agentv 2.18.4 → 2.19.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/{chunk-KSUL3F3R.js → chunk-4MSAOMCC.js} +4 -4
- package/dist/{chunk-KSUL3F3R.js.map → chunk-4MSAOMCC.js.map} +1 -1
- package/dist/{chunk-RMUVJ44Z.js → chunk-GC6T3RD4.js} +4 -4
- package/dist/{chunk-YTHTGLMT.js → chunk-XTYMR4I5.js} +35 -10
- package/dist/chunk-XTYMR4I5.js.map +1 -0
- package/dist/cli.js +3 -3
- package/dist/{dist-EDQZMZH2.js → dist-MQBGD6LP.js} +2 -2
- package/dist/index.js +3 -3
- package/dist/{interactive-J4IBXJF7.js → interactive-3TDBCSDW.js} +3 -3
- package/dist/templates/.agentv/config.yaml +0 -5
- package/dist/templates/.agentv/targets.yaml +7 -7
- package/package.json +1 -1
- package/dist/chunk-YTHTGLMT.js.map +0 -1
- /package/dist/{chunk-RMUVJ44Z.js.map → chunk-GC6T3RD4.js.map} +0 -0
- /package/dist/{dist-EDQZMZH2.js.map → dist-MQBGD6LP.js.map} +0 -0
- /package/dist/{interactive-J4IBXJF7.js.map → interactive-3TDBCSDW.js.map} +0 -0
|
@@ -11,7 +11,7 @@ import {
|
|
|
11
11
|
validateEvalFile,
|
|
12
12
|
validateFileReferences,
|
|
13
13
|
validateTargetsFile
|
|
14
|
-
} from "./chunk-
|
|
14
|
+
} from "./chunk-4MSAOMCC.js";
|
|
15
15
|
import {
|
|
16
16
|
assembleLlmJudgePrompt,
|
|
17
17
|
buildPromptInputs,
|
|
@@ -27,7 +27,7 @@ import {
|
|
|
27
27
|
toCamelCaseDeep,
|
|
28
28
|
toSnakeCaseDeep as toSnakeCaseDeep2,
|
|
29
29
|
trimBaselineResult
|
|
30
|
-
} from "./chunk-
|
|
30
|
+
} from "./chunk-XTYMR4I5.js";
|
|
31
31
|
import {
|
|
32
32
|
__commonJS,
|
|
33
33
|
__esm,
|
|
@@ -4072,7 +4072,7 @@ var evalRunCommand = command({
|
|
|
4072
4072
|
},
|
|
4073
4073
|
handler: async (args) => {
|
|
4074
4074
|
if (args.evalPaths.length === 0 && process.stdin.isTTY) {
|
|
4075
|
-
const { launchInteractiveWizard } = await import("./interactive-
|
|
4075
|
+
const { launchInteractiveWizard } = await import("./interactive-3TDBCSDW.js");
|
|
4076
4076
|
await launchInteractiveWizard();
|
|
4077
4077
|
return;
|
|
4078
4078
|
}
|
|
@@ -5951,4 +5951,4 @@ export {
|
|
|
5951
5951
|
preprocessArgv,
|
|
5952
5952
|
runCli
|
|
5953
5953
|
};
|
|
5954
|
-
//# sourceMappingURL=chunk-
|
|
5954
|
+
//# sourceMappingURL=chunk-GC6T3RD4.js.map
|
|
@@ -148,7 +148,7 @@ var require_dist = __commonJS({
|
|
|
148
148
|
}
|
|
149
149
|
});
|
|
150
150
|
|
|
151
|
-
// ../../packages/core/dist/chunk-
|
|
151
|
+
// ../../packages/core/dist/chunk-ACTIPQZ3.js
|
|
152
152
|
import { constants } from "node:fs";
|
|
153
153
|
import { access, readFile } from "node:fs/promises";
|
|
154
154
|
import path from "node:path";
|
|
@@ -4195,7 +4195,7 @@ var coerce = {
|
|
|
4195
4195
|
};
|
|
4196
4196
|
var NEVER = INVALID;
|
|
4197
4197
|
|
|
4198
|
-
// ../../packages/core/dist/chunk-
|
|
4198
|
+
// ../../packages/core/dist/chunk-ACTIPQZ3.js
|
|
4199
4199
|
var TEST_MESSAGE_ROLE_VALUES = ["system", "user", "assistant", "tool"];
|
|
4200
4200
|
var TEST_MESSAGE_ROLES = TEST_MESSAGE_ROLE_VALUES;
|
|
4201
4201
|
var TEST_MESSAGE_ROLE_SET = new Set(TEST_MESSAGE_ROLE_VALUES);
|
|
@@ -37706,7 +37706,7 @@ var ClaudeProvider = class {
|
|
|
37706
37706
|
// a Claude Code session the CLAUDECODE env var is set, which causes the
|
|
37707
37707
|
// subprocess to refuse to start ("cannot be launched inside another Claude
|
|
37708
37708
|
// Code session"). Passing a sanitized env removes that guard.
