);
   return {
     PROMPT: shellEscape(request.question ?? ""),
     GUIDELINES: shellEscape(request.guidelines ?? ""),
     EVAL_ID: shellEscape(request.evalCaseId ?? ""),
     ATTEMPT: shellEscape(String(request.attempt ?? 0)),
-    FILES: formatFileList(inputFiles, config.filesFormat)
+    FILES: formatFileList(inputFiles, config.filesFormat),
+    OUTPUT_FILE: shellEscape(outputFilePath)
   };
 }
 function normalizeInputFiles(inputFiles) {
@@ -994,11 +994,17 @@ function shellEscape(value) {
     return "''";
   }
   if (process.platform === "win32") {
-    const escaped = value.replace(/"/g, '\\"');
-    return `"${escaped}"`;
+    const escaped = value.replace(/'/g, "''");
+    return `'${escaped}'`;
   }
   return `'${value.replace(/'/g, `'"'"'`)}'`;
 }
+function generateOutputFilePath(evalCaseId) {
+  const safeEvalId = evalCaseId || "unknown";
+  const timestamp = Date.now();
+  const random = Math.random().toString(36).substring(2, 9);
+  return path2.join(os.tmpdir(), `agentv-${safeEvalId}-${timestamp}-${random}.json`);
+}
 function formatTimeoutSuffix(timeoutMs) {
   if (!timeoutMs || timeoutMs <= 0) {
     return "";
@@ -1875,487 +1881,6 @@ var MockProvider = class {
   }
 };
 
-// src/evaluation/providers/targets.ts
-import { z } from "zod";
-var CLI_PLACEHOLDERS = /* @__PURE__ */ new Set(["PROMPT", "GUIDELINES", "EVAL_ID", "ATTEMPT", "FILES"]);
-var BASE_TARGET_SCHEMA = z.object({
-  name: z.string().min(1, "target name is required"),
-  provider: z.string().min(1, "provider is required"),
-  settings: z.record(z.unknown()).optional(),
-  judge_target: z.string().optional(),
-  workers: z.number().int().min(1).optional()
-});
-var DEFAULT_AZURE_API_VERSION = "2024-10-01-preview";
-function normalizeAzureApiVersion(value) {
-  if (!value) {
-    return DEFAULT_AZURE_API_VERSION;
-  }
-  const trimmed = value.trim();
-  if (trimmed.length === 0) {
-    return DEFAULT_AZURE_API_VERSION;
-  }
-  const withoutPrefix = trimmed.replace(/^api[-_]?version\s*=\s*/i, "").trim();
-  return withoutPrefix.length > 0 ? withoutPrefix : DEFAULT_AZURE_API_VERSION;
-}
-function resolveTargetDefinition(definition, env = process.env) {
-  const parsed = BASE_TARGET_SCHEMA.parse(definition);
-  const provider = parsed.provider.toLowerCase();
-  const providerBatching = resolveOptionalBoolean(
-    parsed.settings?.provider_batching ?? parsed.settings?.providerBatching
-  );
-  switch (provider) {
-    case "azure":
-    case "azure-openai":
-      return {
-        kind: "azure",
-        name: parsed.name,
-        judgeTarget: parsed.judge_target,
-        workers: parsed.workers,
-        providerBatching,
-        config: resolveAzureConfig(parsed, env)
-      };
-    case "anthropic":
-      return {
-        kind: "anthropic",
-        name: parsed.name,
-        judgeTarget: parsed.judge_target,
-        workers: parsed.workers,
-        providerBatching,
-        config: resolveAnthropicConfig(parsed, env)
-      };
-    case "gemini":
-    case "google":
-    case "google-gemini":
-      return {
-        kind: "gemini",
-        name: parsed.name,
-        judgeTarget: parsed.judge_target,
-        workers: parsed.workers,
-        providerBatching,
-        config: resolveGeminiConfig(parsed, env)
-      };
-    case "codex":
-    case "codex-cli":
-      return {
-        kind: "codex",
-        name: parsed.name,
-        judgeTarget: parsed.judge_target,
-        workers: parsed.workers,
-        providerBatching,
-        config: resolveCodexConfig(parsed, env)
-      };
-    case "mock":
-      return {
-        kind: "mock",
-        name: parsed.name,
-        judgeTarget: parsed.judge_target,
-        workers: parsed.workers,
-        providerBatching,
-        config: resolveMockConfig(parsed)
-      };
-    case "vscode":
-    case "vscode-insiders":
-      return {
-        kind: provider,
-        name: parsed.name,
-        judgeTarget: parsed.judge_target,
-        workers: parsed.workers,
-        providerBatching,
-        config: resolveVSCodeConfig(parsed, env, provider === "vscode-insiders")
-      };
-    case "cli":
-      return {
-        kind: "cli",
-        name: parsed.name,
-        judgeTarget: parsed.judge_target,
