Add configurable reasoning effort for GPT-5.6 and Claude Opus

[tdpham2]

Summary

• Add --reasoning-effort support to the ChemGraph CLI and TOML configuration flow.
• Support Argo GPT-5.6 Luna, Sol, and Terra plus Claude Opus 4.8 and Opus 5 with model-specific validation.
• Use medium as the default reasoning effort for both model families while preserving explicit user selections.
• Forward Claude effort through Anthropic output_config.effort without explicitly enabling adaptive thinking.
• Preserve compatible effort values across interactive model switches and fall back to the target model default when needed.
• Document supported values, defaults, and CLI usage.

Related issues

None.

Type of change

• Bug fix
• New feature
• Docs
• Chore / refactor / CI

How was this tested?

• git ls-files -z -- "*.py" | xargs -0 ruff check
• pytest tests/test_llm_agent.py tests/test_loader_pr2.py tests/test_main_agent_cli.py -q: 88 passed
• Clean-checkout-equivalent pytest tests/ -k "not tblite": 711 passed, 30 skipped, 2 deselected
• git diff --check

Checklist

• Branched off the latest main and targets main
• PR is focused on a single logical change
• ruff check . passes in a clean checkout
• pytest tests/ -k "not tblite" passes in a clean checkout
• Added/updated tests for the change
• Updated docs for the user-facing change

[tdpham2]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e0a0d39cca

ℹ️ About Codex in GitHub

Your team has set up Codex to 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 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Document the actual medium reasoning defaults

When users instantiate ChemGraph without reasoning_effort, _resolve_reasoning_effort now selects medium for both GPT-5.6 and Claude Opus models, but this public constructor documentation still advertises none and high. This can cause users to misinterpret or incorrectly reproduce an experiment; update the docstring to state that both families default to medium.

Useful? React with 👍 / 👎.

[keceli]

@codex address the feedback

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.