moai-workflow-research

Automate iterative self-research loops by running binary evaluation experiments on moai-adk components like skills, agents, and rules to converge on optimized configurations. It enforces versioned workflows with frontmatter metadata and controlled data directories.

Safety governance in binary eval experiments applies a 5-layer architecture including FrozenGuard, Worktree Sandbox, Canary Regression, Rate Limiter, and Human Approval to ensure controlled, safety-aware experimentation and traceable baselines.

Yes, the workflow requires controlled data directories managed via moai CLI to store eval suites, baselines, experiments, changelogs, observations, and dashboards for traceable experiment baselines.

Yes, developers and researchers can apply this workflow to conduct safety-aware experimentation and architecture refinement inside existing moai-adk projects using versioned workflows and frontmatter-defined metadata.

Track observations and dashboards during component optimization by using predefined data paths and the moai CLI to manage versioned workflows, experiment baselines, and changelogs throughout the iterative research cycle.