convez.ai
MCP Server
NafisRayan

100x-Agent-Toolkit

NafisRayan/100x-Agent-Toolkit

Ver no GitHub
InicialPluginScore · de 0 a 100

A production-grade engineering toolkit for AI-assisted software development. Contains 168 specialized skill workflows (100 core + 68 GSD sub-skills), 142 expert agent personas, 24 design system specifications, 5 reference checklists, 84 slash commands, and 9 MCP server integrations.

Instalação

$npx convez-ai NafisRayan/100x-Agent-Toolkit

1 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

Estrelas
1
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 1 semana
Versão
—
Issues abertas
0
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

100x Agent Toolkit

A production-grade engineering toolkit for AI-assisted software development. Contains 168 specialized skill workflows (100 core + 68 GSD sub-skills), 142 expert agent personas, 24 design system specifications, and 9 MCP server integrations — all designed to enforce professional engineering standards across the full development lifecycle: Define → Plan → Build → Verify → Review → Ship → Scale.


Architecture

The system uses a two-layer composition model:

LayerWhat it isExampleRole
SkillA workflow with steps and exit criteriatest-driven-developmentThe how
PersonaA role with a perspective and output formatcode-reviewerThe who

Personas don't call other personas. The orchestrator skill handles composition.

Important: Skills carry 99% of the system's value. The agents/ directory (142 personas), rules/ directory (104 language-specific rules across 18 languages), references/ directory (5 domain checklists), docs/ directory (3 guides), and commands/ directory (86 slash commands) were all converted into two skills — agent-personas and syntax-rules — because skills provide auto-triggering, progressive loading, on-demand references, and universal compatibility across any AI tool. The rest was redundant with existing skills or no longer needed. If you're utilizing or maintaining this toolkit, prioritize skills first.


Directory Structure

./
├── AGENTS.md              # System prompt — the master configuration
├── README.md              # This file
│
├── skills/                # 168 specialized skill workflows (100 core + 68 GSD sub-skills)
│   └── <skill-name>/
│       ├── SKILL.md       # Skill definition (YAML frontmatter + instructions)
│       ├── references/    # Domain-specific documentation (includes 142 agent personas)
│       ├── scripts/       # Automation scripts (Python, JS, Shell)
│       ├── assets/        # Starter projects, templates, images
│       └── evals/         # Evaluation test cases
│   └── ...
│
├── mcps/
│   └── mcp.json           # 9 MCP server configurations
│
└── designs/
    ├── DESIGN-figma.md    # Figma-inspired design system
    ├── DESIGN-airbnb.md   # Airbnb design system
    ├── DESIGN-apple.md    # Apple design system
    ├── DESIGN-stripe.md   # Stripe design system
    ├── DESIGN-notion.md   # Notion design system
    ├── DESIGN-vercel.md   # Vercel design system
    └── ... (24 total: airbnb, apple, binance, bmw, claude, cursor, expo, figma, framer, meta, minimax, mongodb, notion, nvidia, pinterest, revolut, stripe, supabase, superhuman, uber, vercel, warp, webflow, xai)

Quick Start

This repository is a skill bundle — it configures your AI agent (Claude Code, Cursor, Copilot, etc.) with domain expertise, workflows, and quality gates.

  1. Clone the repository and rename the folder to .agents, then place it in your home directory:
    • Windows: C:\Users\<YourUsername>\.agents
    • macOS / Linux: ~/.agents
  2. The AGENTS.md file is the entry point — your agent reads it on startup and gains access to all skills and personas.
  3. Skills drive the lifecycle through auto-triggering:
    • brainstorming → define what to build
    • planning-and-task-breakdown → plan how to build it
    • incremental-implementation → implement incrementally
    • test-driven-development → verify it works
    • requesting-code-review → quality assurance
    • finishing-a-development-branch → deploy to production

Skills Catalog

Define & Plan

SkillDescription
brainstormingExplore intent and requirements before implementation
planning-and-task-breakdownDecompose specs into verifiable tasks
pro-task-breakdownGenerate structured task breakdown CSVs from features/epics
writing-plansCreate detailed implementation plans
executing-plansExecute written plans with checkpoints

Build & Code

SkillDescription
incremental-implementationThin vertical slices with feature flags
test-driven-developmentRed-Green-Refactor with test pyramid
source-driven-developmentGround decisions in official documentation
subagent-driven-developmentExecute plans with independent sub-agents
karpathy-guidelinesBehavioral guidelines to reduce LLM coding mistakes

Web & Frontend

SkillDescription
nextjs-developerNext.js 14+ App Router, RSC, Server Actions
next-best-practicesFile conventions, RSC boundaries, data patterns
vercel-react-best-practicesReact/Next.js performance optimization
shadcnComponent management for shadcn/ui
motionReact animations with Motion (bundle-optimized)
motion-framerReact animations (Framer Motion / Motion)
gsapGSAP animation reference and patterns
modern-web-design2024-2025 design trends and implementation
tailwind-design-systemTailwind CSS design tokens and component libraries
responsive-designContainer queries, fluid typography, grid
web-design-guidelinesUI code review against design best practices
web-quality-auditComprehensive web quality audit
ui-ux-pro-max50+ styles, 161 color palettes, 57 font pairings
scroll-animationsScroll-triggered animations and parallax
21st-dev-builder-v2Build with 1400+ shadcn/ui components
impeccableCritique, polish, and improve frontend interfaces
design-systemDesign-system driven by specification files
accessibilityWCAG 2.2 audit and compliance
building-native-uiExpo Router native app development
mobile-developerReact Native and Expo mobile development
vercel-react-native-skillsReact Native performance and best practices
muiMaterial-UI v7 component patterns

