bmad-build

Implements code changes through a spec-driven plan, implement, and review workflow.

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

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Turning a feature request, bug fix, or refactor into working code often skips planning and review, producing inconsistent changes that drift from project conventions. This Skill enforces a disciplined pipeline that converts user intent into an approved specification, implements it, and runs adversarial multi-layer review before completion. ## Core Features & Use Cases - Spec-driven routing: Clarifies intent, loads epic or planning context, and routes work to either a one-shot path for zero-blast-radius changes or a full plan-code-review pipeline. - Frozen-intent specifications: Generates a token-bounded spec file with intent, boundaries, I/O edge-case matrix, code map, tasks, and acceptance criteria, locked after human approval. - Adversarial review loop: Runs parallel review layers (Blind Hunter, Edge Case Hunter, Verification Gap) over the diff, classifies findings, and loops back to re-plan or patch until clean. - Use Case: Ask it to add a new API endpoint; it drafts a spec, gets your approval, implements via a subagent, reviews the diff for edge cases and verification gaps, commits, and presents a suggested review order. ## Quick Start Ask the assistant to implement a specific feature or fix, such as adding input validation to the login form, and it will plan, implement, and review the change end to end.

Frequently Asked Questions about bmad-build

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

FAQPage Schema
How do I implement a feature with a spec-driven AI workflow?

Provide your intent and the workflow clarifies it, drafts a specification with tasks and acceptance criteria, waits for your approval, then implements and reviews the change. You approve the spec before any code is written.

What is the difference between the one-shot and plan-code-review routes?

One-shot handles changes with zero blast radius and clear intent, implementing directly with a lightweight review. Plan-code-review handles everything else with full spec drafting, human approval, implementation, and multi-layer adversarial review.

How does the review step classify findings?

Findings are deduplicated, assigned severity by consequence, and routed into intent_gap, bad_spec, patch, defer, or reject categories. Intent and spec problems trigger loopbacks that re-derive the code, while trivial patches are auto-fixed.

Can it resume work on an existing spec or epic story?

Yes. It detects existing specs by status frontmatter (draft, ready-for-dev, in-progress, in-review) and resumes at the matching step. For epic stories it compiles or reuses a cached epic context file and loads continuity from previously completed stories.

What happens if the spec exceeds the token limit?

Specs target 900 to 1600 tokens. If a spec exceeds 1600 tokens, the workflow halts and asks whether to split secondary goals into deferred-work entries or keep the full spec, since oversized specs risk context rot in implementation agents.