Skip to content

VDD.md

Latest commit

 

History

History
24 lines (18 loc) · 1.82 KB

VDD.md

File metadata and controls

24 lines (18 loc) · 1.82 KB
Raw

VDD (Verification-Driven Development)

Source / Inspiration: Verification-Driven Development (VDD) via Iterative Adversarial Refinement

Core Philosophy

VDD is a high-integrity development mode designed for mission-critical or complex components. It shifts the focus from "Validation" (Does it work?) to "Verification" (Is it robust/secure/correct?).

It uses an Adversarial Agent (often called "Sarcasmotron") to actively try to break the code, find logical inconsistencies, or point out "slop" (lazy implementation).

The Cycle (Adversarial Loop)

  1. Build: Developer implements the feature (often using Stub-First TDD).
  2. Attack (The Roast): The Adversarial Agent reviews the code with the specific goal of finding flaws, security holes, or bad practices.
  3. Refine: Developer fixes the identified issues.
  4. Repeat: The loop continues until the objective bar is met — the full test run has executed, zero CRITICAL, zero legitimate logic/security/slop findings, and only bikeshedding remains (the "Objective Convergence" exit). Approval is bound to that bar — never to "the adversary was forced to invent nitpicks".

When to use VDD?

  • Security-sensitive modules (Auth, Payments).
  • Complex algorithmic logic.
  • Core system architecture.
  • Smart contract development (Solidity, Solana).
  • When "good enough" is not enough.

Integration with Security Audit

VDD pairs naturally with the security-audit skill (v3.2). The adversarial loop can include automated scanning (run_audit.py) to verify that the scanner itself catches real-world vulnerability patterns. This was validated in VDD Rounds 1-3 against recent DeFi hacks (Dec 2025 – Mar 2026), resulting in 16 new Solidity-specific detection patterns.