Skillget

don-cheli-sdd

agent
SUMMARY

Don Cheli — SDD Framework. The most comprehensive Specification-Driven Development framework for AI agents. 71+ commands, 42 skills, 15 reasoning models. TDD mandatory, OWASP audit, Anthropic Skills 2.0 compatible. Works with Claude Code, Gemini/Antigravity, Cursor, Codex. ES/EN/PT.

README.md

🌐 Read this in other languages: English | Español | Português

Don Cheli — SDD Framework

Stop guessing. Start Engineering.
Vibe coding is the spark; SDD is the engine. Transition from AI-assisted chaos to professional software delivery.

The first AI-assisted development framework made in and for Latin America.
The most comprehensive SDD framework on the market. Open source. Multilingual (ES/EN/PT).

Install Version License Languages Commands Skills Skills 2.0
CI npm Changelog Last Commit Contributors

Quick Demo

Don Cheli Demo

How does Don Cheli look in action?

Start a task with automatic complexity detection:

/dc:start Implement JWT authentication with refresh tokens

Don Cheli detects Level 2 (Standard), generates the Gherkin spec,
proposes the technical blueprint, breaks down into TDD tasks and executes in Docker.
No vibe coding. With evidence.

In Action

# Without Don Cheli:
"Claude, build me a users API"
# → Code without tests → broken in production

# With Don Cheli (one command):
/dc:start "Users API with JWT auth"
# → Spec → Tests first → Code → Review → Done with evidence

The Problem

You start a project with AI. The first 2 hours go well. Then:

  • Context rot — Claude forgets your architecture decisions
  • Silent stubs — It says "I implemented the service" but the code says // TODO
  • No verification — Does it work? I don't know. Tests? No. Can I deploy? Hopefully

That is vibe coding. And it is the enemy of quality software.

The Solution

Don Cheli transforms chaos into a structured process:

Specify → Clarify → Plan → Break Down → Implement → Review

Each step has quality gates. You don't advance without meeting them. Code is generated with mandatory TDD, stub detection, and 7-dimension peer review.

Why Don Cheli

BMAD
42K ⭐		GSD
39K ⭐		GSD-2
3K ⭐		spec-kit
82K ⭐		✨ Don Cheli
Commands~20~80~40~1085+
Skills~15~15~10~643
Reasoning models————15
Automatic estimates————4 models
Formal quality gates—1246
Mandatory TDD————Iron Law
PoC Mode————✅
OWASP Audit————✅
Stack Migration————✅
Worktree isolation——✅—✅
Crash recovery——✅—✅
Cost tracking——✅—✅
Stuck detection——✅—✅
Stub detection—✅✅—✅
UI Contracts—✅——✅
Nyquist Validation—✅——✅
Multilingual (ES/EN/PT)————✅
Anthropic Skills 2.0————✅ Compatible
Skill Creator (meta-skill)————✅
Skills Marketplace————✅
Multi-platform (Cursor, Antigravity)————✅

Data verified on 2026-03-27. Star counts are approximate.

20 things only Don Cheli has

  1. 15 reasoning models — Pre-mortem, 5 Whys, Pareto, RLM
  2. 4 estimation models — Function Points, AI Planning Poker, COCOMO, Historical
  3. PoC Mode — Validate ideas with timebox and success criteria before committing
  4. Blueprint Distillation — Extract specs from existing code (behavioral reverse engineering)
  5. CodeRAG — Index reference repos and retrieve relevant patterns
  6. OWASP Audit — Static security scanning integrated into the pipeline
  7. Stack Migration — Vue→React, JS→TS with wave plan and equivalences
  8. API Contracts — REST/GraphQL with retries, circuit breaker, idempotency
  9. SOLID Refactoring — Checklist, metrics, structured design patterns
  10. Living Documentation — ADRs, auto-generated OpenAPI, Mermaid diagrams
  11. Captures & Triage — Annotate ideas without pausing work, automatic classification into 5 categories
  12. Auto-generated UAT — Human-executable acceptance scripts after each feature
  13. Doctor — Diagnosis and auto-repair of git, framework, and environment
  14. Skill Creator — Iterative meta-skill: generate → test → evaluate → improve skills automatically
  15. Skills Marketplace — Install skills from official Anthropic, community, or create your own
  16. Project Constitution — Immutable pre-spec principles validated at every quality gate
  17. Formal Pseudocode — Technology-agnostic logic reasoning phase between spec and plan (SPARC)
  18. Multi-layer Spec Validation — 8 checks (implementation leakage, measurability, completeness, constitution adherence)
  19. Adversarial Multi-role Debate — PM vs Architect vs QA with explicit tensions and mandatory objections
  20. Scale-adaptive Planning — Planning level adjusts to complexity (same process for 1 file ≠ 100 files)

