Universal AI Agent Toolkit — 82 agents, 147 skills, 5 MCP servers (50 tools) (50 tools), 15 core modules.
82 agents · 147 skills · 5 MCP servers · 15 core modules · 11 platform adapters.
Quick Start · CLI Reference · Architecture · Configuration · Publishing to npm · Contributing
Super Kit is a universal AI agent toolkit that gives any AI coding assistant — Claude Code, Cursor, Windsurf, Cline, Roo Code, Gemini CLI, Codex, OpenCode, Aider, Continue, or any MCP-compatible platform — instant access to 82 specialized agents organized into 5 teams, with 147 skills and 5 MCP servers for real-world engineering tasks.
The package is designed for local installation — it is distributed as a tarball you build from source and npm install into any project that needs it.
- 82 specialized agents across 5 teams: DEV (development), QA (quality assurance), DATA (data engineering), BIZ (business/marketing), DESIGN (design)
- 147 skills covering architecture, code review, testing, analytics, ETL, performance, security, accessibility, and more
- 313 agent shortcuts for fast lookup and slash-command generation
- YAML-based agent definitions — extend without touching code
| Platform | Adapter | Auto-configures |
|---|---|---|
| Claude Code | claude/ |
CLAUDE.md, .mcp.json, slash commands |
| Cursor | cursor/ |
.cursorrules, MCP servers |
| Windsurf | windsurf/ |
.windsurfrules |
| Cline | cline/ |
.clinerules, MCP config |
| Gemini CLI | gemini/ |
GEMINI.md, .gemini/settings.json |
| Codex | codex/ |
Codex config |
| OpenCode | opencode/ |
OpenCode settings |
| Aider | aider/ |
CONVENTIONS.md, .aider.conf.yml |
| Continue | continue/ |
.continue/config.json |
| Generic MCP | generic/ |
Any MCP-compatible client |
- Model registry with 10+ models across 4 tiers (premium / standard / economy / local)
- Smart router with 4 strategies:
cost,quality,speed,balanced - Budget tracking with daily/monthly limits and automatic downgrade on budget pressure
- Cost-tracker records every call (model, tokens, cost, latency) for analytics
- filesystem — safe file I/O with path validation, glob search, YAML/JSON read/write
- git — branch, diff, log, commit, status, blame, stash (read-only by default)
- database — SQLite / Postgres / MySQL query with schema introspection
- browser — headless Chromium with screenshots, navigation, interaction
- vietnam — ⭐ Unique! Zalo messenger integration, Vietnamese NLP, admin data lookup, VND currency conversion
💡 The Vietnam MCP server makes Super Kit the only AI toolkit with native Vietnamese platform support — perfect for projects targeting the Vietnamese market.
Visualize your Super Kit installation in a single self-contained HTML page:
npx tsx cli/src/index.ts dashboard # Generate + open in browser
npx tsx cli/src/index.ts dashboard -o out.html # Save to a specific file
npx tsx cli/src/index.ts dashboard --agents-onlyThe dashboard command (cli/src/commands/dashboard.ts) walks core/agents/ and core/skills/, parses each YAML manifest, and renders a static HTML page showing:
- Stats — total agents, skills, teams, and skill categories
- Team breakdown — colored badges per team (dev / qa / data / biz / design)
- Agent cards — name, role, description, and keyboard shortcut
- Skill cards — grouped by category
- Footer — version + GitHub link
Output is saved to ~/.super-kit/dashboard.html by default and opened with xdg-open / open / start. The HTML is fully self-contained (inline CSS, no JavaScript, no external assets), so you can host it on GitHub Pages, email it, or share it as a single file.
💡 Perfect for onboarding new contributors or showing stakeholders what your Super Kit installation contains — no server, no build step, just one HTML file.
core/
├── orchestrator/ # Task routing + workflow DAG execution
├── cost-arbitrage/ # Multi-model registry, smart router, cost tracker
├── marketplace/ # Plugin registry (publish, search, install, rating)
├── memory/ # Persistent agent memory (in-memory + JSONL)
├── schemas/ # Agent/Skill/Workflow YAML schema validation
├── types/ # Shared TypeScript types
├── multi-model/ # Provider adapters (anthropic, openai, google, ...)
