Benchmark report
Task verified
5m 28s elapsed
The saved attempt met the documented success criterion.
100/ 100
Agent-Ready- SInstall
- SAuth
- SExecute
- SDocs
- SCurrent
Only hiccup: https://anchorbrowser.io/auth.md was unreachable (proxy CONNECT 403), so auth details came from the OpenAPI security scheme instead.
Where the time went
5:28 agent time- Starting0:01
- Research0:29
- Attempt3:26
- Verify1:32
Judge reason
Saved code (code/anchor_session_lifecycle.py, byte-identical to the heredoc in transcript line 50) runs the four documented REST operations with Python stdlib urllib and the key only in an anchor-api-key header. Create POST /v1/sessions with body {"session":{"timeout":{"max_duration":5,"idle_timeout":2}}} returned HTTP 200 and data.id 48fddbf5-cd99-4f33-a259-80b3cd481264 (evidence/01). GET /v1/sessions/{id} returned HTTP 200 status 'running' with the server echoing max_duration 5 / globalTimeout 5 (evidence/02). GET /v1/sessions/all/status returned the same session_id (evidence/03). DELETE returned HTTP 200 {"data":{"status":"success"}} (evidence/04) and the final GET showed status 'completed' (evidence/05); steps 4-5 sit in the script's finally: block. Printed values in evidence/run_console_output.txt match the saved JSON bodies exactly. No Playwright/CDP client was used; cdp_url and live_view_url were never opened. grep for the key across evidence/code/docs reported CLEAN (transcript 62, 90); the only key-like string saved is the server's own masked 'sk-60ab...d816'. Endpoints and the anchor-api-key scheme are corroborated by the saved docs/api-reference pages and docs/advanced_session-timeout.md.
Attempt record
- 1:26Answered from saved key: Anchor Browser API keyno wait
- —First success verified
Attempt artifacts saved.