-        workers: parsed.workers,
-        providerBatching,
-        config: resolveCliConfig(parsed, env)
-      };
-    default:
-      throw new Error(`Unsupported provider '${parsed.provider}' in target '${parsed.name}'`);
-  }
-}
-function resolveAzureConfig(target, env) {
-  const settings = target.settings ?? {};
-  const endpointSource = settings.endpoint ?? settings.resource ?? settings.resourceName;
-  const apiKeySource = settings.api_key ?? settings.apiKey;
-  const deploymentSource = settings.deployment ?? settings.deploymentName ?? settings.model;
-  const versionSource = settings.version ?? settings.api_version;
-  const temperatureSource = settings.temperature;
-  const maxTokensSource = settings.max_output_tokens ?? settings.maxTokens;
-  const resourceName = resolveString(endpointSource, env, `${target.name} endpoint`);
-  const apiKey = resolveString(apiKeySource, env, `${target.name} api key`);
-  const deploymentName = resolveString(deploymentSource, env, `${target.name} deployment`);
-  const version = normalizeAzureApiVersion(
-    resolveOptionalString(versionSource, env, `${target.name} api version`)
-  );
-  const temperature = resolveOptionalNumber(temperatureSource, `${target.name} temperature`);
-  const maxOutputTokens = resolveOptionalNumber(
-    maxTokensSource,
-    `${target.name} max output tokens`
-  );
-  return {
-    resourceName,
-    deploymentName,
-    apiKey,
-    version,
-    temperature,
-    maxOutputTokens
-  };
-}
-function resolveAnthropicConfig(target, env) {
-  const settings = target.settings ?? {};
-  const apiKeySource = settings.api_key ?? settings.apiKey;
-  const modelSource = settings.model ?? settings.deployment ?? settings.variant;
-  const temperatureSource = settings.temperature;
-  const maxTokensSource = settings.max_output_tokens ?? settings.maxTokens;
-  const thinkingBudgetSource = settings.thinking_budget ?? settings.thinkingBudget;
-  const apiKey = resolveString(apiKeySource, env, `${target.name} Anthropic api key`);
-  const model = resolveString(modelSource, env, `${target.name} Anthropic model`);
-  return {
-    apiKey,
-    model,
-    temperature: resolveOptionalNumber(temperatureSource, `${target.name} temperature`),
-    maxOutputTokens: resolveOptionalNumber(maxTokensSource, `${target.name} max output tokens`),
-    thinkingBudget: resolveOptionalNumber(thinkingBudgetSource, `${target.name} thinking budget`)
-  };
-}
-function resolveGeminiConfig(target, env) {
-  const settings = target.settings ?? {};
-  const apiKeySource = settings.api_key ?? settings.apiKey;
-  const modelSource = settings.model ?? settings.deployment ?? settings.variant;
-  const temperatureSource = settings.temperature;
-  const maxTokensSource = settings.max_output_tokens ?? settings.maxTokens;
-  const apiKey = resolveString(apiKeySource, env, `${target.name} Google API key`);
-  const model = resolveOptionalString(modelSource, env, `${target.name} Gemini model`, {
-    allowLiteral: true,
-    optionalEnv: true
-  }) ?? "gemini-2.5-flash";
-  return {
-    apiKey,
-    model,
-    temperature: resolveOptionalNumber(temperatureSource, `${target.name} temperature`),
-    maxOutputTokens: resolveOptionalNumber(maxTokensSource, `${target.name} max output tokens`)
-  };
-}
-function resolveCodexConfig(target, env) {
-  const settings = target.settings ?? {};
-  const executableSource = settings.executable ?? settings.command ?? settings.binary;
-  const argsSource = settings.args ?? settings.arguments;
-  const cwdSource = settings.cwd;
-  const timeoutSource = settings.timeout_seconds ?? settings.timeoutSeconds;
-  const logDirSource = settings.log_dir ?? settings.logDir ?? settings.log_directory ?? settings.logDirectory;
-  const logFormatSource = settings.log_format ?? settings.logFormat ?? settings.log_output_format ?? settings.logOutputFormat ?? env.AGENTV_CODEX_LOG_FORMAT;
-  const executable = resolveOptionalString(executableSource, env, `${target.name} codex executable`, {
-    allowLiteral: true,
-    optionalEnv: true
-  }) ?? "codex";
-  const args = resolveOptionalStringArray(argsSource, env, `${target.name} codex args`);
