mdflow

Why we included this project

mdflow treats repeatable coding-agent work as markdown files you run from the terminal. Write the job once as a .md file and execute it against Claude, Codex, Gemini, or Copilot, pinning the engine in the filename when you want a specific one. The CLI shows you what each run consumed and lets you pipe results through commands like git diff. Behavioral evals can gate prompt revisions before they go live, and a roster sync can hand matching tasks to coding agents through AGENTS.md and CLAUDE.md. The flow workbench keeps project, global, and PATH flows searchable in one place. One honest caveat: engine context isolation is not a real sandbox, so this is a control-plane tool for teams that want auditable, repeatable runs without building their own orchestration glue.