Skill Awesome - Framework de Engenharia de IA
vaibhav-deveree/skill-awesome
Sistema operacional de engenharia com agentes autônomos e memória SQLite para codebases.
Ver no GitHubMCP server que transforma assistentes de IA em um sistema operacional de engenharia disciplinado, com agentes especializados (Master Architect, Vibe Coder), mecanismo de memória SQLite de zero alucinação, e integração nativa com Git e Claude Code. Ideal para orquestrar codebases massivas mantendo contexto perfeito e aplicando padrões arquiteturais rigorosos.
Instalação
npx convez-ai vaibhav-deveree/skill-awesome1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Onboarding e refatoração de projetos legados sem quebras
- Geração de UI/mockups interativos com vibe-first approach
- Orquestração de arquitetura e revisão automática de PRs com God Review
Veredito

Solução ambiciosa com foco em eliminar alucinações via SQLite memory e aplicar disciplina arquitetural em projetos grandes. Excelente para times que querem IA como co-pilot estruturado, mas requer investimento inicial em configuração e adoção de convenções rigorosas; não é plug-and-play para iniciantes.
README
A tradução para o português deste README ainda está na fila.
skill-awesome: The Ultimate AI Engineering Framework
skill-awesome is a collection of production-ready engineering skills, architecture standards, and autonomous AI subagents designed to turn any standard AI coding assistant (like Claude, Gemini, or ChatGPT) into a highly disciplined, zero-hallucination Engineering Operating System (EOS).
🌟 What It Does
This plugin isn't just a list of guidelines; it provides active agent behaviors and an ultra-fast SQLite Memory Engine to orchestrate massive codebases without blowing up your token limits.
1. The Agents
When you install this plugin, you get access to specialized agents:
- Master Architect: A disciplined orchestrator that enforces mandatory implementation plans, centralizes architecture decisions, and delegates tasks to specialized subagents.
- Vibe Coder: An intuitive, aesthetics-first frontend co-pilot. It handles all boilerplate autonomously and is hardcoded to prioritize modern UI/UX design (glassmorphism, micro-animations).
2. The SQLite Memory Harness (Zero Hallucination)
- Rather than reading huge Markdown files that cause token limits to explode, agents inject an ultra-fast SQLite database engine (
db_engine.py) into your project. - It perpetually tracks User Intent, Current State, History, and a granular Changelog.
- Agents use raw SQL (
gh query "SELECT * FROM state") to fetch only the context they need, ensuring perfect memory forever.
3. Brownfield Project Onboarding
- Drop the agent into a 10-year-old codebase, and it won't break anything.
- It maps the existing directory structure and architecture to the SQLite database.
- It proposes architectural and security upgrades in a dedicated
implementation_plan.mdbefore modifying your old code.
4. Interactive UI Mockup Workflow (Phase Zero)
- When generating a new website or feature, the Vibe Coder will pause and generate a standalone
mockup.html. - This HTML file is a visual mood board presenting color palettes, button styles, and layout variations.
- You click and choose your exact "vibe" before the heavy framework code (React/Vue/Next.js) is even written.
5. Context-Aware Git & PR Workflow
- Conventional Commits: Agents strictly enforce standard Git branching (
feat/,fix/) and Conventional Commits. - Conversational Commits: Agents won't spam PRs on every prompt. They explicitly ask you: "Should I commit these changes?"
- Autonomous God Review: Once approved, the agent automatically creates a Pull Request via the GitHub CLI (
gh pr create) and triggers a multi-agent "God Review" covering security, architecture, and performance on the diff.
6. Native Claude Code Integration
- Slash Commands: Adds native
/master-architect,/vibe-coder, and/god-reviewcommands directly into your Claude Code terminal. No need to copy/paste long prompts. - Global Context Injection: Uses a
SessionStarthook to invisibly inject theengineering-principlesinto Claude's memory the moment the terminal opens. Claude always knows your PR rules. - Cross-Platform MCP Engine: Built-in prompt constraints force the agent to format Windows paths correctly (
C:\), permanently fixing the notorious Git Bash MCP crashing bug. - Inline Token Telemetry: Agents automatically append an
Execution Footerto their responses, providing an inline mathematical estimation of input/output tokens used per transaction.
🚀 How to Install
This repository is structured as a Gemini/Claude Plugin.
Option 1: The AI Prompt (For Native Gemini Agents)
Copy and paste this exact prompt to your AI agent (like Gemini):
"Please install the
skill-awesomeplugin fromhttps://github.com/vaibhav-deveree/skill-awesome.gitinto my agent environment so I can use the Master Architect and Vibe Coder agents."