Use Cases

Project type Recommended level Impact
REST API with auth Level 2 ~40% fewer bugs on first delivery
Vue→React migration /dc:migrate Automatic plan in 10 min
Startup PoC PoC Mode Validated in 2-4h with clear criteria
Security audit /dc:security-audit OWASP Top 10 in < 5 min
Legacy refactoring SOLID + Blueprint Distillation Spec from code with no documentation
Weekly team planning /dc:planning RFCs + WSJF + squad assignment

Visual Pipeline

flowchart LR
    A["📋 SPECIFY\nGherkin + DBML"] -->|Gate 1| B["🔍 CLARIFY\nAuto-QA"]
    B -->|Gates 2+3| C["🏗️ PLAN\nBlueprint"]
    C -->|Gate 4| D["📦 BREAKDOWN\nTDD Tasks"]
    D -->|Gate 5| E["⚙️ IMPLEMENT\nRED→GREEN→REFACTOR"]
    E -->|Gate 6| F["✅ REVIEW\n7 dimensions"]
    F -->|Fail| E

    style A fill:#4A90D9,color:#fff
    style B fill:#7B68EE,color:#fff
    style C fill:#50C878,color:#fff
    style D fill:#FF8C00,color:#fff
    style E fill:#DC143C,color:#fff
    style F fill:#228B22,color:#fff

Installation

# 1. Clone
git clone https://github.com/doncheli/don-cheli-sdd.git

# 2. Install globally
cd don-cheli-sdd && bash scripts/instalar.sh --global

# 3. In any project, initialize
/dc:init

Interactive Installation (NEW)

Run without flags for a guided setup:

bash scripts/instalar.sh

Choose your tools, profile, skills, and commands step by step:

┌──────────────────────────────────────┐
│  Don Cheli SDD — Interactive Setup   │
└──────────────────────────────────────┘

Step 1: Where to install? → Claude Code, Cursor, Antigravity, OpenCode...
Step 2: Developer profile? → 6 archetypes with pre-configured skills
Step 3: Confirm → Summary of everything selected
6 developer profiles
Profile Tag For whom
👻 Phantom Coder Full-stack Real projects with TDD and quality gates
💀 Reaper Sec Security Hackers who also build
🏗 System Architect Architecture The one who designs, not patches
⚡ Speedrunner MVP/Startup Launch first, refine later
🔮 The Oracle Reasoning Hard decisions with 15 mental models
🥷 Dev Dojo Learning Grow while you build
Silent installation (CI/CD, scripts) 
bash scripts/instalar.sh \
  --tools claude,cursor \
  --profile phantom \
  --global --lang en

Flags: --tools, --profile, --skills, --comandos, --dry-run, --global, --lang

Language Selection

The first thing you see after installing is the language selector:

  ╔═══════════════════════════════════════════════════════════╗
  ║                                                           ║
  ║           🏗️  Don Cheli — SDD Framework                   ║
  ║                                                           ║
  ╚═══════════════════════════════════════════════════════════╝

  🌍 Selecciona tu idioma / Select your language / Selecione seu idioma

     1)  🇪🇸  Español
     2)  🇬🇧  English
     3)  🇧🇷  Português

  ▸ _

Once selected, the entire framework adapts to the chosen language: folders, files, templates, messages, and Claude's communication.

