Maintained by: product role
Last updated: 2026-05-14
vstack is a VS Code–native AI engineering workflow system. It provides a set of structured skills — executable by GitHub Copilot in Agent Mode — for planning, reviewing, verifying, and releasing software.
What gets built is determined by the product vision: a microservice, an API, a package, an app, or a fullstack system. The designer role adapts to frontend, backend, or fullstack scope. The tester role adjusts its checks accordingly. vstack itself has no opinion on what you build — only on how you build it well.
Each skill is a procedure. Each role is a persona. Together they form a repeatable engineering pipeline that works inside VS Code without any external binary requirements.
Most AI workflow systems are built around a fixed assumption of what you're building — typically a fullstack app with a browser-driven QA loop. That assumption is baked in and hard to override.
vstack is different:
- The product role defines the vision: what kind of system, what scope, what success looks like.
- Roles downstream (architect, designer, engineer, tester, release) read that vision and adapt their work to it.
- Browser automation, frontend concerns, and mobile targets are available but not assumed.
- VS Code + Copilot Agent Mode is the runtime, and
vstackCLI is the standard install path. - No proprietary platform or closed runtime is required.
The result is a workflow system that fits microservices, packages, APIs, apps, and fullstack products equally — driven by the product vision, not by tooling assumptions.
-
Vision-driven scope. The product role defines what gets built — microservice, package, API, app, or fullstack system. All other roles adapt to that vision. No product type is assumed or privileged by the tooling.
-
VS Code native. Skills are generated from templates and installed into
.github/agents/and.github/skills/byvstack install --target <project>(primary, project-local path). For cross-workspace usage, usevstack install --global. No extension required. -
Templates over hand-editing. Source of truth lives under
src/vstack/_templates/:skills/<name>/template.mdandagents/<name>/{config.yaml,template.md}. Generated files are overwritten on each build. Edit templates, then regenerate. -
Minimal runtime dependencies. The runtime core uses Python 3 stdlib plus one external runtime dependency:
pyyaml>=6.0(no Bun/Node runtime dependency). Developer workflow may use Poetry and dev tooling (ruff, mypy, pytest) for quality checks. -
Role model. Six fixed roles (product, architect, designer, engineer, tester, release) define who produces what. Each role owns a specific set of artifacts. Agents communicate through files on disk.
-
Pipeline-ready. The planner/DAG orchestration model now handles multi-role execution inside VS Code. If coordination needs to evolve further, it should extend the existing workflow contract rather than forcing a rewrite of skills.
An engineer or small team that wants:
- A repeatable, role-based engineering workflow in VS Code
- AI-assisted planning, architecture reviews, verification, security audits, and releases
- A system that adapts to what they're building — not the other way around
- Extensible via Markdown templates, with no proprietary lock-in
Scope is not fixed by vstack — it is set by the product vision for each project.
Supported product types:
- Microservices and backend APIs (REST, gRPC, event-driven)
- Internal libraries and packages (Python, Go, TypeScript/Node, Java)
- Fullstack applications (frontend + backend)
- Standalone apps
Adapts per vision:
- Designer role: backend-mode (API specs, data contracts) or frontend-mode (UI/UX, components)
- Tester role: contract + integration tests for services; E2E for apps with a UI
- Tester role: security + performance auditing, regardless of product type
Tooling non-goals (independent of product scope):
- No browser binary required by default
- No VS Code extension packaging
- No cloud service dependencies
- No Node/Bun runtime dependency