Skill UI

Coordinate multiple AI agents concurrently via dmux’s tmux pane manager, running Claude Code, Codex, OpenCode and other tools in parallel tracks for research, implementation, testing, review, or multi-role coding tasks.

Autonomously orchestrates multi-round research reviews via Codex MCP: review, implement fixes, and re-review until the reviewer declares the work ready or the max rounds expire. Logs progress, maintains compact/recoverable state, and optionally adds human checkpoints for iterative improvement requests.

Provides a Codex-powered workflow that orchestrates multi-round, high-reasoning GPT research reviews, capturing context, criticisms, follow-ups, experiments, and documentation.

Agent Eval Benchmarking orchestrates repeatable CLI comparisons of coding agents like Claude Code, Aider, and Codex, collecting pass rate, cost, time, and consistency metrics on your repository tasks so you can pick the right assistant with data.

Analyze diffs for reuse, clarity, and efficiency issues, coordinating parallel Codex sub-agents before optionally applying behavior-preserving fixes that clean up scoped changes safely.

Guides teams through analyzing experimental results to assess claim support, capturing comparisons, Codex judgment, and routing next actions (pivot/supplement/confirm) before writing papers or running ablations.