diff --git a/packages/evals/package.json b/packages/evals/package.json index 3b6c143ead..0e26ace598 100644 --- a/packages/evals/package.json +++ b/packages/evals/package.json @@ -40,7 +40,7 @@ "@opentelemetry/sdk-trace-node": "^2.9.0", "ai": "^5.0.133", "braintrust": "^3.21.0", - "browse": "0.9.5", + "browse": "workspace:*", "dotenv": "^17.3.1", "langsmith": "^0.5.16", "openai": "^4.104.0", diff --git a/packages/evals/tests/framework/claudeCodeToolAdapter.test.ts b/packages/evals/tests/framework/claudeCodeToolAdapter.test.ts index 5acd0d939a..c214dd8729 100644 --- a/packages/evals/tests/framework/claudeCodeToolAdapter.test.ts +++ b/packages/evals/tests/framework/claudeCodeToolAdapter.test.ts @@ -110,11 +110,14 @@ describe("claude code tool adapter resolution", () => { expect(getBrowseCliAllowedTools()).toEqual(["Skill", "Bash"]); }); - it("exposes browse cli metadata for Braintrust rows", () => { + it("exposes workspace browse cli metadata for Braintrust rows", async () => { + const { version } = JSON.parse( + await fsp.readFile(new URL("../../../cli/package.json", import.meta.url), "utf8"), + ); expect(getBrowseCliToolMetadata()).toMatchObject({ toolCommand: "browse", - browseCliVersion: "0.9.5", - browseCliEntrypoint: expect.stringMatching(/browse[/\\]bin[/\\]run\.js$/u), + browseCliVersion: version, + browseCliEntrypoint: expect.stringMatching(/packages[/\\]cli[/\\]bin[/\\]run\.js$/u), }); }); diff --git a/packages/evals/tests/tui/run.test.ts b/packages/evals/tests/tui/run.test.ts index 4a0cab5e5b..2d7f73c21c 100644 --- a/packages/evals/tests/tui/run.test.ts +++ b/packages/evals/tests/tui/run.test.ts @@ -259,7 +259,7 @@ describe("deriveCategoryFilter", () => { startupProfile: "tool_create_browserbase", toolCommand: "browse", browseCliVersion: expect.any(String), - browseCliEntrypoint: expect.stringMatching(/browse[/\\]bin[/\\]run\.js$/u), + browseCliEntrypoint: expect.stringMatching(/packages[/\\]cli[/\\]bin[/\\]run\.js$/u), harnessConfig: { harness: "claude_code", model: "anthropic/claude-sonnet-4-20250514", @@ -317,7 +317,7 @@ describe("deriveCategoryFilter", () => { startupProfile: "tool_create_browserbase", toolCommand: "browse", browseCliVersion: expect.any(String), - browseCliEntrypoint: expect.stringMatching(/browse[/\\]bin[/\\]run\.js$/u), + browseCliEntrypoint: expect.stringMatching(/packages[/\\]cli[/\\]bin[/\\]run\.js$/u), harnessConfig: { harness: "codex", model: "openai/gpt-5.4-mini", diff --git a/pnpm-lock.yaml b/pnpm-lock.yaml index c769a88398..4e74460909 100644 --- a/pnpm-lock.yaml +++ b/pnpm-lock.yaml @@ -596,8 +596,8 @@ importers: specifier: ^3.21.0 version: 3.28.0(@aws-sdk/credential-provider-web-identity@3.972.74)(zod@4.4.3) browse: - specifier: 0.9.5 - version: 0.9.5(bufferutil@4.1.0)(playwright-core@1.56.1) + specifier: workspace:* + version: link:../cli dotenv: specifier: ^17.3.1 version: 17.4.2