-  const cwd = resolveOptionalString(cwdSource, env, `${target.name} codex cwd`, {
-    allowLiteral: true,
-    optionalEnv: true
-  });
-  const timeoutMs = resolveTimeoutMs(timeoutSource, `${target.name} codex timeout`);
-  const logDir = resolveOptionalString(logDirSource, env, `${target.name} codex log directory`, {
-    allowLiteral: true,
-    optionalEnv: true
-  });
-  const logFormat = normalizeCodexLogFormat(logFormatSource);
-  return {
-    executable,
-    args,
-    cwd,
-    timeoutMs,
-    logDir,
-    logFormat
-  };
-}
-function normalizeCodexLogFormat(value) {
-  if (value === void 0 || value === null) {
-    return void 0;
-  }
-  if (typeof value !== "string") {
-    throw new Error("codex log format must be 'summary' or 'json'");
-  }
-  const normalized = value.trim().toLowerCase();
-  if (normalized === "json" || normalized === "summary") {
-    return normalized;
-  }
-  throw new Error("codex log format must be 'summary' or 'json'");
-}
-function resolveMockConfig(target) {
-  const settings = target.settings ?? {};
-  const response = typeof settings.response === "string" ? settings.response : void 0;
-  return { response };
-}
-function resolveVSCodeConfig(target, env, insiders) {
-  const settings = target.settings ?? {};
-  const workspaceTemplateEnvVar = resolveOptionalLiteralString(settings.workspace_template ?? settings.workspaceTemplate);
-  const workspaceTemplate = workspaceTemplateEnvVar ? resolveOptionalString(workspaceTemplateEnvVar, env, `${target.name} workspace template path`, {
-    allowLiteral: false,
-    optionalEnv: true
-  }) : void 0;
-  const commandSource = settings.vscode_cmd ?? settings.command;
-  const waitSource = settings.wait;
-  const dryRunSource = settings.dry_run ?? settings.dryRun;
-  const subagentRootSource = settings.subagent_root ?? settings.subagentRoot;
-  const defaultCommand = insiders ? "code-insiders" : "code";
-  const command = resolveOptionalLiteralString(commandSource) ?? defaultCommand;
-  return {
-    command,
-    waitForResponse: resolveOptionalBoolean(waitSource) ?? true,
-    dryRun: resolveOptionalBoolean(dryRunSource) ?? false,
-    subagentRoot: resolveOptionalString(subagentRootSource, env, `${target.name} subagent root`, {
-      allowLiteral: true,
-      optionalEnv: true
-    }),
-    workspaceTemplate
-  };
-}
-function resolveCliConfig(target, env) {
-  const settings = target.settings ?? {};
-  const commandTemplateSource = settings.command_template ?? settings.commandTemplate;
-  const filesFormat = resolveOptionalLiteralString(
-    settings.files_format ?? settings.filesFormat ?? settings.attachments_format ?? settings.attachmentsFormat
-  );
-  const cwd = resolveOptionalString(settings.cwd, env, `${target.name} working directory`, {
-    allowLiteral: true,
-    optionalEnv: true
-  });
-  const envOverrides = resolveEnvOverrides(settings.env, env, target.name);
-  const timeoutMs = resolveTimeoutMs(settings.timeout_seconds ?? settings.timeoutSeconds, `${target.name} timeout`);
-  const healthcheck = resolveCliHealthcheck(settings.healthcheck, env, target.name);
-  const commandTemplate = resolveString(
-    commandTemplateSource,
-    env,
-    `${target.name} CLI command template`,
-    true
-  );
-  assertSupportedCliPlaceholders(commandTemplate, `${target.name} CLI command template`);
-  return {
-    commandTemplate,
-    filesFormat,
-    cwd,
-    env: envOverrides,
-    timeoutMs,
-    healthcheck
-  };
-}
-function resolveEnvOverrides(source, env, targetName) {
-  if (source === void 0 || source === null) {
-    return void 0;
-  }
-  if (typeof source !== "object" || Array.isArray(source)) {
-    throw new Error(`${targetName} env overrides must be an object map of strings`);
-  }
-  const entries = Object.entries(source);
-  const resolved = {};
-  for (const [key, value] of entries) {
-    if (typeof value !== "string") {
-      throw new Error(`${targetName} env override '${key}' must be a string`);
-    }
-    const resolvedValue = resolveString(value, env, `${targetName} env override '${key}'`);
-    resolved[key] = resolvedValue;
-  }
-  return Object.keys(resolved).length > 0 ? resolved : void 0;
-}
-function resolveTimeoutMs(source, description) {
