feat: Add hosted evals options.

[d42me]

Note

Introduces a platform-hosted evaluation flow alongside local runs, with CLI ergonomics and async polling/log streaming.

• New --hosted mode in prime eval run and deprecated prime env eval; requires owner/name slug and resolves environment ID from hub
• Adds hosted run options: --poll-interval, --no-stream-logs, --timeout-minutes, --allow-sandbox-access, --allow-instances-access, --custom-secrets, --eval-name
• Implements utils/hosted_eval.py providing HostedEvalConfig, HostedEvalResult, run_hosted_evaluation (create via /hosted-evaluations, poll /evaluations/{id}, stream logs, fetch final stats) and print_hosted_result
• Updates env.run_eval to parse JSON args (--env-args, --custom-secrets), create hosted config, run via asyncio, print results, and treat non-COMPLETED as failure; retains existing local eval path and installation behavior

^{Written by Cursor Bugbot for commit 1d4747c. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[cursor]

Missing validation for negative poll interval causes crash

Low Severity

The poll_interval parameter is passed directly to asyncio.sleep() without validation. If a user provides a negative value via --poll-interval, Python raises ValueError: sleep length must be non-negative, resulting in a traceback rather than a graceful CLI error. The parameter is exposed as a CLI option in evals.py without any bounds checking.

Additional Locations (1)

• packages/prime/src/prime_cli/commands/evals.py#L648-L652