Ai Project Factory Kernel

AI Project Factory Kernel v7

[](README.md) [](LICENSE)

Reusable meta-kernel for AI-assisted software development. Generates release-candidate projects from prompts — not universal app templates, not MVP scaffolds by default.

Works on Windows, Linux, and macOS. Maintenance scripts run via Node.js 18+ (npm run ...).

Works with Cursor, Claude Code, Cline, Roo, Windsurf, and other coding agents that read AGENTS.md, project rules, or MCP configs.

What it does

For each user request, the kernel guides the AI through:

1. Material clarification only
2. Access, tools, and API key checks
3. Live browser-based discovery
4. Comparison of AGENTS, Skills, Rules, MCP servers, repo patterns, references
5. Project classification and complexity tier
6. Project-specific AI development environment (Tier 2–3)
7. Product, frontend, backend, design, security, testing, deployment analysis
8. Dynamic project structure synthesis
9. Stopping criteria before implementation
10. Release-candidate implementation
11. Visual QA for UI projects
12. Documentation and code comments
13. Release-readiness evidence

Core rule

Default target is release candidate, not MVP.

Minimal output is allowed only when minimality is the correct production architecture (e.g. a static landing page without backend).

Why this exists

Most AI project generators fail in one of two ways:

• stop too early with a demo/MVP
• never stop improving without a clear done definition

This kernel addresses both with explicit stopping criteria, evidence-based release gates, and tier-aware workflow depth.

Quick start

1. Clone or copy the kernel

git clone https://github.com/kakakak241/ai-project-factory-kernel.git
cd ai-project-factory-kernel

Or copy the folder into your workspace.

2. Open in your AI coding tool

The kernel ships entry points for common tools:

| File | Purpose | |---|---| | AGENTS.md | Short bridge for agents that look for root AGENTS.md | | FACTORY.md | Master workflow — read for full detail | | .cursorrules | Cursor | | .clinerules | Cline | | .windsurfrules | Windsurf |

3. Run the quick-start prompt

Use AI Project Factory Kernel v7.

I want to build: [project idea].

Do not default to MVP. First ask only necessary clarification questions. Then perform live browser-based discovery for AGENTS, Skills, Rules, MCP, repo patterns and references. Create a project-specific AI development environment. Define stopping criteria before implementation. Analyze product, frontend, backend, design, security, testing, deployment. Synthesize structure dynamically. Implement a release-candidate version if not blocked. Run visual QA for any UI. Generate release-readiness and generation reports.

Or use prompts/start-project.md and examples/ for typed requests.

4. Verify kernel structure (optional)

Requires Node.js 18+:

npm run setup
npm run doctor
npm test

Works the same on Windows, Linux, and macOS. Legacy ./scripts/*.sh wrappers delegate to the same Node scripts.

Complexity tiers

Tier is chosen during intake and controls workflow depth:

| Tier | Profile | Pipeline | |---|---|---| | 1 — Simple | Static/landing, no backend, auth, or DB | 0 → 1 (focused) → 3 → 4 → 6 → 7 → 8 (minimal) | | 2 — Standard | App with backend, auth, DB | Full pipeline | | 3 — Complex | Platform, SaaS, marketplace, payments, AI | Full + extended vendor/security |

Tier 1 minimal artifacts: STOPPING_CRITERIA.md, GENERATION_REPORT.md, RELEASE_READINESS_REPORT.md

Tier 2–3: full artifact set per FACTORY.md

Repository layout

AGENTS.md              — short entry point for coding agents
FACTORY.md             — master workflow
policies/              — what must be true
protocols/             — how to perform procedures
roles/                 — sequential AI role behavior
templates/             — files generated INTO projects
prompts/               — user-invocable prompts
mcp/                   — MCP server seed docs
configs/               — per-IDE MCP config examples
tools/                 — external tool catalog
context/, memory/, tasks/, references/ — per-project layout templates
examples/              — example user prompts
scripts/               — kernel maintenance (doctor, setup, security)

Operating model

• One AI context — roles adopted sequentially, not separate agent swarm
• Dynamic structure — no forced src/app/components/api/db scaffold
• Evidence over claims — release readiness requires artifacts and checks
• Vendor provenance — copy / adapt / overlay / reference-only / reject with license tracking

How it differs from similar tools

| | This kernel | BMAD / Spec Kit | Cursor rules kits | |---|---|---|---| | Primary artifact | Decision process + gates | Specs / workflows | IDE rules | | Default output | Release candidate | Spec-driven implementation | Code assistance | | Stopping criteria | Required before code | Partial / varies | Rare | | Live discovery | Required | Optional | Rare | | Vendor composition | Formal provenance | Varies | No | | Multi-IDE | Yes | Mostly yes | Usually one IDE |

See references/ai-project-factory-systems.md for a longer comparison.

Generated project outputs

When applicable:

PROJECT_BLUEPRINT.md
PROJECT_KERNEL_MANIFEST.md
EXTERNAL_DISCOVERY_REPORT.md
AI_ENVIRONMENT_USAGE_MAP.md
PRD.md
ARCHITECTURE.md
DESIGN_SYSTEM.md
SECURITY.md
TESTING_STRATEGY.md
DEPLOYMENT.md
STOPPING_CRITERIA.md
RELEASE_GATES.md
DOCUMENTATION_MATRIX.md
CODE_COMMENT_COVERAGE.md
docs/VISUAL_QA_REPORT.md
GENERATION_REPORT.md
RELEASE_READINESS_REPORT.md

For external AI artifacts:

vendor/
project-overrides/
compiled-ai-environment/
VENDOR_MANIFEST.md
ADAPTATION_DIFF.md
LICENSE_NOTICES.md

Scripts

Cross-platform via npm (Node.js 18+):

| Command | Purpose | |---|---| | npm run setup | Prepare kernel scripts | | npm run doctor | Validate kernel structure | | npm run validate-env | Check local CLI tools | | npm run create-project -- "..." | Save a project request for AI | | npm run security | Scan for secret-like strings | | npm test | Run doctor + security checks | | npm run build:zip | Package kernel (zip on Windows, zip/tar.gz elsewhere) | | npm run install-mcp | Print MCP install guidance |

Equivalent shell wrappers: scripts/*.sh (optional on Unix; call the same Node scripts).

Contributing

See [CONTRIBUTING.md](CONTRIBUTING.md).

Security

See [SECURITY.md](SECURITY.md). Never commit real secrets — use configs/.env.example placeholders.

License

[](LICENSE)

MIT — see [LICENSE](LICENSE).

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: kakakak241
• Source: kakakak241/ai-project-factory-kernel
• License: MIT

Install and usage instructions live in the source repository linked above.