Structure by Language

The installation creates folders with names in the selected language:

Content🇪🇸 Español🇬🇧 English🇧🇷 Português
Skillshabilidades/skills/habilidades/
Rulesreglas/rules/regras/
Templatesplantillas/templates/modelos/
Hooksganchos/hooks/ganchos/
Agentsagentes/agents/agentes/

Project files (.especdev/) are also created in the configured language:

File🇪🇸 Español🇬🇧 English🇧🇷 Português
Statusestado.mdstatus.mdestado.md
Findingshallazgos.mdfindings.mddescobertas.md
Planplan.mdplan.mdplano.md
Progressprogreso.mdprogress.mdprogresso.md
Proposalpropuesta.mdproposal.mdproposta.md

The language is persisted in locale and folder-map.json so Claude knows exactly which files to look for. To change language, simply reinstall:

bash scripts/instalar.sh --global
Remote installation (without cloning)

The installer automatically downloads the repository when run via pipe:

# Interactive (prompts for language)
curl -fsSL https://raw.githubusercontent.com/doncheli/don-cheli-sdd/main/scripts/instalar.sh | bash -s -- --global

# Non-interactive (set language directly)
curl -fsSL https://raw.githubusercontent.com/doncheli/don-cheli-sdd/main/scripts/instalar.sh | bash -s -- --global --lang en

Available languages: es (Español), en (English), pt (Português)

Via npx

npx don-cheli-sdd init

Verified installation (recommended for security)

curl -fsSL https://raw.githubusercontent.com/doncheli/don-cheli-sdd/main/scripts/instalar.sh \
  -o instalar.sh
cat instalar.sh  # Review before executing
bash instalar.sh --global

Requirements: Claude Code (or compatible AI agent) + Git

Quick Start — Zero to productive in 2 minutes

Step 1: Install (1 min)

curl -fsSL https://raw.githubusercontent.com/doncheli/don-cheli-sdd/main/scripts/instalar.sh | bash -s -- --global --lang en

Step 2: Use (1 command)

You only need one command to start:

/dc:start "Implement JWT authentication"

That's it. Don Cheli will:

  1. Auto-detect the complexity level of your task
  2. Generate a Gherkin spec with acceptance criteria
  3. Write tests BEFORE the code (TDD iron law)
  4. Implement the minimum code to pass
  5. Run a 7-dimension peer review

You don't need to memorize 72 commands. Just /dc:start and follow the prompts.

Want more control? Use individual pipeline commands 
/dc:specify    # Gherkin spec + DBML schema
/dc:clarify    # Auto-QA + resolve ambiguities
/dc:tech-plan  # Blueprint + constitution check
/dc:breakdown  # TDD tasks with parallelism
/dc:implement  # RED → GREEN → REFACTOR
/dc:review     # 7-dimension peer review
Complexity levels (auto-detected, no action needed)
Level Name When
0 Atomic 1 file, < 30 min
P PoC Validate viability (2-4h timebox)
1 Micro 1-3 files, known solution
2 Standard Multiple files, 1-3 days
3 Complex Multi-module, 1-2 weeks
4 Product New system, 2+ weeks

The 3 Iron Laws

Non-negotiable. Always applied.

Law Principle In practice
TDD All code requires tests RED → GREEN → REFACTOR, no exceptions
Debugging Root cause first Reproduce → Isolate → Understand → Fix → Verify
Verification Evidence before assertions "Tests pass" > "I think it works"

Anthropic Skills 2.0

Don Cheli is 100% compatible with the Anthropic Skills ecosystem. It supports both skill formats:

Format File Usage
Anthropic SKILL.md Compatible with the official marketplace
Don Cheli HABILIDAD.md Extended format with version, tags, freedom degree

Skill Creator

Create skills without writing a single line of YAML:

/dc:create-skill "Weekly team report generator"

5 iterative phases: Discover → Generate SKILL.md → Test with real prompt → Evaluate quality → Iterate until optimal.