├── collaboration/ # Inter-agent messaging (pub/sub, channels)
├── benchmark/ # Performance benchmarks
├── observability/ # Tracing, logging, health checks
└── ... # 6 more modules
Super Kit is installed from a local tarball. Build the tarball once from this repo, then npm install it into any project.
git clone https://github.com/qkhalk/super-kit.git
cd super-kit
npm install
npm run build
npm pack
# → creates super-kit-3.1.0.tgz (~800 KB)# In any project that needs Super Kit:
npm install /absolute/path/to/super-kit-3.1.0.tgzFor convenience, copy the tarball somewhere stable (e.g. ~/packages/) and reference it from there:
mkdir -p ~/packages
cp super-kit-3.1.0.tgz ~/packages/
# In a project:
npm install ~/packages/super-kit-3.1.0.tgznode -e "const {MemoryStore}=require('super-kit/core/memory');const m=new MemoryStore();m.save('check','ok');console.log(m.recall('check'))"
# → okRequires Node.js ≥ 18.
import { MemoryStore } from 'super-kit/core/memory';
import { SchemaValidator } from 'super-kit/core/schemas';
import { SmartRouter, ModelRegistry } from 'super-kit/core/cost-arbitrage';
import { TaskRouter, WorkflowEngine } from 'super-kit/core/orchestrator';
// 1. Validate agent YAML definitions
const validator = new SchemaValidator();
const results = validator.validateAllAgents('./agents');
console.log(validator.getSummary(results));
// 2. Store and recall memory
const memory = new MemoryStore();
memory.save('arch-choice', 'Use microservices', { type: 'decision' });
console.log(memory.recall('arch-choice'));
// 3. Route a task to the best agent
const taskRouter = new TaskRouter(/* agent loader, intent classifier */);
const decision = taskRouter.routeByIntent('design a rate limiter');
console.log(decision[0].agent.id, decision[0].confidence);
// 4. Pick the cheapest model that meets a quality bar
const registry = new ModelRegistry();
const smartRouter = new SmartRouter(registry, /* agent map */, /* cost tracker */);
const routing = smartRouter.route('code-review-agent', { prioritize: 'cost' });After installing the tarball, the superkit CLI is available inside node_modules/.bin/:
# After npm install /path/to/super-kit-3.1.0.tgz in your project:
./node_modules/.bin/superkit init --platform claude
./node_modules/.bin/superkit agents
./node_modules/.bin/superkit skills
./node_modules/.bin/superkit --helpTo use superkit without the ./node_modules/.bin/ prefix, add it to your PATH:
export PATH="$PWD/node_modules/.bin:$PATH"
superkit init --platform claudeOr add it to your shell profile for permanent access in a project directory.
The init command writes the platform-specific rules file (e.g. CLAUDE.md), MCP server registrations, and slash command files into the current project.
# Direct invocation from the installed package:
node node_modules/super-kit/dist/core/mcp-servers/filesystem/server.js
node node_modules/super-kit/dist/core/mcp-servers/git/server.js
node node_modules/super-kit/dist/core/mcp-servers/database/server.js
node node_modules/super-kit/dist/core/mcp-servers/browser/server.js
node node_modules/super-kit/dist/core/mcp-servers/vietnam/server.js| Import path | What it exports |
|---|---|
super-kit |
Top-level re-exports (MemoryStore, TaskRouter, SmartRouter, ...) |
super-kit/core/memory |
MemoryStore, MemoryEntry, MemoryType |
super-kit/core/schemas |
SchemaValidator, AgentDefinitionSchema, SkillDefinitionSchema |
super-kit/core/orchestrator |
TaskRouter, WorkflowEngine |
super-kit/core/cost-arbitrage |
SmartRouter, ModelRegistry, CostTracker |
super-kit/core/marketplace |
MarketplaceRegistry |
super-kit/core/types |
PlatformType, AgentTeam, TaskStatus |
super-kit/core/multi-model |
Multi-model provider adapters |
super-kit/mcp/filesystem |
Filesystem MCP server |
super-kit/mcp/git |
Git MCP server |
super-kit/mcp/database |
Database MCP server |
super-kit/mcp/browser |
Browser MCP server |
super-kit/mcp/vietnam |
Vietnam MCP server |
super-kit/adapters/claude |
Claude Code adapter |
super-kit/adapters/cursor |
Cursor adapter |
| ... and 5 more adapters | See package.json exports field |
superkit init [options] # Initialize platform config in current project
superkit agents list # List all 82 agents
superkit agents show <id> # Show agent details
superkit skills list # List all skills
superkit workflow run <file> # Execute a workflow YAML
superkit marketplace search <query> # Search community plugins
superkit marketplace install <id> # Install a plugin
superkit memory search <query> # Search agent memory
superkit cost report # Show cost report (daily / monthly)
superkit doctor # Diagnose installation issuesAll commands support --help for full option list.
