Coherence-Aware Reinforcement Learning and bidirectional local agent workflows.
Use CARL to observe coherence, validate training configuration, train and evaluate models, run CARL skills, or delegate bounded work to Codex, Claude Code, and OpenCode.