Skills Marketplace

Install skills from the official Anthropic marketplace or from the community:

/dc:marketplace --install document-skills --source anthropic
/dc:marketplace --search "weekly report"

Supported sources: Anthropic Official • skillsmp.com • aitmpl.com • Don Cheli built-in (42 skills)

Progressive Disclosure

Skills use a 3-layer design for maximum token efficiency:

Layer 1: Metadata (YAML)     → ~20 tokens per skill, always in context
Layer 2: Body (Markdown)     → Loaded only when the skill is activated
Layer 3: File References     → Loaded on demand within the body

This allows having dozens of skills without impacting the context window.

Commands (85+)

Backward compatible: All /dc:* commands are also available as /especdev:* for backward compatibility.

Main (32)
Command Description
/dc:init Initialize in a project
/dc:start Start task (auto-detects level)
/dc:quick Fast mode (Level 1)
/dc:poc Proof of Concept with timebox
/dc:full Full mode (Level 3)
/dc:status Current status
/dc:diagnostic Setup health check
/dc:doctor Diagnosis and auto-repair of git, framework, and environment
/dc:continue Recover previous session
/dc:reflect Self-reflection (+8-21% quality)
/dc:capture Fire-and-forget ideas with automatic triage
/dc:uat Auto-generated acceptance scripts per feature
/dc:agent Load specialized agent
/dc:roundtable Multi-perspective discussion (CPO, UX, Business)
/dc:tech-panel Senior dev experts table (Tech Lead, Backend, Frontend, Architect, DevOps)
/dc:planning Weekly team planning: RFC review, WSJF prioritization, pair/squad assignment
/dc:estimate Development estimates
/dc:distill Extract specs from code
/dc:mine-refs Search reference repos
/dc:ui-contract UI design contracts
/dc:api-contract API/webhook contracts
/dc:security-audit OWASP Top 10 audit
/dc:migrate Stack migration
/dc:reverse Architecture reverse engineering
/dc:explore Explore codebase (assumptions mode)
/dc:propose Change proposal
/dc:analyze-sessions Usage pattern analysis
/dc:present Generate interactive HTML presentation
/dc:create-skill Create skills iteratively (Anthropic Skills 2.0 compatible)
/dc:marketplace Install skills from Anthropic, community, or built-in
/dc:pseudocode Technology-agnostic logic between spec and plan (SPARC)
/dc:validate-spec Multi-layer spec validation (8 checks, BMAD-inspired)
/dc:debate Adversarial multi-role deliberation (PM vs Architect vs QA)
/dc:update Detect and apply framework updates
Gherkin Pipeline (5)
Command Description
/dc:specify Gherkin spec with P1/P2/P3+ priorities
/dc:clarify Auto-QA + schema-spec verification
/dc:tech-plan Blueprint + constitution check
/dc:breakdown TDD tasks with [P] markers
/dc:review 7-dimension peer review
Reasoning (15)
Command What it does
/razonar:first-principles Decompose to fundamentals
/razonar:5-whys Root cause analysis
/razonar:pareto 80/20 focus
/razonar:inversion Solve in reverse
/razonar:second-order Consequences of consequences
/razonar:pre-mortem Anticipate failures
/razonar:minimize-regret Long-term decisions
/razonar:opportunity-cost Evaluate alternatives
/razonar:circle-of-competence Know your limits
/razonar:map-territory Model vs reality
/razonar:probabilistic Reason in probabilities
/razonar:reversibility Can it be undone?
/razonar:rlm-verification Verification with sub-LLMs
/razonar:rlm-chain-of-thought Multi-step reasoning
/razonar:rlm-decomposition Divide and conquer

Skills (42)

Category Skills
Quality Iron Laws, Nyquist Validation, Stub detection, Loop detection, Quality gates, Proof of work, Progressive rigor
Context Context engineering, Context optimizer, Persistent memory (Engram), CodeRAG + LightRAG, Code reference mining
Reasoning 12 mental models, 3 RLM models (PrimeIntellect)
Architecture Living architectural map, SOLID Refactoring, DBML Schemas
Design UI/UX Design System (67 styles, 161 palettes), UI Contracts, HTML Presentations
Documentation Living documentation (ADRs, OpenAPI), DevLog, Traceability, Delta specs, Obsidian
Autonomy Autonomous orchestration, Auto-correction, Session recovery
Discovery Structured brainstorming, Git Worktrees
Efficiency Token optimization, Token accounting, Subagent development, Dynamic model routing, Cost projections
Observability Skill health (success rate and consumption telemetry per skill)
Security Permissions and security, OWASP Audit
Integration MCP servers, Extensions and presets

Each gate blocks progress if criteria are not met. No shortcuts.