User input
│
▼
Platform (Claude / Cursor / Gemini / ...)
│
▼
Slash command / Agent invocation
│
▼
TaskRouter ─────► Command match? ─────► Direct agent
│
▼
Intent scoring (82 agents × keywords × weights)
│
▼
Top agent (with confidence) + alternatives
│
▼
SmartRouter ────► Model selection (cost / quality / speed / balanced)
│
▼
Provider (anthropic / openai / google / kimi / glm / local)
│
▼
Result → AgentMemory → CostTracker → Metrics
Workflow YAML
│
▼
WorkflowEngine.execute()
│
├── parallel: true → run ready steps via Promise.allSettled
│ (steps with met deps run concurrently)
│
└── parallel: false → run sequentially
(skip step if deps not met)
│
▼
Step results → context Map → next iteration
Agent save(key, value)
│
▼
MemoryStore (in-memory Map) ──► auto-flush to JSONL
│
▼
Persistent file: ~/.super-kit/memory/memory.jsonl
| Variable | Description | Default |
|---|---|---|
SUPERKIT_HOME |
Super Kit home directory | ~/.super-kit |
SUPERKIT_LOG_LEVEL |
Log level (debug, info, warn, error) |
info |
ANTHROPIC_API_KEY |
Anthropic API key for Claude models | — |
OPENAI_API_KEY |
OpenAI API key for GPT models | — |
GOOGLE_API_KEY |
Google API key for Gemini models | — |
KIMI_API_KEY |
Moonshot API key for Kimi models | — |
GLM_API_KEY |
Zhipu API key for GLM models | — |
GROK_API_KEY |
xAI API key for Grok models | — |
SUPERKIT_BUDGET_DAILY |
Daily budget in USD | 5.00 |
SUPERKIT_BUDGET_MONTHLY |
Monthly budget in USD | 100.00 |
# ~/.super-kit/config.yaml
budget:
daily: 5.00
monthly: 100.00
routing:
default: balanced
fallback: cost
memory:
defaultTtlMs: 86400000 # 24 hours
persistPath: ~/.super-kit/memory
models:
preferLocal: false
preferProvider: anthropicsuper-kit/
├── core/ # Core TypeScript modules (15)
├── adapters/ # Platform adapters (10)
├── cli/ # CLI entry point
├── docs/ # Additional documentation
├── examples/ # Example workflows and agents
├── scripts/ # Build / dev scripts
├── .github/workflows/ # CI / CD
├── package.json
├── tsconfig.json
├── vitest.config.ts
└── README.md
npm install
npm run build # Compile TypeScript → dist/
npm run dev # Watch mode (tsx)npm test # Full suite
npm test -- --run # CI mode (no watch)
npm test -- core/__tests__/router # Single filenpm run lint # ESLint
npx tsc --noEmit # Type check only- Create
core/agents/<team>/<your-agent>.yaml— seecore/agents/dev/architect.yamlfor example - Run
npm run lintto validate - Add tests in
core/__tests__/agents.test.tsif needed
- Create
adapters/<platform>/index.tsextendingBaseAdapter - Implement required methods:
generateRulesContent,generateMCPConfig,postInstall - Register in
adapters/index.ts - Add tests
Contributions welcome. Please:
- Fork the repo
- Create a feature branch (
git checkout -b feature/my-change) - Add tests for new functionality
- Run
npm test && npm run lintbefore committing - Open a pull request
For major changes, open an issue first to discuss.
Status (v3.2.0): Build is green, package metadata is complete, and the npm name
super-kitis available on the public registry. This section documents the exact steps to publish.
- An npm account with publish rights for the package name (or its scoped variant).
- Two-factor authentication enabled (recommended).