[!WARNING] For Claude CLI Users: Please see Option 2 below. You must add this repository as a custom marketplace first.
Option 2: Claude CLI Installation (Marketplace)
Because Claude Code uses a decentralized marketplace system, you must add this repository as a registry before installing the plugin:
claude plugins marketplace add vaibhav-deveree/skill-awesome
claude plugins install skill-awesome-plugin
Option 3: Global Installation (Manual)
- Clone this repository.
- Copy the entire folder into your global plugins directory:
C:\Users\<your_user>\.gemini\config\plugins\skill-awesome
Option 4: Project-Local Installation
- Navigate to your specific project's root folder.
- Clone this repo into the local plugins directory:
mkdir -p .gemini/plugins cd .gemini/plugins git clone https://github.com/vaibhav-deveree/skill-awesome.git
📊 Benchmarking & Observability (LangSmith)
To benchmark how the agents perform (tracking token usage, execution time, and cost of every /master-architect command), you should integrate LangSmith alongside this framework.
Because skill-awesome acts as your prompt framework, it natively pairs with the official LangSmith tracing plugin.
To install the telemetry tracer in Claude Code:
claude plugins marketplace add langchain-ai/langsmith-claude-code-plugins
claude plugins install langsmith-tracing@langsmith-claude-code-plugins
Configure your .env or settings file with your LangSmith keys, and every action performed by the Vibe Coder or Master Architect will be fully visible and benchmarked in your dashboard!
🗣️ Commands & Skill Triggers
skill-awesome provides two ways to interact with the framework: Native Slash Commands (for Claude Code CLI) and Natural Language Triggers (for Gemini / Antigravity).
How to Invoke Agents
🤖 In Gemini / Antigravity (Natural Language)
Because Gemini and Antigravity are natural language conversational agents, you do NOT need to use slash commands. You can invoke any subagent by simply asking the agent in plain English.
- "Use the
@master-architectto plan this feature." - "Trigger the
@qa-testerto validate my API." - "Run the
@code-revieweragent on my PR." The agent will automatically read the corresponding skill and use its nativeinvoke_subagenttool to execute the workflow.
💻 In Claude Code (Slash Commands)
If you are using Claude Code or a compatible CLI, you can instantly switch the active agent profile using these commands:
| Command | What it does | When to use it |
|---|---|---|
/master-architect | Invokes the Master Architect agent. | For backend, system design, architectural planning, and complex multi-agent orchestration. |
/vibe-coder | Invokes the Vibe Coder agent. | For frontend development, UI/UX implementation, and aesthetics-first coding. |
/god-review | Invokes the God Review pipeline. | Run this autonomously on a diff to review security, architecture, and UI slop before merging a PR. |
/code-reviewer | Invokes the Enterprise Code Review Orchestrator. | Use to spawn a multi-agent team (3-5 agents) for inline checks, cross-file checks, and final verification boards. |
/qa-tester | Invokes the QA Tester agent. | Use to deeply analyze a feature, test all success/failure/edge case flows, and generate a total mockup report. |
/goal | Invokes Claude Code's native continuous loop mode. | Use when you want agents to work autonomously until a specific condition is met (e.g., all tests pass). |
Natural Language Skill Triggers
Agents are equipped with dozens of passive "Skills" that sit in the background. To activate them, simply include their trigger words in your prompt:
| Skill | Trigger Phrase | What it does |
|---|---|---|
| Web Quality Audit | "run lighthouse audit", "check page quality", "audit my site" | Executes a massive 150+ check audit against Core Web Vitals, SEO, and Accessibility standards. |
| Transitions.dev | "add a transition", "make the modal open smoothly", "stagger animation" | Injects production-ready, zero-dependency CSS micro-animations into your frontend. |
| Clean SaaS UI | "build a clean UI", "SaaS look", "professional design" | Enforces the frontend-design-skill rules to build structured, high-utility, glassmorphism-free interfaces. |
| Creative UI | "bold design", "maximalist chaos", "out of the box" | Bypasses the SaaS rules and activates the frontend-creative-design skill for highly unique, experimental UI. |
| Scoped Review | "check for bugs", "verify the fix", /scoped-review | Enforces a strict, file-by-file code review limited ONLY to the files or diff you specify. Prevents scope creep and stops the agent from scanning unrelated files. |
| Memory Harness | (Automatic) | Agents automatically run SQLite commands (log_state, log_history) in the background to track project memory. |
🛠️ How to Use It
Once installed, the framework runs natively in your agent workspace.
