@agentv/core 2.18.3 → 2.19.0

package/dist/evaluation/validation/index.js

@@ -6,7 +6,7 @@ import {
   findGitRoot,
   isEvaluatorKind,
   resolveFileReference
-} from "../../chunk-V42NUK73.js";
+} from "../../chunk-ACTIPQZ3.js";
 
 // src/evaluation/validation/file-type.ts
 import { readFile } from "node:fs/promises";

package/dist/index.cjs

@@ -2409,6 +2409,11 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
     }
     if (typeValue === "code-judge") {
       let command;
+      if (rawEvaluator.script !== void 0 && rawEvaluator.command === void 0) {
+        console.warn(
+          `${ANSI_YELLOW4}Warning: 'script' is deprecated in evaluator '${name}' in '${evalId}'. Use 'command' instead.${ANSI_RESET4}`
+        );
+      }
       const rawCommand = rawEvaluator.command ?? rawEvaluator.script;
       if (typeof rawCommand === "string") {
         const trimmed = rawCommand.trim();
@@ -3167,6 +3172,11 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
     let resolvedPromptScript;
     let promptScriptConfig;
     if (isJsonObject2(rawPrompt)) {
+      if (rawPrompt.script !== void 0 && rawPrompt.command === void 0) {
+        console.warn(
+          `${ANSI_YELLOW4}Warning: 'prompt.script' is deprecated in evaluator '${name}' in '${evalId}'. Use 'prompt.command' instead.${ANSI_RESET4}`
+        );
+      }
       const commandArray = asStringArray(
         rawPrompt.command ?? rawPrompt.script,
         `prompt.command for evaluator '${name}' in '${evalId}'`
@@ -4697,6 +4707,9 @@ var loadEvalCaseById = loadTestById;
 function parseWorkspaceScriptConfig(raw, evalFileDir) {
   if (!isJsonObject(raw)) return void 0;
   const obj = raw;
+  if (obj.script !== void 0 && obj.command === void 0) {
+    logWarning6("'script' is deprecated. Use 'command' instead.");
+  }
   const commandSource = obj.command ?? obj.script;
   if (!Array.isArray(commandSource) || commandSource.length === 0) return void 0;
   const commandArr = commandSource.filter((s) => typeof s === "string");
@@ -5552,7 +5565,7 @@ var ClaudeProvider = class {
       // a Claude Code session the CLAUDECODE env var is set, which causes the
       // subprocess to refuse to start ("cannot be launched inside another Claude
       // Code session"). Passing a sanitized env removes that guard.
-      env: sanitizeEnvForClaudeSdk()
+      env: sanitizeEnvForClaudeSdk(request.braintrustSpanIds)
     };
     if (this.config.model) {
       queryOptions.model = this.config.model;
@@ -5856,10 +5869,14 @@ function summarizeMessage(msg) {
       return void 0;
   }
 }
-function sanitizeEnvForClaudeSdk() {
+function sanitizeEnvForClaudeSdk(braintrustSpanIds) {
   const env = { ...process.env };
   env.CLAUDECODE = void 0;
   env.CLAUDE_CODE_ENTRYPOINT = void 0;
+  if (braintrustSpanIds) {
+    env.CC_PARENT_SPAN_ID = braintrustSpanIds.parentSpanId;
+    env.CC_ROOT_SPAN_ID = braintrustSpanIds.rootSpanId;
+  }
   return env;
 }
 function isClaudeLogStreamingDisabled() {
@@ -16267,6 +16284,11 @@ async function executeWorkspaceScript(config, context2, failureMode = "fatal") {
   });
   const timeoutMs = config.timeout_ms ?? (failureMode === "fatal" ? 6e4 : 3e4);
   const cwd = config.cwd ?? context2.evalDir;
+  if (config.script !== void 0 && config.command === void 0) {
+    console.warn(
+      "\x1B[33mWarning: 'script' is deprecated in workspace config. Use 'command' instead.\x1B[0m"
+    );
+  }
   const rawCommand = config.command ?? config.script ?? [];
   const commandArray = interpolateArgs(rawCommand, context2);
   const result = await execFileWithStdin(commandArray, stdin, {
@@ -17937,6 +17959,7 @@ async function invokeProvider(provider, options) {
     signal.addEventListener("abort", () => controller.abort(), { once: true });
   }
   try {
+    const braintrustSpanIds = streamCallbacks?.getActiveSpanIds?.() ?? void 0;
     return await provider.invoke({
       question: promptInputs.question,
       guidelines: promptInputs.guidelines,
@@ -17952,7 +17975,8 @@ async function invokeProvider(provider, options) {
       cwd,
       workspaceFile,
       captureFileChanges: captureFileChanges2,
-      streamCallbacks
+      streamCallbacks,
+      braintrustSpanIds: braintrustSpanIds ?? void 0
     });
   } finally {
     if (timeout !== void 0) {
@@ -18507,9 +18531,16 @@ var OTEL_BACKEND_PRESETS = {
   braintrust: {
     name: "braintrust",
     endpoint: "https://api.braintrust.dev/otel/v1/traces",
-    headers: (env) => ({
-      Authorization: `Bearer ${env.BRAINTRUST_API_KEY ?? ""}`
-    })
+    headers: (env) => {
+      const headers = {
+        Authorization: `Bearer ${env.BRAINTRUST_API_KEY ?? ""}`
+      };
+      const parent = env.BRAINTRUST_PARENT ?? (env.BRAINTRUST_PROJECT_ID ? `project_id:${env.BRAINTRUST_PROJECT_ID}` : void 0) ?? (env.BRAINTRUST_PROJECT ? `project_name:${env.BRAINTRUST_PROJECT}` : void 0);
+      if (parent) {
+        headers["x-bt-parent"] = parent;
+      }
+      return headers;
+    }
   },
   confident: {
     name: "confident",
@@ -18857,11 +18888,23 @@ var OtelStreamingObserver = class {
     this.rootSpan = null;
     this.rootCtx = null;
   }
+  /** Return the active eval span's trace ID and span ID for Braintrust trace bridging */
+  getActiveSpanIds() {
+    if (!this.rootSpan) return null;
+    try {
+      const spanCtx = this.rootSpan.spanContext?.() ?? this.rootSpan._spanContext;
+      if (!spanCtx?.traceId || !spanCtx?.spanId) return null;
+      return { parentSpanId: spanCtx.spanId, rootSpanId: spanCtx.traceId };
+    } catch {
+      return null;
+    }
+  }
   /** Get ProviderStreamCallbacks for passing to providers */
   getStreamCallbacks() {
     return {
       onToolCallEnd: (name, input, output, durationMs, toolCallId) => this.onToolCall(name, input, output, durationMs, toolCallId),
-      onLlmCallEnd: (model, tokenUsage) => this.onLlmCall(model, tokenUsage)
+      onLlmCallEnd: (model, tokenUsage) => this.onLlmCall(model, tokenUsage),
+      getActiveSpanIds: () => this.getActiveSpanIds()
     };
   }
 };