NAVIER-CFD 1.1.0

NAVIER-CFD 1.1.0 introduces the first Codex-integrated AutoResearch foundation.

Main additions

• structured ResearchObjective, ResearchContract, ResearchBudget, and StopPolicy;
• persistent AutoResearchSession workspaces;
• assistant, guided, and bounded autonomy modes;
• action risk classification and approval recording;
• deterministic budget and stopping decisions;
• a local read-only MCP server;
• six repository Codex skills;
• deterministic field, worst-case, and multiphase-interface diagnostics;
• FigureSpec, figure auditing, Matplotlib renderers, and FigureManifest;
• repository scientific-integrity rules in AGENTS.md;
• project MCP configuration example;
• focused tests and CI integration.

Documentation

The release includes a complete documentation set:

• AutoResearch overview
• Architecture
• MCP tools
• Codex skills
• Contracts and sessions
• CFD diagnostics
• FigureLab

Validation

The release branch is validated through:

• Python 3.10, 3.11, and 3.12 tests;
• Ruff;
• browser recommender tests;
• native model/provider/metric/training and AutoResearch smoke tests;
• strict documentation build;
• official-source security and live upstream validation.

Compatibility

Existing dataset providers, model hub APIs, recommendation, training, checkpoints, metrics, experiments, CLI workflows, and the original AgentOrchestrator remain available.

AutoResearch composes these systems rather than replacing them.

Safety and scope

The v1.1.0 MCP tools are intentionally read-only. The release does not claim automatic training, solver execution, cluster submission, active simulation generation, or autonomous laboratory control.

Those capabilities require separate approved execution adapters.