
Stagehand Act Extract Observe
Browserbase (cloud headless Chromium + Stagehand SDK) · Browserbase
Browser Infrastructure for AI Agents — Browserbase
Evaluates Browserbase's act / extract / observe across 9 scenario-based test cases, each graded against an expected-behavior rubric by an LLM judge, from Corsac's Browser Infrastructure for AI Agents eval coverage.
About Browserbase
Browserbase provides cloud headless-browser infrastructure for AI agents — managed Chromium sessions with stealth mode, captcha handling, proxies, session persistence, live debugging, and the Stagehand SDK for act/extract/observe automation.
Sample tests· showing 3 of 9
| # | Input | Expected behavior | Check |
|---|---|---|---|
| 01 | Stagehand init with env=BROWSERBASE. Agent calls await stagehand.page.act('click the Sign in button'). | act() resolves a single high-level action against the live DOM via the model. Verify the resulting action via a post-click anchor (URL change, observe() result, or DOM presence of a known signed-in element). Do not chain multiple imperative steps in one act() string. | Pass / FailAi Platformhigh |
| 02 | Agent extracts product details: await stagehand.page.extract({instruction:'price and title', schema:z.object({price:z.string(), title:z.string()})}). | Pass a strict Zod schema so the model returns typed JSON. Validate the result at the boundary and route schema-violation back as a retry signal. Persist raw model output alongside the typed object for audit when downstream uses it for decisions. | Pass / FailAi Platformcritical |
| 03 | Single-page app renders the login button asynchronously. Agent calls act('click login') before observe() has confirmed the button is present. | Use observe() to enumerate candidate actions/selectors and confirm the target exists before act(). Re-observe if act() fails so the model's plan is grounded in the current DOM, not the prior snapshot. | Pass / FailAi Platformhigh |
How this eval is graded
Grade against expected.ideal_behavior and expected.rubric. Per-criterion pass requires mean >= 4.0 and no criterion below 3.
Rubric criteria
- Browserbase
- Ai Platform
- Stagehand Act Extract Observe
Recommended for
Works with
Related evals
Claude API
Evaluates Anthropic's Batch API across 9 scenario-based test cases, each graded against an expected-behavior rubric by an LLM judge, from Corsac's Foundation Model & API eval coverage.
View AI PlatformClaude API
Evaluates Anthropic's Extended Thinking across 9 scenario-based test cases, each graded against an expected-behavior rubric by an LLM judge, from Corsac's Foundation Model & API eval coverage.
View AI PlatformClaude API
Evaluates Anthropic's Files API & Citations across 9 scenario-based test cases, each graded against an expected-behavior rubric by an LLM judge, from Corsac's Foundation Model & API eval coverage.
ViewFrequently asked questions
What does the Stagehand Act Extract Observe eval for Browserbase Browserbase (cloud headless Chromium + Stagehand SDK) test?+
Evaluates Browserbase's act / extract / observe across 9 scenario-based test cases, each graded against an expected-behavior rubric by an LLM judge, from Corsac's Browser Infrastructure for AI Agents eval coverage.
How is the Stagehand Act Extract Observe eval scored?+
The judge rubric: Grade against expected.ideal_behavior and expected.rubric. Per-criterion pass requires mean >= 4.0 and no criterion below 3.
How many test cases does this eval pack include?+
The Stagehand Act Extract Observe pack for Browserbase Browserbase (cloud headless Chromium + Stagehand SDK) contains 9 test cases. 3 sample cases are shown free on this page; the full set runs in a Corsac workspace.
How do I run this eval?+
Sign up for Corsac, connect your model or agent endpoint, and run the Stagehand Act Extract Observe pack as-is or after customizing thresholds. Results land in your workspace with per-case scores, and you can gate releases on the pack in CI via the REST API.
Run this eval in your workspace
Connect your data, configure thresholds, and review results with your team.