-  const seconds = resolveOptionalNumber(source, `${description} (seconds)`);
-  if (seconds === void 0) {
-    return void 0;
-  }
-  if (seconds <= 0) {
-    throw new Error(`${description} must be greater than zero seconds`);
-  }
-  return Math.floor(seconds * 1e3);
-}
-function resolveCliHealthcheck(source, env, targetName) {
-  if (source === void 0 || source === null) {
-    return void 0;
-  }
-  if (typeof source !== "object" || Array.isArray(source)) {
-    throw new Error(`${targetName} healthcheck must be an object`);
-  }
-  const candidate = source;
-  const type = candidate.type;
-  const timeoutMs = resolveTimeoutMs(
-    candidate.timeout_seconds ?? candidate.timeoutSeconds,
-    `${targetName} healthcheck timeout`
-  );
-  if (type === "http") {
-    const url = resolveString(candidate.url, env, `${targetName} healthcheck URL`);
-    return {
-      type: "http",
-      url,
-      timeoutMs
-    };
-  }
-  if (type === "command") {
-    const commandTemplate = resolveString(
-      candidate.command_template ?? candidate.commandTemplate,
-      env,
-      `${targetName} healthcheck command template`,
-      true
-    );
-    assertSupportedCliPlaceholders(commandTemplate, `${targetName} healthcheck command template`);
-    const cwd = resolveOptionalString(candidate.cwd, env, `${targetName} healthcheck cwd`, {
-      allowLiteral: true,
-      optionalEnv: true
-    });
-    return {
-      type: "command",
-      commandTemplate,
-      timeoutMs,
-      cwd
-    };
-  }
-  throw new Error(`${targetName} healthcheck type must be 'http' or 'command'`);
-}
-function assertSupportedCliPlaceholders(template, description) {
-  const placeholders = extractCliPlaceholders(template);
-  for (const placeholder of placeholders) {
-    if (!CLI_PLACEHOLDERS.has(placeholder)) {
-      throw new Error(
-        `${description} includes unsupported placeholder '{${placeholder}}'. Supported placeholders: ${Array.from(CLI_PLACEHOLDERS).join(", ")}`
-      );
-    }
-  }
-}
-function extractCliPlaceholders(template) {
-  const matches = template.matchAll(/\{([A-Z_]+)\}/g);
-  const results = [];
-  for (const match of matches) {
-    if (match[1]) {
-      results.push(match[1]);
-    }
-  }
-  return results;
-}
-function resolveString(source, env, description, allowLiteral = false) {
-  const value = resolveOptionalString(source, env, description, {
-    allowLiteral,
-    optionalEnv: false
-  });
-  if (value === void 0) {
-    throw new Error(`${description} is required`);
-  }
-  return value;
-}
-function resolveOptionalString(source, env, description, options) {
-  if (source === void 0 || source === null) {
-    return void 0;
-  }
-  if (typeof source !== "string") {
-    throw new Error(`${description} must be a string`);
-  }
-  const trimmed = source.trim();
-  if (trimmed.length === 0) {
-    return void 0;
-  }
-  const envVarMatch = trimmed.match(/^\$\{\{\s*([A-Z0-9_]+)\s*\}\}$/i);
-  if (envVarMatch) {
-    const varName = envVarMatch[1];
-    const envValue = env[varName];
-    if (envValue !== void 0) {
-      if (envValue.trim().length === 0) {
-        throw new Error(`Environment variable '${varName}' for ${description} is empty`);
-      }
-      return envValue;
-    }
-    const optionalEnv = options?.optionalEnv ?? false;
-    if (optionalEnv) {
-      return void 0;
-    }
-    throw new Error(`Environment variable '${varName}' required for ${description} is not set`);
-  }
-  const allowLiteral = options?.allowLiteral ?? false;
-  if (!allowLiteral) {
-    throw new Error(`${description} must use \${{ VARIABLE_NAME }} syntax for environment variables or be marked as allowing literals`);
-  }
-  return trimmed;
-}
-function resolveOptionalLiteralString(source) {
-  if (source === void 0 || source === null) {
-    return void 0;
-  }
-  if (typeof source !== "string") {
-    throw new Error("expected string value");
-  }
-  const trimmed = source.trim();
-  return trimmed.length > 0 ? trimmed : void 0;
-}
-function resolveOptionalNumber(source, description) {
-  if (source === void 0 || source === null || source === "") {
-    return void 0;
-  }
-  if (typeof source === "number") {