NODE_AUTH_TOKENexported (or runnpm loginfirst).- Node.js ≥ 18 and
npm≥ 9.
The package.json for super-kit@3.2.0 is already configured with:
| Field | Value |
|---|---|
name |
super-kit |
version |
3.2.0 |
description |
Universal AI Agent Toolkit — 82 agents, 147 skills, ... |
keywords |
ai, agent, mcp, toolkit, claude, cursor, windsurf, cline, ... |
author |
tqk09 <qkhalk@users.noreply.github.com> |
license |
MIT |
repository |
git+https://github.com/qkhalk/super-kit.git |
homepage |
https://github.com/qkhalk/super-kit#readme |
bugs |
https://github.com/qkhalk/super-kit/issues |
engines.node |
>=18.0.0 |
bin |
superkit → ./dist/cli/index.js |
exports |
Library sub-paths + per-adapter + MCP servers |
prepublishOnly |
npm run lint && npm test -- --run && npm run build |
files |
dist, README.md, LICENSE, CHANGELOG.md, core, adapters, cli (+ test/.ts excludes) |
# 1. Authenticate (one-time per machine / CI runner)
npm login # interactive, or set NPM_TOKEN env var
npm whoami # confirm identity
# 2. Verify the registry sees the name as available
npm view super-kit name # should 404 (means available for a fresh publish)
npm view super-kit version # if this prints, someone owns it — see fallback below
# 3. Make sure working tree is clean and on the release tag
git checkout main
git pull --rebase origin main
git status # nothing uncommitted
git tag v3.2.0 --list # confirm tag exists (it does for v3.2.0)
# 4. Pre-flight (this also runs automatically via `prepublishOnly`)
npm ci
npm run lint
npm test -- --run
npm run build
# 5. Inspect what will ship
npm pack --dry-run # 943 files → super-kit-3.2.0.tgz (~853 KB)
# 6. Dry-run the publish itself (no actual upload)
npm publish --dry-run # validates auth + tarball + access
# 7. Real publish (with OTP if 2FA enabled)
npm publish --access public --otp=123456
# 8. Verify on the registry
npm view super-kit
npm view super-kit versions- Tarball:
super-kit-3.2.0.tgz - Package size: 853.1 kB (unpacked 4.0 MB)
- Total files: 943
- Includes: everything under
dist/,core/,adapters/,cli/,README.md,LICENSE,CHANGELOG.md - Excludes:
*.test.ts,*.spec.ts,**/__tests__/**,**/__test_plugins__/**,.env*, secrets, runtime caches,.super-kit/, etc. (perfiles+.npmignore)
If someone publishes super-kit between this readiness check and the real publish, pick one of these pre-validated alternatives:
| Candidate name | npm registry status (2026-06-22) | Notes |
|---|---|---|
superkit |
404 (available) | Drop the hyphen, classic fallback |
@qkhalk/super-kit |
404 (available) | Scoped under npm user qkhalk |
superkit-cli |
taken (superkit-cli@1.0.0) |
❌ do not use |
superkit-core |
check before use | Run npm view superkit-core |
universal-agent-kit |
check before use | More descriptive |
To switch, edit package.json → name (and the bin.superkit path if relevant), then re-run npm pack --dry-run and npm publish --dry-run. For a scoped package, also add a publishConfig:
{
"publishConfig": {
"access": "public"
}
}# Deprecate a version (preferred over unpublish)
npm deprecate super-kit@3.2.0 "use 3.2.1 instead"
# Unpublish (only within 72h of publish per npm policy)
npm unpublish super-kit@3.2.0name: publish
on:
push:
tags: [ "v*.*.*" ]
jobs:
publish:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with: { node-version: 20, registry-url: https://registry.npmjs.org/ }
- run: npm ci
- run: npm run lint && npm test -- --run && npm run build
- run: npm publish --access public
env:
NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}MIT — see LICENSE.
Built on:
- @modelcontextprotocol/sdk — MCP protocol
- commander — CLI framework
- inquirer — interactive prompts
- vitest — test framework
- zod — schema validation
Inspired by:
- Claude Code (slash commands, sub-agents)
- Cursor Agents
- OpenHands
- Continue
- AutoGPT
Maintainers: Tran Quoc Khanh (@tqk09) Status: Active development