Vibe coding is fast until it breaks: code nobody understands, a project you cannot change safely. Blueprint runs focused skills that plan, build, and prove one feature at a time, turning every decision into durable state you can inspect.
Version 0.9.0 adds read-only terminal status, versioned JSON output, and an optional global blueprint command.
Starting fresh? Scaffold an app in your preferred stack first. Already have a working codebase? Install Blueprint inside it and use the adoption path.
Create the app in your preferred stack, then install Blueprint and tune the workflow to the new repo.
Blueprint surveys shipped features and existing conventions, then generates plans that reflect the real project.
The setup creates durable project context. Then every feature moves through a spec, implementation, proof, and review cycle.
Give the agent a durable understanding of what you are building.
One scoped feature moves through visible human review gates.
You own the plan. Blueprint turns it into durable context, one active feature, and a readable history of what shipped.
Clear the chat or come back next week and the next session resumes from those files, not from a conversation it no longer has.
AI projects drift when scope, decisions, and proof live only in chat.
Blueprint is a workflow layer, not an app starter. Your framework, architecture, and product choices stay yours.
The agent writes the feature spec first, then stops. You approve the shape before code exists.
Plans, current work, standards, and history live beside the project in readable markdown.
The agent shows diffs, runs the real app, and proves each done-when before work lands.
Tune a fresh scaffold
Survey existing code
Check workflow health
Develop plans through conversation
Generate project context
Preview upcoming work
Lock the visual direction
Write a reviewed spec
Reverse completed work
Build in small steps
Run one bounded pass
Archive and close out
Add the unit test gate
Automate the Verify recipe
Diagnose without editing
Observe the real app
Generate manual steps
Audit broadly or by lens
Blueprint stays lightweight by default. Add unit testing when application logic appears, run the same Verify recipe locally and on GitHub, then prepare release checks when the project is ready to ship.
Add the stack-native test runner when you need it, then let Blueprint define one Verify command from the configured typecheck, tests, and build. GitHub Actions runs that exact recipe automatically.
Blueprint checks the build, start command, output directory, environment names, and health path. It can prepare local Render or Vercel configuration when the project needs it.
Blueprint does not compete with app starters or coding agents. It gives them durable context and clear review boundaries.
| Option | What it gives you | What is still missing |
|---|---|---|
| App boilerplate | Starter code and patterns | Your product plan and workflow history |
| Chat-only AI coding | Fast implementation | Durable project state |
| AI Blueprint | A file-backed workflow | App code is still yours to choose |
No. It installs workflow files and agent skills, not auth, billing, database code, or UI components. Scaffold the app you actually want first.
No. The project state lives in markdown. Adapters currently support Codex and Claude Code, and the workflow can be understood by other coding agents.
No. The core loop is feature, implement, check, and complete. Discovery, brief, debug, try, audit, tests, CI, release, prototype, rollback, and autopilot are optional tools for specific situations.
Yes. Run the adopt workflow. It surveys the real repo and creates plans that account for features you have already shipped.
Install AI Blueprint inside the app you already scaffolded. Your code stays yours. Your workflow gets clearer.