Agile V: Complementary framework for traceable AI development

[KochC]

This is excellent work on deterministic AI verification! The focus on fail-closed architecture and formal verification aligns perfectly with what we're building in Agile V.

We've developed a complementary framework focused on traceable requirements → implementation → verification loops for AI-assisted development:

• https://github.com/Agile-V/agile_v_skills (agent skills for evidence-based engineering)
• https://github.com/Agile-V/agentic_agile_v (practical scaffold with validation gates)

The combination of your Z3 prover approach with our human-in-the-loop gates could create a powerful verification stack. Would love to explore potential collaboration or cross-referencing in our documentation.

Key overlap areas:

• Hallucination detection through evidence bundles
• Deterministic verification (your formal methods + our traceability)
• MCP integration for secure AI workflows

Our approach:

Agile V enforces a verified engineering loop:

1. Clarify intent → Structured requirements
2. Create traceable requirements → Machine-verifiable artifacts
3. Build against approved requirements → Implementation with traceability
4. Independent verification → Red Team testing
5. Evidence bundles → Compliance-ready audit trails
6. Human approval gates → Final release decision

Happy to discuss further and explore integration possibilities!

[rahuldass19]

Hi KochC (@KochC) thanks for sharing this. I'm grateful to you for contacting me🙇

I can definitely see some overlap around auditability, traceability, and making AI-assisted workflows more accountable.

One clarification I would like to make qwed-learning is primarily the learning repository. However, most of the core verification work happens in qwed-verification https://github.com/QWED-AI/qwed-verification, where the focus is deterministic verification, explicit trust boundaries, and fail-closed behavior. QWED also maintains many other modules and protocols under its deterministic philosophy which you can check.

At a high level, Agile V appears to be operating more at the process and traceability layer, while QWED focuses on verification at the computation layer. Those seem more complementary than directly overlapping.

I’ll take a closer look at the repositories when I get some time. Appreciate you sharing the work and starting the conversation.

Let me know if something clicks your mind, will be happy to discuss.