7 specialized agents

Agent Model Role
planner opus Decomposition and planning
architect opus System design
executor sonnet Code implementation
reviewer opus Architectural code review
tester sonnet Testing and QA
documenter haiku Documentation
estimator opus Effort estimates 
/dc:agent planner
/dc:roundtable "Monolith or microservices?"
/dc:tech-panel "Redis or Memcached for session caching?"
/dc:planning --team "Ana,Carlos,Luis" --week "2026-03-24"

PoC Mode

Validate ideas before committing to implementation:

/dc:poc --hypothesis "SQLite is sufficient for the MVP"
Phase What
Hypothesis Define what to validate and success/failure criteria
Build Throwaway code, relaxed rules, no TDD
Evaluate Results vs criteria with evidence
Verdict VIABLE / WITH RESERVATIONS / NOT VIABLE / INCONCLUSIVE

If viable → /dc:poc --graduate → full pipeline.

Automatic Estimates

/dc:estimate docs/prd.md

4 complementary models:

Model Technique
Function Points Functional complexity
AI Planning Poker 3 agents estimate independently
COCOMO Estimated LOC → effort
Historical Comparison with similar tasks

Output: optimistic, expected, and pessimistic estimate with breakdown by feature.

Security Audit

/dc:security-audit

Scans all 10 OWASP categories:

  • A01 Broken Access Control — endpoints without auth, IDOR, CORS
  • A02 Cryptographic Failures — passwords in plaintext, JWT without expiration
  • A03 Injection — SQL, XSS, command injection
  • A04-A10 — Configuration, vulnerable components, logging

Each finding includes severity, file, line, and suggested fix.

Stack Migration

/dc:migrate --from "Vue 3" --to "React 19"

6 phases: Inventory → Equivalences → Strategy → Plan → Execution → Verification

Supports: framework (Vue→React), version (Next 14→15), language (JS→TS), paradigm (REST→GraphQL).

Supported Platforms

Don Cheli works natively with multiple AI agents and IDEs. All content adapts to the installed language (ES/EN/PT).

Claude Code — Full support (85+ commands, 43 skills)

What you get

  • CLAUDE.md — Framework instructions (translated to installed language)
  • 85+ slash commands via /dc:* (translated names per locale)
  • /especdev:* — Backward-compatible alias
  • /razonar:* — 15 reasoning models
  • 43 modular skills in habilidades/ (or skills/ in EN)
  • 7 specialized agents (planner, architect, executor, reviewer, tester, documenter, estimator)
  • Full i18n: folder names, file content, rules, commands — all in your language

Commands

/dc:start "Implement JWT auth"       # Start task (auto-detect complexity)
/dc:specify                           # Gherkin spec + DBML
/dc:implement                         # TDD: RED → GREEN → REFACTOR
/dc:debate "Monolith vs microservices" # Adversarial multi-role debate
/dc:estimate docs/prd.md             # 4 estimation models
/razonar:pre-mortem                   # Anticipate failures

Limitations

None. Claude Code is the primary platform with full feature access.

