Related project
VC-AD
Value-Continuous AI Development: a development protocol in which AI writes the implementation while humans keep the core of judgment and responsibility.
VCDesign is law; VC-AD is its implementation at the current technical level.
Source: VCDesign README
What VC-AD does
When an AI agent writes code, VC-AD hands it, as YAML inside the repository, what was decided, what must and must not happen, where things connect, and which files the rules apply to. The AI implements within that scope, and tools check that the scope is kept.
In VCDesign terms, it makes the scope authorized by a decision closed by a human (axiom A2) machine-readable in day-to-day development.
Repository
The norms are in VCDesign-org/architecture-yaml-protocols. See its README for usage and the latest layout. As of September 2026 the main parts are:
| Directory | Role |
|---|---|
decisions/ | Why / What: architectural decisions, policies, technology choices |
contracts/ | Must / Must Not: language-agnostic policies (resources, side effects) and IO contracts |
connections/ | How / Interface: connections between components, data flows, API contracts |
components/ | Where: maps the rules above to actual code (file paths) |
tools/ | Gatekeeper: tools that check the contracts are kept |
About the earlier pages
The VC-AD pages previously on this site (the Quickstart prompt, the "11 boundaries" and the .vc-ad/boundaries.yaml Static Badge, and the Basic Constitution) were based on an earlier layout of the repository. The current repository no longer has boundaries/ or closures/, so those pages were moved to the archive. The January 2026 research report is also listed in the archive.