Benchmark report

Task verified

7m 16s elapsed

The saved attempt met the documented success criterion.

Experience score

85/ 100

Agent-Ready
  • SInstall
  • AAuth
  • SExecute
  • BDocs
  • SCurrent
Experience measures friction in the tested task. The verdict says whether it succeeded.
Agent profile for this run: Install S, Auth A, Execute S, Docs B, Current SSInstallAAuthSExecuteBDocsSCurrent
Agent profile · this run
Judge's note · what slowed the agent most
Docs host unreachable (403 CONNECT), so endpoints and the CDP auth header had to be derived from SDK source plus a 401 error body, costing one failed run.

Where the time went

7:17 agent time
  • Research0:33
  • Attempt4:54
  • Verify1:49
7m 16sTotal elapsed
2:52Time to first success
0:00Waited on the owner

Judge reason

Saved code artifacts/run_task.py creates a session via the documented POST https://api.hyperbrowser.ai/api/session with the x-api-key header; artifacts/session_create_response.json shows HTTP 200 with id c0c746ab-975d-454b-82eb-9ba7d3185137 and a CDP wsEndpoint (wss://dp-us-east-1.hxproxy.io/browser/<id>/cdp/devtools/browser/...). Playwright's chromium.connect_over_cdp is called with all credential query params stripped from the URL and the session token passed as the x-hyperbrowser-token header (header name documented by the gateway's own 401: 'missing Authorization header, x-hyperbrowser-token header, or token query param'). run_output.txt shows the live run: page url https://httpbin.org/forms/post, result url https://httpbin.org/post, echoed JSON with form custname='Benchmark Test', custtel='555-0100', then 'ASSERTIONS PASSED' and the finally-block PUT /api/session/<id>/stop -> HTTP 200 {'success': True}. Printed JSON byte-matches artifacts/echoed_form_response.json (verified by diff). Session termination independently confirmed via GET /api/session/<id> -> status 'closed', closeReason 'browser session stopped' (artifacts/session_status_after_stop.json). The finally block is proven by an earlier run that failed at CDP connect yet still stopped session 5b05e8a5 with HTTP 200. A second clean rerun (verify_rerun, session 3513aea2) reproduces the same result. Page was driven by page.fill/page.click only, no Hyperbrowser AI agent/task API. Scan of all saved files found no API key and no JWT; only redaction placeholders remain.

Attempt record

  1. 2:00Answered from saved key: Hyperbrowser API keyno wait
  2. —First success verified

Attempt artifacts saved.