Google Antigravity (Gemini 3.1) — 14 skills, 9 workflows

What you get

  • GEMINI.md — Adapted instructions with Gemini model routing (Flash/Pro) — translated to installed language
  • 14 skills in .agent/skills/ loaded via semantic matching
  • 9 workflows in .agent/workflows/ as slash commands
  • doncheli-skills meta-router for access to all 43 habilidades
  • Model routing: Gemini 3 Flash (default) → Gemini 3.1 Pro (complex tasks)

Skills (14)

Skill What it does
@doncheli-spec Gherkin BDD specifications
@doncheli-plan Technical blueprint
@doncheli-implement TDD execution (RED-GREEN-REFACTOR)
@doncheli-review 7-dimension peer review
@doncheli-security OWASP Top 10 audit
@doncheli-estimate 4 estimation models
@doncheli-debate Adversarial multi-role debate
@doncheli-reasoning 15 reasoning models
@doncheli-migrate Stack migration with wave plan
@doncheli-distill Blueprint Distillation
@doncheli-planning Weekly team planning
@doncheli-tech-panel Senior dev expert table
@doncheli-api-contract API/webhook contract design
@doncheli-skills Router to all 43 habilidades

Workflows

/doncheli-start          # Init or start task
/doncheli-pipeline       # Full spec→review pipeline
/doncheli-estimate       # Run 4 estimation models
/doncheli-debate         # Launch adversarial debate
/doncheli-reasoning      # Apply reasoning model
/doncheli-migrate        # Plan stack migration
/doncheli-planning       # Weekly team planning
/doncheli-review         # Standalone review
/doncheli-security       # Standalone security audit

Limitations

  • No direct access to /dc:* slash commands (use workflows and @skills instead)
  • Skills are invoked via semantic matching, not explicit command names
  • Some advanced features (doctor, capture, guardian) require using the @doncheli-skills router
OpenCode (SST) — Agent + 27 skills via compatibility layer

What you get

  • opencode.json — Configuration pointing to CLAUDE.md and AGENTS.md
  • .opencode/agents/doncheli.md — Primary agent with all 85+ commands
  • 27 skills loaded automatically via .agent/skills/ compatibility layer
  • CLAUDE.md read as fallback instructions (native OpenCode behavior)

How to use

Invoke the Don Cheli agent with @doncheli:

@doncheli start "Implement JWT authentication"
@doncheli estimate docs/prd.md
@doncheli debate "Monolith vs microservices"
@doncheli pre-mortem "Database migration to MongoDB"
@doncheli security-audit
@doncheli tech-panel "Redis vs Memcached for caching"

Or just describe your task — skills activate via semantic matching:

"I need to implement user authentication"
→ OpenCode activates @doncheli-spec automatically

Available commands via @doncheli

Category Commands
Lifecycle start, init, specify, clarify, tech-plan, breakdown, implement, review
Reasoning pre-mortem, 5-whys, pareto, inversion, first-principles, second-order + 9 more
Advanced estimate, debate, tech-panel, planning, security-audit, migrate, drift, tea, pr-review
Quality spec-score, tech-debt, context-health, diagram, changelog-auto

Limitations

  • No native /dc:* slash commands — use @doncheli <command> instead
  • Skills loaded via .agent/skills/ compatibility (not .opencode/skills/)
  • Model routing uses OpenCode's model config, not Don Cheli's agent table
Cursor IDE — Full command reference (.cursorrules)

What you get

  • .cursorrules — Comprehensive instructions (115 lines) — translated to installed language
  • All /dc:* commands documented with descriptions
  • All /razonar:* reasoning models documented
  • 6 quality gates, complexity levels, pipeline reference
  • Iron laws enforced

How to use

Commands work as prompts in Cursor's AI chat:

/dc:start "Implement JWT auth"
/dc:specify
/dc:implement
/razonar:pre-mortem "What could go wrong with this migration?"

Limitations

  • Commands are prompt-based (not native slash commands like in Claude Code)
  • No .agent/skills/ semantic matching (Cursor doesn't support Antigravity's skill format)
  • No specialized agents (planner, architect, etc.) — Cursor uses a single model
  • Skills/habilidades must be referenced manually, not auto-loaded
Other agents (Codex, Amp, etc.)

What you get

  • AGENTS.md — Cross-tool instructions
  • prompt.md — Minimal generic instructions for any AI agent
  • All rules and habilidades are standard Markdown — any agent can read them

How to use

Point your agent to read CLAUDE.md or prompt.md at the start of a session. The framework's files are standard Markdown, so any agent that can read files can use Don Cheli.

