Normalize eval enable_thinking sampling args

[peter941221]

Summary

• normalize local prime eval run --sampling-args payloads that use top-level enable_thinking
• rewrite that flag into extra_body.chat_template_kwargs.enable_thinking before handing off to verifiers
• add a regression test that checks the rewritten command-line payload

Why

Users currently follow the documented --sampling-args '{"enable_thinking": false}' form and hit AsyncCompletions.create() got an unexpected keyword argument 'enable_thinking' during local eval runs. Hosted and training paths already model enable_thinking as a first-class sampling control, so the local eval bridge should preserve that user-facing contract.

Testing

• python -m py_compile packages/prime/src/prime_cli/verifiers_bridge.py packages/prime/tests/test_eval_billing.py
• uv run ruff check packages/prime/src/prime_cli/verifiers_bridge.py packages/prime/tests/test_eval_billing.py
• uv run pytest packages/prime/tests/test_eval_billing.py -q

---

Note

Low Risk
CLI-only sampling-arg normalization and short-lived temp config files; behavior is additive for users who already pass nested extra_body, with regression tests.

Overview
Local prime eval run now accepts top-level enable_thinking in --sampling-args JSON and in eval TOML sampling_args, rewriting it to extra_body.chat_template_kwargs.enable_thinking before verifiers runs so inference no longer rejects an unknown keyword.

CLI passthrough rewrites every --sampling-args occurrence; config-driven runs write a temporary TOML when needed and delete it in a finally block. Tests cover CLI rewrite (including multiple flags), config rewrite, and temp-file cleanup on failure.

^{Reviewed by Cursor Bugbot for commit 3576156. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 6281eeb. Configure here.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6281eeb0dd

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".