bmad-assist
⚠️ Deprecated & unmaintained. This tool only ever supported BMAD v6.0.0-alpha.22. Use bmad-loop instead.
bmad-assist: a deprecated AI story implementer
A CLI tool that read PRD, architecture, and epic docs and implemented them story by story with several LLMs in parallel. It is end of life, works only with an early BMAD alpha, and points to bmad-loop instead.
Read the status first
The first thing the README does is warn you. bmad-assist is deprecated and unmaintained, only ever supported BMAD v6.0.0-alpha.22, and the recommendation is to use bmad-loop instead. It was a solo project, it only works with an early alpha release of BMAD, and yet it still runs against that version.
The loop it runs
BMAD structures AI-driven projects into docs including PRD, architecture, epics, and stories. bmad-assist reads those docs and runs an automated loop. It creates the next story from the epics, validates the story using multiple LLMs in parallel, implements the story by writing code and tests, and reviews the code with multiple LLMs. The whole thing is one repeated cycle.
Why multiple LLMs
One LLM, called the Master, writes all the code while the others only validate and review and never touch the files. The reasoning is stated plainly. Validation and review benefit from multiple perspectives, while creation and implementation need a single source of truth. Different models can be configured per phase, with more powerful models for critical phases.
Platform and differences
Installation requires Python 3.11 and at least one LLM CLI tool like Claude Code, Gemini CLI, or Codex. Linux and macOS are the only supported platforms, because the subprocess management layer depends on POSIX APIs. On Windows, WSL2 is recommended, with projects kept on the Linux filesystem for correct Unix socket permissions. The README also lists differences from vanilla BMAD: no user interaction so workflows run non-interactively, context injection where the compiler embeds all needed files instead of loading them at runtime, and stdout output with reports written to stdout. Added workflows include validate-story with INVEST criteria and an evidence score, plus validation synthesis.
Editorial conclusion
Deprecated projects usually bury their history. This README is upfront about being a solo, unmaintained tool tied to one alpha release, which makes it easier to judge whether to use it at all.
Community notes