Buildwright: A New Generation Intent-Driven Autonomous Development Workflow

Buildwright is an agent-first autonomous development workflow for Claude Code, OpenCode, and OpenClaw. Its core idea is radical: humans approve intent, agents deliver automatically. Developers only need to approve a requirements specification, and AI agents handle the entire process from research, planning, building, security scanning, code review to PR submission. This shifts the developer's role from writing code to making key decisions, liberating them from repetitive tasks.

Software development is undergoing a paradigm shift—from "writing code" to "approving code". With mature AI programming agents like Claude Code, OpenCode, and OpenClaw, Intent-Driven Development is becoming a reality. Buildwright's design philosophy centers on humans in the loop as decision-makers (since code can be generated, but product direction and technical trade-offs need human judgment). Its workflow has seven stages: RESEARCH → RESOLVE → PLAN → VALIDATE → APPROVE → BUILD → UPDATE DOCS → SHIP. The key node is the "APPROVE" step—humans only decide at this critical point, with all other stages automated.

Buildwright offers two main modes:

1. /bw-new-feature: For full feature development (research → clarify → plan → validate → build → deliver).
2. /bw-quick: For quick fixes/small changes (fast research → TDD → validation → security scan → code review → docs update → PR).

It also supports greenfield projects: agents ask for product vision and technical constraints, generate a recommended tech stack, then wait for human approval/adjustment before proceeding.

Auto-approve Mode: Enabled via BUILDWRIGHT_AUTO_APPROVE env var—skips manual approval, submits specs to git for audit, and proceeds to build. Suitable for routine features, CI/CD pipelines, or batch processing. Multi-agent Collaboration: Involves specialized agents: research (code analysis), planning (spec generation), senior engineer (spec review), build (TDD implementation), validation (tests), security (scans), review (PR prep). Each agent focuses on a specific task to enhance quality and efficiency.

Integration: Supports Claude Code, OpenCode, OpenClaw via a unified SKILL.md. Installation options:

• Project-level: Setup script creates configs (.claude/, .opencode/, etc.) and AGENTS.md.
• Global: Install as an OpenClaw skill.

Quality Assurance: Includes TDD (test-first implementation), auto validation (test runs), security scans (vulnerability checks), code review (AI-based), and failure handling (commits work and submits PR with failure report if any stage fails).

Use Cases:

• Startups: Fast iteration, focus on product decisions.
• Enterprises: Standardized workflows, compliance via audit trails.
• Open source: Automate PR reviews/docs updates.
• Individual devs: Focus on creativity over coding.

Limitations:

• Complex architecture changes need human architects.
• Domain-specific knowledge gaps (e.g., finance/medical).
• Creative work (UI/UX) still requires humans.
• High-security scenarios may not use auto-approve.

Buildwright represents the future of AI-assisted development—freeing developers from repetitive tasks to focus on creative, high-value work. It is not a replacement for developers but an enabler. For teams aiming to boost efficiency and standardize workflows, Buildwright is a tool worth exploring as AI agent capabilities continue to evolve.