feat(framework): show tool call cwd on Braintrust tool spans

apps/framework/scripts/upload-braintrust.test.ts

@@ -7,6 +7,7 @@ import {
   unwrapShell,
   type SpanSink,
   tokenMetrics,
+  transcriptSchema,
   utcStamp,
 } from './upload-braintrust.js';
 
@@ -333,6 +334,54 @@ describe('logTranscript', () => {
   });
 });
 
+describe('tool span cwd', () => {
+  const toolMetadata = (transcript: unknown) => {
+    const metadata: unknown[] = [];
+    const sink = (): SpanSink => ({
+      startSpan: sink,
+      log(event) {
+        if (event.metadata && 'tool_name' in event.metadata) {
+          metadata.push(event.metadata);
+        }
+      },
+      end() {},
+    });
+    logTranscript(sink(), {
+      prompt: 'go',
+      agentReport: '',
+      checks: [],
+      judgeCalls: [],
+      passed: true,
+      modelId: 'm',
+      startTime: 100,
+      endTime: 120,
+      agentEndTime: 110,
+      scoringEndTime: 120,
+      toolLabels: [],
+      transcript: transcriptSchema.parse(transcript),
+    });
+    return metadata;
+  };
+
+  it('puts a call cwd in tool span metadata and omits it otherwise', () => {
+    expect(
+      toolMetadata([
+        { type: 'tool_call', name: 'Bash', ts: 101_000, cwd: '/work/a' },
+        { type: 'tool_call', name: 'Bash', ts: 102_000 },
+      ])
+    ).toEqual([{ tool_name: 'Bash', cwd: '/work/a' }, { tool_name: 'Bash' }]);
+  });
+
+  it('parses transcript parts with and without cwd', () => {
+    const parsed = transcriptSchema.parse([
+      { type: 'tool_call', name: 'Bash', input: {}, cwd: '/work/a' },
+      { type: 'tool_call', name: 'Bash', input: {} },
+    ]);
+    expect(parsed[0]).toMatchObject({ cwd: '/work/a' });
+    expect(parsed[1]).not.toHaveProperty('cwd');
+  });
+});
+
 describe('judge spans', () => {
   it('nests judge calls under the score span', () => {
     const spans: Record<string, unknown>[] = [];

apps/framework/scripts/upload-braintrust.ts

@@ -61,11 +61,12 @@ const transcriptPartSchema = z.discriminatedUnion('type', [
     ts: z.number().optional(),
     resultTs: z.number().optional(),
     id: z.string().optional(),
+    cwd: z.string().optional(),
     requestId: z.string().optional(),
     usage: requestUsageSchema.optional(),
   }),
 ]);
-const transcriptSchema = z.array(transcriptPartSchema).catch([]);
+export const transcriptSchema = z.array(transcriptPartSchema).catch([]);
 const judgeCallsSchema = z.array(judgeCallSchema).catch([]);
 type TranscriptPart = z.infer<typeof transcriptPartSchema>;
 
@@ -595,7 +596,10 @@ export function logTranscript(
       input: part.input,
       // The span name adds a label, so keep the raw name filterable.
       // https://github.com/braintrustdata/braintrust-spec/blob/b068e39112e081e45b6070e035877f1e2e83f9b7/skills/instrumentation-spec/references/features/skill-load-metadata.md
-      metadata: { tool_name: part.name },
+      metadata: {
+        tool_name: part.name,
+        ...(part.cwd ? { cwd: part.cwd } : {}),
+      },
       ...(part.output !== undefined ? { output: part.output } : {}),
       ...(part.error ? { error: part.error } : {}),
     });

packages/core/src/index.ts

@@ -184,6 +184,8 @@ export type TranscriptPart =
       resultTs?: number;
       /** Correlates the call with its result, as `TranscriptEvent.tool.id`. */
       id?: string;
+      /** Absolute directory the call ran in, when the agent records one per call. */
+      cwd?: string;
       requestId?: string;
       usage?: RequestUsage;
     };

packages/core/src/parsers/adapt.test.ts

@@ -33,4 +33,12 @@ describe('adaptTranscript', () => {
     expect(second.cwd).toBeUndefined();
     expect(second.resultTs).toBeUndefined();
   });
+
+  it("copies a call's cwd onto its transcript part only when present", () => {
+    const parts = adaptTranscript(events).transcript.filter(
+      (part) => part.type === 'tool_call'
+    );
+    expect(parts[0]).toMatchObject({ cwd: '/work/client-a' });
+    expect(parts[1]).not.toHaveProperty('cwd');
+  });
 });

packages/core/src/parsers/adapt.ts

@@ -79,6 +79,7 @@ export function adaptTranscript(events: TranscriptEvent[]): AdaptedTranscript {
         ...timed(event.timestamp),
         ...(resolved?.ts ? { resultTs: resolved.ts } : {}),
         ...(event.tool.id ? { id: event.tool.id } : {}),
+        ...(event.tool.cwd ? { cwd: event.tool.cwd } : {}),
         ...(event.requestId ? { requestId: event.requestId } : {}),
         ...(event.usage ? { usage: event.usage } : {}),
       });