bmad-build Skill (BMad Method) for Claude Code: Build Flow

• Runs uv run --no-cache on _bmad/scripts/render_skill.py

  Run once from the project root with an optional --set workflow.route (oneshot or full) and --set workflow.review (none, quick or thorough). If the script is missing, Claude offers to set BMad up, and on any other failure it reports the output and stops.

  Try: “bmad-build fix the pagination bug in the orders list, skip review”

  Runs code · Render the workflow

• Resolve the ticket or intent and check the git tree

  Looks for an existing plan or ticket, ingests your request, and checks that the working tree is clean and the branch makes sense, halting to ask if not.

  Read · Step 1: Clarify and route

• Write a plan file with acceptance criteria and a risk score

  Writes the plan from a template with status, risk (low, medium or high), a code map and Given/When/Then acceptance criteria, and stops for your approval on full runs.

  Write · Step 2: Plan

• Capture the baseline commit and hand off to a subagent

  Records baseline_revision, sets the plan to in-progress, launches a no-context coding subagent that reads the plan, then writes a diff and runs the commands in the plan’s Verification section. SKILL rules say no push and no remote operations.

  Runs code · Step 3: Implement

• One quick lens, or four thorough lenses, then triage

  Quick runs one reviewer. Thorough runs Blind Hunter, Edge Case Hunter, Verification Gap Reviewer and Intent Alignment Auditor. Findings are logged in the plan and patched or deferred. Reviewers inspect only and do not modify code.

  Write · Step 4: Review

• Mark the plan built and make a local commit

  If version control is available and the tree is dirty, it creates a local commit with a conventional message from the plan title. The rule is never to auto-push.

  Runs code · Step 5: Present

• customize.toml overrides for route, review, handoff and plan opening

  Defaults can be overridden in _bmad/custom/bmad-build.toml. Settings include route, review, the implementation handoff, the review lenses and an open_plan command for your editor.

  Customization

Watch out for

• It runs a script from your project folder

  SKILL.md runs _bmad/scripts/render_skill.py with uv. That script comes from BMad setup in your project, so read it before you trust it. github.com

• Subagents edit code and run commands

  Step 3 hands the plan to a no-context subagent and runs the plan’s Verification commands. A plan’s commands are whatever it contains, so read the plan before you approve it. github.com

• It makes a local commit

  Step 5 commits locally with a message derived from the plan title and never pushes by itself. github.com

Security advisories