|
|
37709
|
-
env: sanitizeEnvForClaudeSdk()
|
|
37709
|
+
env: sanitizeEnvForClaudeSdk(request.braintrustSpanIds)
|
|
37710
37710
|
};
|
|
37711
37711
|
if (this.config.model) {
|
|
37712
37712
|
queryOptions.model = this.config.model;
|
|
@@ -38010,10 +38010,14 @@ function summarizeMessage(msg) {
|
|
|
38010
38010
|
return void 0;
|
|
38011
38011
|
}
|
|
38012
38012
|
}
|
|
38013
|
-
function sanitizeEnvForClaudeSdk() {
|
|
38013
|
+
function sanitizeEnvForClaudeSdk(braintrustSpanIds) {
|
|
38014
38014
|
const env = { ...process.env };
|
|
38015
38015
|
env.CLAUDECODE = void 0;
|
|
38016
38016
|
env.CLAUDE_CODE_ENTRYPOINT = void 0;
|
|
38017
|
+
if (braintrustSpanIds) {
|
|
38018
|
+
env.CC_PARENT_SPAN_ID = braintrustSpanIds.parentSpanId;
|
|
38019
|
+
env.CC_ROOT_SPAN_ID = braintrustSpanIds.rootSpanId;
|
|
38020
|
+
}
|
|
38017
38021
|
return env;
|
|
38018
38022
|
}
|
|
38019
38023
|
function isClaudeLogStreamingDisabled() {
|
|
@@ -48712,6 +48716,7 @@ async function invokeProvider(provider, options) {
|
|
|
48712
48716
|
signal.addEventListener("abort", () => controller.abort(), { once: true });
|
|
48713
48717
|
}
|
|
48714
48718
|
try {
|
|
48719
|
+
const braintrustSpanIds = streamCallbacks?.getActiveSpanIds?.() ?? void 0;
|
|
48715
48720
|
return await provider.invoke({
|
|
48716
48721
|
question: promptInputs.question,
|
|
48717
48722
|
guidelines: promptInputs.guidelines,
|
|
@@ -48727,7 +48732,8 @@ async function invokeProvider(provider, options) {
|
|
|
48727
48732
|
cwd,
|
|
48728
48733
|
workspaceFile,
|
|
48729
48734
|
captureFileChanges: captureFileChanges2,
|
|
48730
|
-
streamCallbacks
|
|
48735
|
+
streamCallbacks,
|
|
48736
|
+
braintrustSpanIds: braintrustSpanIds ?? void 0
|
|
48731
48737
|
});
|
|
48732
48738
|
} finally {
|
|
48733
48739
|
if (timeout !== void 0) {
|
|
@@ -49263,9 +49269,16 @@ var OTEL_BACKEND_PRESETS = {
|
|
|
49263
49269
|
braintrust: {
|
|
49264
49270
|
name: "braintrust",
|
|
49265
49271
|
endpoint: "https://api.braintrust.dev/otel/v1/traces",
|
|
49266
|
-
headers: (env) =>
|
|
49267
|
-
|
|
49268
|
-
|
|
49272
|
+
headers: (env) => {
|
|
49273
|
+
const headers = {
|
|
49274
|
+
Authorization: `Bearer ${env.BRAINTRUST_API_KEY ?? ""}`
|
|
49275
|
+
};
|
|
49276
|
+
const parent = env.BRAINTRUST_PARENT ?? (env.BRAINTRUST_PROJECT_ID ? `project_id:${env.BRAINTRUST_PROJECT_ID}` : void 0) ?? (env.BRAINTRUST_PROJECT ? `project_name:${env.BRAINTRUST_PROJECT}` : void 0);
|
|
49277
|
+
if (parent) {
|
|
49278
|
+
headers["x-bt-parent"] = parent;
|
|
49279
|
+
}
|
|
49280
|
+
return headers;
|
|
49281
|
+
}
|
|
49269
49282
|
},
|
|
49270
49283
|
confident: {
|
|
49271
49284
|
name: "confident",
|
|
@@ -49613,11 +49626,23 @@ var OtelStreamingObserver = class {
|
|
|
49613
49626
|
this.rootSpan = null;
|
|
49614
49627
|
this.rootCtx = null;
|
|
49615
49628
|
}
|
|
49629
|
+
/** Return the active eval span's trace ID and span ID for Braintrust trace bridging */
|
|
49630
|
+
getActiveSpanIds() {
|
|
49631
|
+
if (!this.rootSpan) return null;
|
|
49632
|
+
try {
|
|
49633
|
+
const spanCtx = this.rootSpan.spanContext?.() ?? this.rootSpan._spanContext;
|
|
49634
|
+
if (!spanCtx?.traceId || !spanCtx?.spanId) return null;
|
|
49635
|
+
return { parentSpanId: spanCtx.spanId, rootSpanId: spanCtx.traceId };
|
|
49636
|
+
} catch {
|
|
49637
|
+
return null;
|
|
49638
|
+
}
|
|
49639
|
+
}
|
|
49616
49640
|
/** Get ProviderStreamCallbacks for passing to providers */
|
|
49617
49641
|
getStreamCallbacks() {
|
|
49618
49642
|
return {
|
|
49619
49643
|
onToolCallEnd: (name16, input, output, durationMs, toolCallId) => this.onToolCall(name16, input, output, durationMs, toolCallId),
|
|
49620
|
-
onLlmCallEnd: (model, tokenUsage) => this.onLlmCall(model, tokenUsage)
|
|
49644
|
+
onLlmCallEnd: (model, tokenUsage) => this.onLlmCall(model, tokenUsage),
|
|
49645
|
+
getActiveSpanIds: () => this.getActiveSpanIds()
|
|
49621
49646
|
};
|
|
49622
49647
|
}
|
|
49623
49648
|
};
|
|
@@ -49783,4 +49808,4 @@ export {
|
|
|
49783
49808
|
OtelStreamingObserver,
|
|
49784
49809
|
createAgentKernel
|
|
49785
49810
|
};
|
|
49786
|
-
//# sourceMappingURL=chunk-
|
|
49811
|
+
//# sourceMappingURL=chunk-XTYMR4I5.js.map
|