To invoke the Master Architect: Use the native slash command to instantly switch personas:
/master-architect Review my project and generate a new feature implementation plan.
To enforce the Strict Iterative Development Workflow (Task-by-Task): If you want to ensure zero bugs accumulate during a large feature, combine the master orchestrator with the strict iteration rule:
/master-architect We need to build the checkout flow. Enforce the Strict Iterative Development Workflow: write code for one task at a time, and run a mandatory God Review after each task before moving forward.
To do aesthetics-first frontend development:
/vibe-coder Build a modern dashboard for this project with glassmorphism.
To trigger an autonomous God Review on a PR:
/god-review
To run an autonomous loop until a condition is met (Goal Mode):
Combine the native /goal command with a subagent to let the agent work completely autonomously until verifiable success is achieved:
/goal The test suite in test/api passes using the @master-architect agent
To interact with the Memory Harness manually: You can always view your project's memory by running the python CLI the agents use:
python .memory/db_engine.py query "SELECT * FROM state"
💡 Copy & Paste Sample Prompts
To get the most out of the framework and its specialized agents, you can simply copy and paste these prompts directly into your AI assistant:
For New Feature Development
"I want to build a new feature. Use the
@master-architectagent to analyze the current system, write an implementation plan to.memory/plans/, and orchestrate the backend and frontend development using the required subagents."
For Strict Iterative Development (Bug-Free Execution)
"Use the
@master-architectand enforce the Strict Iterative Development Workflow. You must implement exactly ONE task at a time. After writing the code for a task, pause and invoke a God Review on that specific task. Ensure all frontend edge cases (routing, state) and backend edge cases (timeouts, load) pass before you move on to the next task."
For Frontend & UI Tasks
"Use the
@vibe-coderagent to build a new clean SaaS UI for the dashboard. Generate the interactive Phase Zeromockup.htmlfirst so I can select the exact vibe before we write any framework code."
For Deep Code Reviews
"Run the Enterprise Code Review on my recent changes using the
@code-revieweragent. Spawn a team of 3-5 specialized subagents to perform inline checks and cross-file validation, then present the Final Executive Report."
For Scoped, File-by-File Reviews
"Run a
/scoped-reviewonauth_service.pyanduser_controller.py. Check for bugs and security issues in these files only.""Verify the fix in my recent changes. Do a scoped review of the diff to ensure there are no unhandled exceptions."
For QA Testing & Edge Case Simulation
"Use the
@qa-testeragent to test the new authentication API. Read the source code to acquire deep knowledge, and perform a static dry run tracing the execution path from Controller to Repository for each API. Match the exact return values against the happy/unhappy scenarios, simulate edge cases, and generate the Total Mockup Report."
For Goal Mode (Autonomous Loops)
"/goal All tests in
test/api/pass and the lint step is clean using the@master-architectagent."
🤝 How to Contribute
Contributions are highly encouraged! We want to build the most comprehensive library of AI engineering protocols.
Contribution Guidelines
- Never merge directly to
main. The repository is protected by strict branch protection rules. - Branch Naming: Use
feat/<name>,fix/<name>, orchore/<name>. - Commit Messages: Use Conventional Commits (
feat(core): add new memory table). - Create a Pull Request: Submit a PR.
- The God Review: Your PR must pass an automated "God Review" and receive at least 1 approving review from a maintainer before it can be merged.
Examples of great contributions:
- New specialized skills (e.g.,
cloud-native-skill,ai-agent-testing-skill). - Improvements to the SQLite Memory Harness.
- New Agent Profiles.
🏢 About Zero Assist
Many of these autonomous skills and PR workflows were developed, refined, and battle-tested while building Zero Assist.
Zero Assist is a next-generation engineering platform. By leveraging the advanced multi-agent architecture and rigorous engineering standards found in this repository, Zero Assist delivers an unparalleled experience for engineering teams.
If you want to see these AI engineering protocols in action in a massive production environment, check out Zero Assist!
📜 Philosophy
Every skill in this repository follows four core principles:
- Evidence Over Assumptions: Code generation is based on the SQLite DB facts.
- Practical Over Theoretical: Focus on realistic production issues.
- Reusable Over Project Specific: The DB engine and skills work on ANY tech stack.
- Validation Over Opinion: Code must pass the God Review before merging.
Feel free to use, modify, extend, and adapt these skills for your own engineering workflows.