@agentv/core 2.18.3 → 2.19.0

package/dist/index.d.cts

@@ -14,6 +14,11 @@ interface ProviderStreamCallbacks {
     onToolCallStart?: (toolName: string, toolCallId?: string) => void;
     onToolCallEnd?: (toolName: string, input: unknown, output: unknown, durationMs: number, toolCallId?: string) => void;
     onLlmCallEnd?: (model: string, tokenUsage?: ProviderTokenUsage) => void;
+    /** Returns active OTel span IDs for Braintrust trace bridging (optional) */
+    getActiveSpanIds?: () => {
+        parentSpanId: string;
+        rootSpanId: string;
+    } | null;
 }
 interface ProviderRequest {
     readonly question: string;
@@ -36,6 +41,11 @@ interface ProviderRequest {
     readonly captureFileChanges?: boolean;
     /** Real-time observability callbacks (optional) */
     readonly streamCallbacks?: ProviderStreamCallbacks;
+    /** Braintrust span IDs for trace-claude-code plugin (optional) */
+    readonly braintrustSpanIds?: {
+        readonly parentSpanId: string;
+        readonly rootSpanId: string;
+    };
 }
 /**
  * A tool call within an output message.
@@ -3207,6 +3217,11 @@ declare class OtelStreamingObserver {
     onLlmCall(model: string, tokenUsage?: ProviderTokenUsage): void;
     /** Finalize root span with score/verdict after evaluation completes */
     finalizeEvalCase(score: number, error?: string): void;
+    /** Return the active eval span's trace ID and span ID for Braintrust trace bridging */
+    getActiveSpanIds(): {
+        parentSpanId: string;
+        rootSpanId: string;
+    } | null;
     /** Get ProviderStreamCallbacks for passing to providers */
     getStreamCallbacks(): ProviderStreamCallbacks;
 }

package/dist/index.d.ts

@@ -14,6 +14,11 @@ interface ProviderStreamCallbacks {
     onToolCallStart?: (toolName: string, toolCallId?: string) => void;
     onToolCallEnd?: (toolName: string, input: unknown, output: unknown, durationMs: number, toolCallId?: string) => void;
     onLlmCallEnd?: (model: string, tokenUsage?: ProviderTokenUsage) => void;
+    /** Returns active OTel span IDs for Braintrust trace bridging (optional) */
+    getActiveSpanIds?: () => {
+        parentSpanId: string;
+        rootSpanId: string;
+    } | null;
 }
 interface ProviderRequest {
     readonly question: string;
@@ -36,6 +41,11 @@ interface ProviderRequest {
     readonly captureFileChanges?: boolean;
     /** Real-time observability callbacks (optional) */
     readonly streamCallbacks?: ProviderStreamCallbacks;
+    /** Braintrust span IDs for trace-claude-code plugin (optional) */
+    readonly braintrustSpanIds?: {
+        readonly parentSpanId: string;
+        readonly rootSpanId: string;
+    };
 }
 /**
  * A tool call within an output message.
@@ -3207,6 +3217,11 @@ declare class OtelStreamingObserver {
     onLlmCall(model: string, tokenUsage?: ProviderTokenUsage): void;
     /** Finalize root span with score/verdict after evaluation completes */
     finalizeEvalCase(score: number, error?: string): void;
+    /** Return the active eval span's trace ID and span ID for Braintrust trace bridging */
+    getActiveSpanIds(): {
+        parentSpanId: string;
+        rootSpanId: string;
+    } | null;
     /** Get ProviderStreamCallbacks for passing to providers */
     getStreamCallbacks(): ProviderStreamCallbacks;
 }

package/dist/index.js

@@ -17,7 +17,7 @@ import {
   readTextFile,
   resolveFileReference,
   resolveTargetDefinition
-} from "./chunk-V42NUK73.js";
+} from "./chunk-ACTIPQZ3.js";
 import {
   OtlpJsonFileExporter
 } from "./chunk-HFSYZHGF.js";
@@ -783,6 +783,11 @@ 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();
@@ -1541,6 +1546,11 @@ 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}'`
@@ -3071,6 +3081,9 @@ 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");
@@ -3811,7 +3824,7 @@ 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;
@@ -4115,10 +4128,14 @@ 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() {
@@ -13395,6 +13412,11 @@ async function executeWorkspaceScript(config, context, failureMode = "fatal") {
   });
   const timeoutMs = config.timeout_ms ?? (failureMode === "fatal" ? 6e4 : 3e4);
   const cwd = config.cwd ?? context.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, context);
   const result = await execFileWithStdin(commandArray, stdin, {
@@ -15065,6 +15087,7 @@ 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,
@@ -15080,7 +15103,8 @@ async function invokeProvider(provider, options) {
       cwd,
       workspaceFile,
       captureFileChanges: captureFileChanges2,
-      streamCallbacks
+      streamCallbacks,
+      braintrustSpanIds: braintrustSpanIds ?? void 0
     });
   } finally {
     if (timeout !== void 0) {
@@ -15635,9 +15659,16 @@ 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",
@@ -15985,11 +16016,23 @@ 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()
     };
   }
 };