Automation & Workflow

SkillDescription
n8n-code-javascriptJavaScript Code node patterns for n8n workflows

Backend & API

SkillDescription
fastapi-expertAsync Python APIs with FastAPI + Pydantic V2
nodejs-express-serverProduction-ready Express.js servers
nodejs-backend-patternsExpress/Fastify middleware, auth, API design
nodejs-best-practicesNode.js development principles and decision-making
nestjs-best-practicesNestJS architecture and DI patterns
shopify-expertShopify themes, apps, and Storefront API

Database

SkillDescription
prisma-proSchema design, migrations, query optimization, relations
sql-proComplex joins, window functions, indexing
supabase-postgres-best-practicesPostgres performance from Supabase
neon-postgresServerless Postgres with Neon

Cloud & Infrastructure

SkillDescription
cloudflareWorkers, Pages, D1, R2, KV, AI, Agents SDK
docker-expertMulti-stage builds, orchestration, security hardening
viteBuild tool config, plugin API, SSR, Rolldown
migrate-to-vinextMigrate Next.js projects to vinext (Vite-based)

Quality & Testing

SkillDescription
systematic-debuggingStructured bug investigation
verification-before-completionRun commands before claiming success
receiving-code-reviewHandle feedback with technical rigor
requesting-code-reviewGet quality feedback before merge
performanceMeasure-first Core Web Vitals improvement
core-web-vitalsOptimize LCP, INP, CLS
full-code-reviewDeep multi-agent code review across codebase
ultrareviewMulti-agent parallel code review
playwright-cliPlaywright CLI automation, testing, and session management

3D & Graphics

SkillDescription
3d-web-experienceThree.js, React Three Fiber, Spline, WebGL
threejs-webglInteractive 3D scenes, materials, lighting
algorithmic-artGenerative art and creative coding

Documentation

SkillDescription
docs-writingDiataxis framework, Stripe-style clarity
code-wikiGenerate wiki docs + Mermaid diagrams for codebase
doc-coauthoringStructured workflow for co-authoring docs
mermaidFlowcharts, sequence diagrams, C4 architecture
design-mdSynthesize design systems into DESIGN.md files
humanizerRemove AI writing patterns, make text sound natural
SkillDescription
--------------------
docxWord document generation and manipulation
pptxPowerPoint deck creation and editing
xlsxSpreadsheet creation, editing, data cleaning
pdfPDF read, merge, split, OCR, form filling

Business

SkillDescription
business-analystProduct discovery and requirements analysis
sales-enablementSales collateral, pitch decks, demo scripts
marketing-psychologyBehavioral science applied to marketing
brag-documentWork accomplishments reports and self-reviews
internal-commsInternal communications writing

Design & Creativity

SkillDescription
figma-developerFigma to React component conversion
canvas-designVisual art in .png and .pdf documents
algorithmic-artGenerative art and creative coding with p5.js
theme-factoryStyling artifacts with pre-set themes

Languages

SkillDescription
javascript-proModern ES2023+ JavaScript patterns
typescript-advanced-typesGenerics, conditional types, mapped types
python-performance-optimizationProfile and optimize Python code
syntax-rulesLanguage-specific coding standards for 18 languages

SEO

SkillDescription
seoSearch engine visibility optimization
seo-auditDiagnose SEO issues on websites

Meta & Dev Tools

SkillDescription
enhance-promptTransform vague UI ideas into polished prompts
skill-creatorBuild new skills with evaluation
find-skillsDiscover skills for specific tasks
agent-browserBrowser automation for AI agents
mcp-builderBuild MCP server integrations
context7Retrieve up-to-date library docs
duckduckgo-searchWeb search without API keys
using-git-worktreesIsolated workspace for feature work
using-superpowersFind and use skills effectively
dispatching-parallel-agentsRun independent tasks concurrently
finishing-a-development-branchGuide completion of dev work
clone-websiteReverse-engineer and clone websites from URLs
cavemanUltra-compressed communication mode
grill-meInterview user about a plan or design
gsdGet Shit Done project management (68 sub-skills)
agent-personas142 expert agent personas across 23 categories

MCP Server Integration

The system connects to 9 external tools via Model Context Protocol (mcps/mcp.json):

ServerPurposeStatus
@vkhanhqui/figma-mcp-goFigma design integrationConfigured
@21st-dev/magic21st.dev component marketplaceConfigured
playwrightBrowser automationConfigured
chrome-devtoolsChrome DevTools accessConfigured
context7Upstash Context7 for library docsConfigured
storybook-mcpStorybook component explorerRequires URL
firecrawlWeb crawlingRequires API key
typescript-lspTypeScript language serverConfigured
vercelVercel deploymentRequires API token

Adding a New Skill

  1. Create skills/<skill-name>/SKILL.md with YAML frontmatter:
    ---
    name: your-skill-name
    description: What triggers this skill
    ---
    
  2. Add workflow steps, verification gates, and anti-rationalization sections.
  3. Optionally add references/, scripts/, assets/, or evals/ subdirectories.
  4. Register the skill in AGENTS.md under the appropriate category.
  5. Use the skill-creator skill for detailed guidance.

