Don Cheli SDD Framework: Transforming AI-Assisted Development from "Vibe Coding" to Professional Software Delivery

Don Cheli is a Specification-Driven Development (SDD) framework designed to address issues in AI-assisted development caused by "Vibe Coding"—such as context loss, uncontrollable code quality, and lack of testing. Through 85+ commands, 43+ skills, and 15 reasoning models, the framework transforms AI-assisted development into a structured, professional software engineering process, enforcing TDD, security audits, and multi-dimensional code reviews to ensure each phase has clear standards and verification mechanisms.

The purely AI-driven "Vibe Coding" model has three typical dilemmas:

1. Context Rot: AI assistants forget architectural decisions after long conversations;
2. Silent Stubs: AI claims to implement services but code contains // TODO comments;
3. Lack of Verification: No guarantees for code functionality, testing, or deployment status. These issues stem from AI generating code based on probability matching, lacking rigorous engineering verification.

Don Cheli's SDD methodology defines six strictly sequential phases: Specify→Clarify→Plan→Break Down→Implement→Review, with quality gates at each phase. The framework takes TDD (Test-Driven Development) as an iron rule: before implementing a feature, you must first have a failing test, then write the minimal code to pass it, and finally refactor and optimize—ensuring code testability and verifiability.

Don Cheli provides 85+ commands, 43+ skills, and 15 dedicated reasoning models. Core capabilities include:

• Six-Layer Quality Gates: Specification verification, architecture review, implementation validation, security audit, peer review, acceptance testing;
• 15 Reasoning Models: Pre-mortem analysis, 5 Whys, complexity assessment, estimation models, etc.;
• Unique Features: PoC mode (rapid idea validation), Blueprint Distillation (legacy system reverse engineering), CodeRAG (contextual code suggestions), tech stack migration support, Skill Creator meta-skill (self-enhancing capability ecosystem), etc.

Traditional AI-Assisted Development: Directly letting AI generate untested code, which easily leads to production crashes; Don Cheli Approach: Trigger the full process via /dc:start "JWT-authenticated user API": Complexity detection → Generate Gherkin specifications → Technical blueprint → TDD task breakdown → Docker execution → Completion with evidence, ensuring each link is traceable.

Don Cheli supports mainstream AI tools (Claude Code, Gemini/Antigravity, Cursor, Codex) and follows the Anthropic Skills 2.0 specification. Documentation is available in English, Spanish, and Portuguese, making it the first AI-assisted development framework for the Latin American community. The project practices its own specifications: GitHub Actions CI, semantic versioning (v1.14.0), Apache 2.0 license, and a skill marketplace supporting multi-source installation.

Don Cheli is suitable for various project scenarios:

Project Type	Recommendation Level	Expected Benefit
Authenticated REST API	Level2	Reduce defects by ~40%
Microservices Architecture	Level3	Improve architectural consistency
Legacy System Refactoring	Level4	Lower modernization risks
Rapid Prototype Validation	PoC Mode	Fast failure, low-cost validation
For projects pursuing code quality, team collaboration, or critical business needs, it can significantly reduce technical debt accumulation.

Don Cheli represents the evolutionary direction of AI-assisted development: from "letting AI write code" to "using AI to do good software engineering". Its core philosophy—"Stop guessing, start engineering"—reminds teams that the competitive advantage in the AI era lies in reliably delivering maintainable, verifiable, and evolvable software systems, not just generating code quickly. Don Cheli provides a comprehensive methodology and toolset for this purpose.