🌐 Read in: English | Español | Português
Don Cheli — SDD Framework
Stop guessing. Start engineering.
The only framework where TDD is law, not a suggestion.
One command. Verified code.
/dc:auto "Implement JWT authentication with refresh tokens"
Don Cheli takes your idea and delivers tested, reviewed, verified code — automatically.
✅ /dc:specify 8 Gherkin scenarios generated
✅ /dc:clarify 2 ambiguities resolved
✅ /dc:tech-plan Blueprint: 3-layer architecture
✅ /dc:breakdown 7 TDD tasks created
✅ /dc:implement 14 tests pass, 91% coverage
✅ /dc:review 7/7 review dimensions passed
Result: ALL PASSED — Project updated with verified code
Your project is untouched until everything passes. If anything fails, nothing changes.
Install
npm install -g don-cheli-sdd
don-cheli install --global
Other install methods
# Git clone
git clone https://github.com/doncheli/don-cheli-sdd.git
cd don-cheli-sdd && bash scripts/instalar.sh
# One liner
curl -fsSL https://raw.githubusercontent.com/doncheli/don-cheli-sdd/main/scripts/instalar.sh | bash -s -- --global --lang en
# VS Code Extension
code --install-extension doncheli.don-cheli-sdd
The Runtime — Your code never breaks
Other frameworks give you prompts and hope the AI follows them. Don Cheli has a real TypeScript orchestrator that makes it impossible to skip rules.
How it works in 4 steps:
Step 1: The runtime creates a copy of your project (git worktree). Your real project is never touched.
Step 2: It spins up a Docker container and runs each /dc:* command inside — completely isolated.
Step 3: After each phase, real quality gates verify the output:
- Does the Gherkin spec exist? Does it have P1 scenarios?
- Do tests exist? Do they pass? Are there zero
// TODO stubs?
- Is coverage >= 85%?
- Do your custom rules in
.dc/gates/*.yml pass?
Step 4: Only when ALL 6 phases pass ALL gates does the code merge into your project. If anything fails → your project stays exactly as it was.
3 providers: Use your Claude Code subscription (free), OpenAI Codex, or Ollama for 100% free local execution.
How it works
flowchart LR
P["📋 PRD"] -.->|optional| B
A["💡 Idea"] --> B["/dc:specify"]
B --> C["/dc:clarify"]
C --> D["/dc:tech-plan"]
D --> E["/dc:breakdown"]
E --> F["/dc:implement"]
F --> G["/dc:review"]
G --> H["✅ Verified"]
style P fill:#2d3436,color:#dfe6e9,stroke:#6c5ce7,stroke-dasharray: 5 5
style F fill:#e17055,color:#fff
style H fill:#00b894,color:#fff
| Phase | Command | What happens |
|---|
| Specify | /dc:specify | Your idea → Gherkin specs with test scenarios |
| Clarify | /dc:clarify | QA finds ambiguities before you code |
| Plan | /dc:tech-plan | Architecture + API contracts + schema |
| Break Down | /dc:breakdown | TDD tasks with parallelism markers |
| Implement | /dc:implement | Test FIRST → code → refactor (Iron Law) |
| Review | /dc:review | 7-dimension peer review |
Each phase has a quality gate. You don't advance without passing. No shortcuts.
3 ways to use it
# Interactive — you drive each phase
/dc:start "JWT auth"
# Autonomous — runtime drives everything, Docker isolates
/dc:auto "JWT auth"
# Terminal — without AI agent open
don-cheli auto "JWT auth"
The Iron Laws
These are not suggestions. The runtime enforces them.