Migration Log — Folders Converted to Skills

During the shift to a skill-first architecture, several top-level directories were absorbed into skills. Skills are the better home for this content because they get progressive loading (metadata first, full content on trigger), automatic context-based activation, bundled references that load only when needed, and a single skills/ entry point instead of scattered root-level directories.

agents/ → skills/agent-personas/ (converted)

142 agent persona .md files across 23 categories (core, development, data-ai, infrastructure, quality-testing, security, reviewers, build-resolvers, specialized, orchestrators, etc.) moved into skills/agent-personas/references/.

Why skills work better: Agent personas were standalone .md files in a directory tree with no triggering mechanism — they only activated when manually referenced or via commands. As a skill, the personas get a unified SKILL.md that explains when and how to use each agent, auto-triggers on agent-related queries ("which agent should I use", "set up a code review team"), and bundles the full persona catalog as reference files that load on demand instead of sitting in context permanently.

rules/ → skills/syntax-rules/ (converted)

104 rule files across 18 languages (TypeScript, Python, Rust, Go, Java, Kotlin, Swift, C++, C#, PHP, Dart, Angular, ArkTS, F#, Perl, web, zh) plus 10 common universal rules moved into skills/syntax-rules/references/.

Why skills work better: Rules lived in a flat directory tree with no triggering — they had to be manually installed into ~/.claude/rules/ and were always in context even when writing a different language. As a skill, the rules auto-trigger only when writing or reviewing code in the relevant language, load language-specific rules on demand (no need to keep all 18 languages in context), and include a structured workflow (detect language → load rules → apply → report violations) that the raw directory never had.

references/ (deleted)

5 domain checklists (security-checklist.md, testing-patterns.md, performance-checklist.md, accessibility-checklist.md, orchestration-patterns.md) removed.

Why deleted: These were static markdown files that required manual lookup and had no triggering mechanism. Their content is now redundant — every topic they covered is handled more thoroughly by dedicated skills (security skill for security checks, test-driven-development for testing patterns, performance for performance optimization, accessibility for WCAG compliance, and dispatching-parallel-agents for orchestration patterns). Skills provide the same information plus actionable workflows, verification gates, and auto-triggering.

docs/ (deleted)

3 documentation files (best-practices.md, creating-agents.md, dependencies.md) removed.

Why deleted: These were one-time setup guides that don't benefit from being in context during normal operation. best-practices.md and creating-agents.md were about how to configure the agent system itself — useful for humans setting up the repo, not something the AI needs loaded at runtime. dependencies.md was a Context7 MCP setup instruction that's now covered by the context7 skill. Keeping these as root-level docs meant they were always visible but rarely useful — the skill-based approach loads equivalent guidance only when relevant.

commands/ (deleted)

86 slash command definitions (/spec, /plan, /build, /test, /review, /ship, etc.) removed.

Why deleted: Slash commands were a convention specific to Claude Code's .claude/commands/ directory structure — they don't port to other AI systems (Cursor, Copilot, Windsurf, etc.) and required manual installation into ~/.claude/commands/. The workflows these commands encoded — spec → plan → build → test → review → ship — are already captured as skills (brainstorming, planning-and-task-breakdown, incremental-implementation, test-driven-development, best-practices, finishing-a-development-branch). Skills are universal, self-documenting, and work across any AI tool that reads a markdown file. Commands were thin orchestration wrappers around skills; removing them eliminates a layer of indirection without losing any capability.

Skills removed

SkillReason
documentation-writerOverlapped with docs-writing (both Diátaxis-based). Consolidated into the more comprehensive docs-writing skill which has 52 rules across 9 categories.
sql-optimization-patternsOverlapped with sql-pro which already covers query optimization, EXPLAIN analysis, and indexing strategies more comprehensively.
mirror-websiteEmpty skill directory with no SKILL.md — never had content.
prisma + prisma-expertOverlapped significantly. Consolidated into a single prisma-pro skill covering schema design, migrations, query optimization, and relations modeling.

Design Philosophy

  • Progressive Disclosure — Skill metadata loads first; full content loads on demand.
  • Verification Gates — Every skill includes proof steps. Never skip "prove it works."
  • Anti-Rationalization — Skills rebut common excuses to enforce discipline.
  • No Meta-Orchestration — Personas don't call other personas. Commands handle composition.
  • Quality Is Not Optional — Skills enforce standards used by top engineering organizations.

License

Refer to individual skill directories for license information. Several skills include LICENSE.txt files specifying their terms.

Ver o original no GitHub

Bom para

Relacionados