-    return Number.isFinite(source) ? source : void 0;
-  }
-  if (typeof source === "string") {
-    const numeric = Number(source);
-    if (Number.isFinite(numeric)) {
-      return numeric;
-    }
-  }
-  throw new Error(`${description} must be a number`);
-}
-function resolveOptionalBoolean(source) {
-  if (source === void 0 || source === null || source === "") {
-    return void 0;
-  }
-  if (typeof source === "boolean") {
-    return source;
-  }
-  if (typeof source === "string") {
-    const lowered = source.trim().toLowerCase();
-    if (lowered === "true" || lowered === "1") {
-      return true;
-    }
-    if (lowered === "false" || lowered === "0") {
-      return false;
-    }
-  }
-  throw new Error("expected boolean value");
-}
-function resolveOptionalStringArray(source, env, description) {
-  if (source === void 0 || source === null) {
-    return void 0;
-  }
-  if (!Array.isArray(source)) {
-    throw new Error(`${description} must be an array of strings`);
-  }
-  if (source.length === 0) {
-    return void 0;
-  }
-  const resolved = [];
-  for (let i = 0; i < source.length; i++) {
-    const item = source[i];
-    if (typeof item !== "string") {
-      throw new Error(`${description}[${i}] must be a string`);
-    }
-    const trimmed = item.trim();
-    if (trimmed.length === 0) {
-      throw new Error(`${description}[${i}] cannot be empty`);
-    }
-    const envVarMatch = trimmed.match(/^\$\{\{\s*([A-Z0-9_]+)\s*\}\}$/i);
-    if (envVarMatch) {
-      const varName = envVarMatch[1];
-      const envValue = env[varName];
-      if (envValue !== void 0) {
-        if (envValue.trim().length === 0) {
-          throw new Error(`Environment variable '${varName}' for ${description}[${i}] is empty`);
-        }
-        resolved.push(envValue);
-        continue;
-      }
-      throw new Error(`Environment variable '${varName}' for ${description}[${i}] is not set`);
-    }
-    resolved.push(trimmed);
-  }
-  return resolved.length > 0 ? resolved : void 0;
-}
-
 // src/evaluation/providers/vscode.ts
 import path5 from "node:path";
 import { dispatchAgentSession, dispatchBatchAgent, getSubagentRoot, provisionSubagents } from "subagent";
@@ -2918,7 +2443,6 @@ var CodeEvaluator = class {
         expected_outcome: context.evalCase.expected_outcome,
         reference_answer: context.evalCase.reference_answer,
         candidate_answer: context.candidate,
-        system_message: context.promptInputs.systemMessage ?? "",
         guideline_paths: context.evalCase.guideline_paths,
         input_files: context.evalCase.file_paths,
         input_segments: context.evalCase.input_segments
@@ -3160,7 +2684,7 @@ function validateConcurrency(concurrency) {
 // src/evaluation/orchestrator.ts
 async function runEvaluation(options) {
   const {
-    testFilePath,
+    testFilePath: evalFilePath,
     repoRoot,
     target,
     targets,
@@ -3179,11 +2703,11 @@ async function runEvaluation(options) {
     onProgress
   } = options;
   const load = loadEvalCases;
-  const evalCases = await load(testFilePath, repoRoot, { verbose });
+  const evalCases = await load(evalFilePath, repoRoot, { verbose, evalId });
   const filteredEvalCases = filterEvalCases(evalCases, evalId);
   if (filteredEvalCases.length === 0) {
     if (evalId) {
-      throw new Error(`Test case with id '${evalId}' not found in ${testFilePath}`);
+      throw new Error(`Eval case with id '${evalId}' not found in ${evalFilePath}`);
     }
     return [];
   }
@@ -3562,8 +3086,7 @@ async function evaluateCandidate(options) {
   const rawRequest = {
     question: promptInputs.question,
     ...isAgentProvider(provider) ? {} : { guidelines: promptInputs.guidelines },
-    guideline_paths: evalCase.guideline_paths,
-    system_message: promptInputs.systemMessage ?? ""
+    guideline_paths: evalCase.guideline_paths
   };
   return {
     eval_id: evalCase.id,
@@ -3827,7 +3350,6 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs,
     question: promptInputs.question,
     ...isAgentProvider(provider) ? {} : { guidelines: promptInputs.guidelines },
     guideline_paths: evalCase.guideline_paths,
-    system_message: promptInputs.systemMessage ?? "",
     error: message
   };
   return {