aiblueprinthq/ai-blueprint

What it solves

AI coding agents often become unreliable when product intent is trapped in chat histories, features overlap, and claims of "working" code are not backed by evidence. AI Blueprint provides a structured, file-backed workflow framework that ensures AI agents follow a controlled loop of planning, implementation, and verification, keeping the project state durable and tool-independent.

How it works

It implements a spec-driven workflow where agents must write a feature spec and obtain review before coding. The process follows a strict loop: feature -> review spec -> implement -> check -> audit current -> complete. This state is stored in readable Markdown files (like project-plan.md and current-feature.md) within the project repository, allowing the project to move between different coding agents (such as Claude Code, GitHub Copilot, or Codex) without losing context.

Who it’s for

Developers using AI coding agents who want more control, better traceability, and a way to prevent AI-generated regressions or hallucinations in real software projects.

Highlights

• File-backed state: Project plans, specs, and audit findings are stored as files, not in chat history.
• Spec-before-code: Forces a review gate before implementation begins.
• Tool-agnostic: Works with any application stack and supports multiple agents via adapters.
• Verification gates: Includes dedicated commands for checking acceptance criteria and auditing branch deltas.
• Context efficiency: Optimizes how context is loaded into agents to reduce token usage.
• Automation modes: Offers "Autopilot" and "Continuous" modes for bounded local automation while maintaining safety boundaries.