The Prompting Company

Benchmark report

Task verified

5m 24s elapsed

The saved attempt met the documented success criterion.

Experience score

95/ 100

Agent-Ready
  • SInstall
  • SAuth
  • SExecute
  • ADocs
  • ACurrent
Experience measures friction in the tested task. The verdict says whether it succeeded.
Agent profile for this run: Install S, Auth S, Execute S, Docs A, Current ASInstallSAuthSExecuteADocsACurrent
Agent profile · this run
Judge's note · what slowed the agent most
Docs never show the API key as a header on the CDP connect (only apiKey in the URL), so the agent crawled 182 doc pages before writing a header-based workaround.

Where the time went

5:24 agent time
  • Starting0:01
  • Research0:22
  • Attempt3:32
  • Verify1:30
5m 24sTotal elapsed
2:47Time to first success
0:00Waited on the owner

Judge reason

Saved code code/anchor_session.mjs calls the documented POST https://api.anchorbrowser.io/v1/sessions with the anchor-api-key header; evidence/create-session-response.json shows HTTP 200 with id 2ea1a3a3-476d-4e54-90cf-38c25564652f and cdp_url wss://connect.anchorbrowser.io?sessionId=...&rt=c51d072e. Playwright (playwright-core, npm install exit 0) connected via chromium.connectOverCDP to that URL with the key stripped from the query string and sent only in the anchor-api-key header; run-output.txt (transcript entry 94, EXIT=0) prints 'Page title: Example Domain' and page-title.txt records url=https://example.com/, title=Example Domain. The finally block issued DELETE /v1/sessions/{id}; end-session-response.json shows http_status 200 OK with {"data":{"status":"success"}}. Printed values match saved responses; all saved files show REDACTED_API_KEY and an in-transcript grep for the key across home and work dirs reported 'no credential in saved files'. Real network values (live UUID, rt token) rule out mocking.

Attempt record

  1. 1:19Answered from saved key: Anchor Browser API keyno wait
  2. —First success verified

Attempt artifacts saved.