Limitations

  • No native slash command integration
  • No semantic skill matching
  • Manual workflow — you need to tell the agent which files to read
  • No model routing optimization

Project Structure

don-cheli/
├── comandos/
│   ├── especdev/          # 71 /dc:* commands
│   └── razonar/           # 15 /razonar:* commands
├── habilidades/           # 42 modular skills
├── reglas/
│   ├── constitucion.md    # 8 governing principles
│   ├── leyes-hierro.md    # 3 non-negotiable laws
│   ├── puertas-calidad.md # 6 quality gates
│   ├── i18n.md            # Internationalization rules
│   ├── skills-best-practices.md  # Anthropic Skills 2.0 best practices
│   └── reglas-trabajo-globales.md
├── locales/               # 🌍 i18n strings
│   ├── es.json            # Spanish (158 strings)
│   ├── en.json            # English (158 strings)
│   └── pt.json            # Portuguese (158 strings)
├── plantillas/
│   └── especdev/
│       ├── es/            # Templates in Spanish
│       ├── en/            # Templates in English
│       └── pt/            # Templates in Portuguese
├── agentes/               # 7 specialized agents
├── ganchos/               # Pre/Post tool + Stop hooks
├── scripts/               # instalar.sh, bucle.sh, validar.sh
├── .opencode/             # 🟢 OpenCode (SST) compatibility
│   └── agents/            # @doncheli agent
├── .agent/                # 🔮 Antigravity/Gemini compatibility
│   ├── skills/            # 14 skills (spec, plan, implement, review, security, estimate, debate, reasoning, migrate, distill, planning, tech-panel, api-contract, skills-router)
│   └── workflows/         # 9 workflows (start, pipeline, review, security, estimate, debate, reasoning, migrate, planning)
├── CLAUDE.md              # Instructions for Claude Code
├── GEMINI.md              # Instructions for Google Antigravity
├── AGENTS.md              # Cross-tool instructions (Cursor, Codex)
├── prompt.md              # Instructions for Amp
├── NOTICE                 # Attributions
└── LICENCIA               # Apache 2.0

After installing with a language, the installed structure uses localized names:

~/.claude/don-cheli/          # Global installation
├── skills/                   # (or habilidades/ in ES, habilidades/ in PT)
├── rules/                    # (or reglas/ in ES, regras/ in PT)
├── templates/                # (or plantillas/ in ES, modelos/ in PT)
├── hooks/                    # (or ganchos/ in ES/PT)
├── agents/                   # (or agentes/ in ES/PT)
├── locales/                  # es.json, en.json, pt.json
├── locale                    # 2-letter file: "es", "en" or "pt"
├── folder-map.json           # Name mapping for Claude
├── CLAUDE.md
└── VERSION

Philosophy

"Context Window = RAM, File System = Disk"

  1. Persistence over conversation — Write it down, don't just say it
  2. Structure over chaos — Clear files, clear roles
  3. Recovery over restart — Never lose progress
  4. Evidence over assertions — Show, don't tell
  5. Simplicity over complexity — Everything in your language

Contributing

See CONTRIBUTING for the complete guide.

# Fork → Clone → Branch → Changes → PR
git checkout -b feature/my-improvement

Community & Support

Channel Purpose
GitHub Discussions Questions, ideas, show & tell
GitHub Issues Bugs and feature requests
YouTube @doncheli Tutorials and video demos
Instagram @doncheli.tv Updates and quick tips

Found a bug?

  1. Search existing issues
  2. If it doesn't exist, open one with the bug report template
  3. Include: framework version, AI agent, and steps to reproduce

Changelog

See CHANGELOG.md for the full history of changes.

License

Apache 2.0 — Copyright 2026 Jose Luis Oronoz Troconis (@DonCheli)

You may use, modify, and distribute Don Cheli freely. You must maintain attribution to the original author and indicate any changes made.

Stop guessing. Start Engineering.
Vibe coding is the spark; SDD is the engine. Transition from AI-assisted chaos to professional software delivery.
Don Cheli — SDD Framework

Yorumlar (0)

Sonuc bulunamadi