bmad-planner

The BMAD Planner is an interactive, callable Python-based agent skill designed to streamline the planning phase of software engineering projects. It acts as a multi-persona facilitator—specifically an Analyst, Architect, and Project Manager—to guide developers through a structured Q&A process. By automating the creation of comprehensive planning documents, it ensures consistency and significantly reduces token consumption during subsequent AI-assisted coding tasks. It is intended for use in the main repository, specifically on contribution branches, before any feature implementation begins.

• Conducts interactive requirements gathering, technical architecture design, and epic breakdown using persona-based questioning.

• Automatically generates a suite of planning documents including requirements.md, architecture.md, and epics.md.

• Sets up standardized project structures, including documentation for Claude Code context (CLAUDE.md) and human-readable overviews.

• Reduces token usage by approximately 92% (up to 2,700 tokens per feature) by providing pre-contextualized planning data that Claude Code can reuse throughout the lifecycle of the feature development.

• Integrates with local development workflows, including support for Git-based feature tracking and automatic commit operations.

• Prerequisites: Must be invoked from the main repository while currently operating on a contrib branch.

• Execution: Runs via the command line with parameters for the feature slug and the GitHub username to ensure proper documentation scoping.

• Output: Generates a dedicated planning directory containing all project-specific design artifacts, keeping the main repository organized.

• Workflow: This is typically the first step in the development cycle, ensuring that the feature requirements and technical constraints are well-defined and committed before a worktree is created.

• Templates: Relies on a centralized set of markdown templates that are dynamically populated with user-provided responses during the interactive session.