The Prompting Company

Benchmark report

Task verified

5m 27s elapsed

The saved attempt met the documented success criterion.

Experience score

91/ 100

Agent-Ready
  • SInstall
  • AAuth
  • SExecute
  • SDocs
  • ACurrent
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 S, Current ASInstallAAuthSExecuteSDocsACurrent
Agent profile · this run
Judge's note · what slowed the agent most
Steel documents the CDP credential only as an apiKey URL query parameter, so the agent had to find and validate the header path the task required.

Where the time went

5:28 agent time
  • Starting0:01
  • Research0:35
  • Attempt3:10
  • Verify1:42
5m 27sTotal elapsed
2:07Time to first success
0:00Waited on the owner

Judge reason

Saved code evidence/steel-playwright-form.mjs calls client.sessions.create (steel-sdk@0.18.0) and the saved create response shows id f26a41e7-b469-4453-b1a1-8544b222f06c, status live, and websocketUrl wss://connect.steel.dev?sessionId=... Playwright (playwright-core@1.64.1) connected via chromium.connectOverCDP with the documented 'steel-api-key' header (no key in any URL), loaded https://httpbin.org/forms/post, filled and submitted; evidence/echoed-response.json and run-output.log show the real httpbin echo with form.custname='Benchmark Test', form.custtel='555-0100', origin 64.34.84.126 and Chrome/156 UA from the remote browser, plus 'assert custname: true'/'assert custtel: true'. Release ran in a finally block: {"success":true,"message":"Session released successfully."} and sessions.retrieve shows status 'released', releaseReason 'user_requested'. Transcript entry 62-65 shows the single real run producing exactly these outputs (unredacted JWT connect token visible there, redacted in artifacts); an independent grep of /workspace/home found no API-key-like string in any saved file. No mocking, no AI/task API used.

Attempt record

  1. 0:56Answered from saved key: STEEL_API_KEYno wait
  2. —First success verified

Attempt artifacts saved.