bmad-build-auto

Runs one unattended development iteration from intent through planning, implementation, and multi-layer review.

Updated Sep 3, 2026
One-click install
npx skills add https://github.com/watchthelight/shatterfish --skill bmad-build-auto-watchthelight
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: bmad-build-auto
Source: https://github.com/watchthelight/shatterfish/tree/main/.claude/skills/bmad-build-auto
Command: npx skills add https://github.com/watchthelight/shatterfish --skill bmad-build-auto-watchthelight

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires uv, and includes references (resource) components.

What problem does it solve? It automates a full unattended development loop: turning a raw intent, story ID, or spec file into a planned specification, implemented code change, and multi-reviewer-audited result without human interaction. ## Core Features & Use Cases - Intent routing and planning: Resolves invocation intent (free-form, epic story, or existing spec), compiles epic context from planning artifacts, and produces a ready-for-development spec with code maps and acceptance criteria. - Subagent-based implementation: Hands the spec to a context-free implementation subagent, then runs verification commands and an I/O edge-case matrix test audit. - Multi-layer automated review: Runs parallel review layers (blind hunter, edge-case hunter, verification-gap, intent-alignment), triages findings into intent_gap/bad_spec/patch/defer/reject, and loops fixes until done or blocked. - Use Case: Point it at a story in stories.yaml and it plans, implements, reviews, commits, and writes a structured Auto Run Result into the story spec file. ## Quick Start Invoke the bmad-build-auto skill with a short description of the change you want implemented, or a spec folder and story ID, and let it run the full loop unattended.

Frequently Asked Questions about bmad-build-auto

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an unattended AI development loop on a story?

Invoke the skill with a spec folder and story ID. It reads stories.yaml, resolves the story, plans a spec, implements it via a subagent, runs parallel review layers, triages findings, and commits the result with status done or blocked.

How does automated spec-driven code review work?

The skill constructs a diff from the baseline revision, then runs four review layers in parallel: blind hunter, edge-case hunter, verification-gap, and intent-alignment. Findings are deduplicated, severity-rated, and triaged into intent_gap, bad_spec, patch, defer, or reject categories.

What happens when a review finds problems in the implementation?

Patch findings are auto-fixed and re-verified. Bad_spec findings trigger a spec amendment and full re-implementation loopback, capped at 5 iterations. Intent gaps halt the run with status blocked and save the attempted change as a patch file.

Does this workflow require version control and subagents?

Subagents are mandatory; the run halts with status blocked if they are unavailable. Version control is used for baseline diffs and clean-tree checks, but the workflow degrades gracefully with a NO_VCS fallback when it is unavailable.

What are the limitations of unattended development loops?

The loop halts rather than guesses on ambiguous intent, unresolved intent gaps, failed verification, or non-converging review loops. It never asks questions mid-run, so unclear requirements produce a blocked status with the open questions recorded.