A complete product-development system for AI agents
OpenCraft gives your AI assistant a disciplined path from a rough idea to a meaningful, original, secure, tested, and deployment-ready product—without expecting you to speak engineering jargon.
Meaning before features
Evidence before claims
Root cause before fixes
Human judgment before trade-offs
A working screen is not the same as a useful product. Real delivery means knowing whose problem matters, choosing what not to build, designing every failure state, protecting sensitive data, and knowing exactly what “ready” means.
OpenCraft turns that judgment into a repeatable workflow your AI assistant can follow, while keeping the important decisions visible to you.
The complete journey
Each stage answers a plain-language question and leaves evidence for the next stage.
Separate what you know from what you assume. Understand the people, their current alternatives, the opportunity, and the evidence needed before investing.
Turn the validated opportunity into a readable product agreement: outcomes, journeys, permissions, requirements, success measures, edge cases, and explicit non-goals.
Design the journey, content, interactions, and technical foundation together. Give the product a recognizable point of view without sacrificing clarity or accessibility.
Break the product into small, demonstrable slices. Every task has boundaries, dependencies, risks, and a clear way to prove it is complete.
Build real behavior through tests, connect the complete journey, and diagnose failures from their root cause rather than layering on speculative fixes.
Review what was built against what was promised. Test the complete journey across roles, devices, permissions, failures, sensitive data, migrations, and recovery paths.
Prepare the exact release, observation, and recovery plan. Deployment remains permission-gated, and production feedback becomes evidence for the next iteration.
Nothing gets lost between stages
OpenCraft gives every important decision a stable identity. That lets you follow a promise from the PRD to the design, code, security control, test result, and release evidence.
Product craft
OpenCraft rejects design that looks polished but says nothing. Originality must come from the people, language, data, and rituals of the product's real domain.
21 focused skills
Skill names are for the AI agent. The outcomes are for everyone.
Right-sized process
OpenCraft does not force a new product, a small feature, and an urgent incident through the same amount of process.
Build a durable product from first evidence through launch.
Add value without breaking the system people already use.
Test one important uncertainty, then discard or productionize.
Close quality, security, performance, or operational gaps.
Contain, find the cause, restore safely, and prevent recurrence.
Built to be trusted
Every skill is validated, evaluated, checksum-locked, and tested across supported agent directories before release.
Start in minutes
Install every skill for your preferred coding agent, then ask it to take your product through the complete journey.
npx opencraft-skills install \
--target all \
--project . \
--with-project-filespython3 scripts/install.py \
--target all \
--project . \
--with-project-filesNode.js 18+ · No runtime dependencies · MIT licensed · View package on npm
Your first prompt
“Use $ship-web-product to take this idea from product analysis and PRD through original design, implementation, responsive validation, security and data testing, and deployment readiness. Do not deploy without my explicit approval.”Explore OpenCraft on GitHub
Common questions
No. The workflow is designed to explain decisions in product language. Your coding agent handles repository details while keeping consequential choices, risks, and evidence visible to you.
No. It reads and respects your existing repository first. For a new product, it chooses the simplest architecture supported by the requirements rather than imposing a fashionable framework.
No. OpenCraft can prepare the complete deployment and rollback plan, but production deployment and other consequential external actions require explicit permission.
No. It gives AI agents a stronger working method and makes their output easier for people to inspect. Human judgment remains essential for product decisions, risk acceptance, and production authority.
Yes. Invoke a focused skill for one stage, or use $ship-web-product to orchestrate the complete lifecycle.