메뉴
Brain Ecosystem — 3 autonomous self-learning AI brains as MCP servers for Claude Code. Brain (280 tools): error memory, code intelligence, autonomous research. Trading Brain (181 tools): adaptive trading, paper trading, signal learning, backtesting.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Entry-point live-market scan. Returns current quote, H1/H4 regime, session state, and watch levels for a default watchlist (gold, US30, US500, US100, EURUSD, GBPUSD, BTCUSD, ETHUSD) — or any custom list. Uses the free yfinance path by default, MT5 if available. USE FOR - scan markets, what's moving, market overview, current prices, what's on the watchlist, show me the markets.
Entry-point ranked-setup feed. Runs the full pair-analyze pipeline against a watchlist and returns the top N highest-grade trade setups right now, sorted by confidence × R:R, with one-line rationale and a copy-pasteable trade plan. USE FOR - what should I trade, recommend trades, best setups, top picks, ranked opportunities, where's the edge.
Entry-point skill discovery. Given a keyword or task description, returns the shortest list of Tradecraft skills that match, with direct invocation recipes. No keyword = show the 5 main entry points. USE FOR - find skill, which skill, how do I, search skills, list skills, what can this do, find command.
Entry-point strategy selector. Lists the available trading strategies with a one-line when-to-use for each, plus a deep-dive drill-down by name. Routes to the right specialist skill (ICT/SMC, breakout, scalping, swing, trend-following, mean-reversion). USE FOR - which strategy, what strategies, strategy selector, strategy list, how should I trade, pick a strategy, ICT or SMC, breakout or trend.
This skill should be used when the user asks to "create an agent", "add an agent", "write a subagent", "agent frontmatter", "when to use description", "agent examples", "agent tools", "agent colors", "autonomous agent", or needs guidance on agent structure, system prompts, triggering conditions, or agent development best practices for Claude Code plugins.
Agentic storage architecture — persistent memory for AI agents using MCP, immutable versioning, sandboxing, and intent validation. Covers the stateless problem in LLM agents, RAG limitations (read-only), MCP protocol (JSON-RPC, resources, tools), storage safety layers, and agent file system design. Source: IBM Technology (Martin Keen), March 2026. Use this skill for "agentic storage", "agent memory", "persistent agent state", "MCP storage", "agent file system", "agent hard drive", "stateless agent problem", "agent work product", "immutable versioning", "agent sandboxing", "intent validation", "agent safety layers", "agent context window limits", "storage for AI agents", "agent persistence", "autonomous agent storage", "agent audit trail". Works with mcp-integration, trading-brain, agents, trade-psychology-coach.
| name | brain-ecosystem-mcp |
| description | Brain Ecosystem — 3 autonomous self-learning AI brains as MCP servers for Claude Code. Brain (280 tools): error memory, code intelligence, autonomous research. Trading Brain (181 tools): adaptive trading, paper trading, signal learning, backtesting. |
USE FOR:
3 autonomous self-learning AI brains running as MCP servers, designed for Claude Code.
npm install -g @timmeck/brain && brain setuplocalhost:7790 (13 pages of metrics)"117+ autonomous engines run in feedback loops — observing, detecting anomalies, forming hypotheses, testing and falsifying them statistically"
| Brain | Port | Tools | Specialty |
|---|---|---|---|
| Brain | 7777-7778 | 280 | Error memory, code intelligence, autonomous research |
| Trading Brain | 7779-7780 | 181 | Adaptive trading, paper trading, signal learning, backtesting |
| Marketing Brain | 7781-7782 | 177 | Content strategy, cross-platform optimization |
| Command Center | 7790 | — | Unified dashboard, 13 monitoring pages |
# Main brain
npm install -g @timmeck/brain
brain setup # configures Claude Code MCP automatically
# Trading brain
npm install -g @timmeck/trading-brain
trading setup
# Marketing brain
npm install -g @timmeck/marketing-brain
marketing setup
For Cursor/Windsurf/Cline (HTTP/SSE):
{
"mcpServers": {
"brain": {
"url": "http://localhost:7778/sse"
}
}
}
Goal → decompose → sub-goals → hypotheses
→ test statistically (17 falsification methods)
→ confirm/reject → update knowledge graph
→ synthesize report
Trade executed → outcome recorded
→ signal that preceded it gets weighted up/down
→ strategy parameters auto-adjusted
→ anomaly detection flags regime changes
→ hypothesis: "this pattern no longer works" → test → confirm → disable
Input (error/trade/content event)
↓
117 autonomous engines in parallel feedback loops:
- AnomalyDetector
- HypothesisGenerator
- StatisticalFalsifier
- PatternRecognizer
- KnowledgeGraphUpdater
- DreamConsolidator (offline)
↓
Hebbian synapse: brains share relevant discoveries
↓
Knowledge persists in SQLite → available next session
When idle (no active requests):
Brain enters "dream mode":
1. Replays recent experiences
2. Identifies patterns not obvious during active processing
3. Prunes weak connections (low-weight knowledge)
4. Strengthens high-value patterns
5. Prepares summaries for fast retrieval next session
After brain setup, Claude Code gets access to all 280 Brain tools:
# In Claude Code session — brain remembers across sessions:
> "You made an error with X last week"
→ Brain recalls error memory: exact context + solution applied
> "Research async patterns in Python"
→ Brain creates 5-step research roadmap, executes autonomously,
synthesizes findings into knowledge graph entry
> "What patterns have worked for BTCUSDT this month?"
→ Trading Brain queries signal performance history → ranked list
| Feature | Brain Ecosystem | Standard MCP Memory |
|---|---|---|
| Error learning | ✓ (auto, cross-session) | Manual |
| Hypothesis testing | ✓ (statistical) | ✗ |
| Dream consolidation | ✓ | ✗ |
| Self-modification | ✓ (human-gated) | ✗ |
| Trading brain | ✓ (181 tools) | ✗ |
| Inter-brain comms | ✓ (Hebbian) | ✗ |
| Vision | ✓ (Anthropic+Ollama) | ✗ |
name: mcp-server-chart description: > AntV MCP Server Chart - MCP server generating 26+ chart types via AntV. Tools: generate_bar_chart, generate_line_chart, generate_pie_chart, generate_network_graph, generate_sankey, generate_treemap, generate_spreadsheet, etc. npx @antv/mcp-server-chart. Works with Claude, VSCode, Dify. USE FOR:
MCP server generating 26+ visualization types using AntV.
npm install -g @antv/mcp-server-chart{
"mcpServers": {
"mcp-server-chart": {
"command": "npx",
"args": ["-y", "@antv/mcp-server-chart"]
}
}
}
Standard: area, bar, column, line, pie, scatter, dual_axes
Statistical: boxplot, histogram, violin
Flow: funnel, sankey, treemap
Hierarchy: mind_map, fishbone, org_chart
Network: network_graph, venn
Geographic: district_map, path_map, pin_map
Other: radar, word_cloud, liquid, spreadsheet
User: "Plot this data as a bar chart: [data]"
Claude: calls generate_bar_chart({ data: [...], xField: "x", yField: "y" })
→ returns chart image/URL
name: dify-llm-platform description: > Dify - open-source LLM app development platform. Visual canvas for AI workflows, RAG pipelines (PDF/PPT ingestion), agent builder (50+ tools: Google, DALL-E, Wolfram), LLMOps observability, BaaS APIs. Supports GPT, Claude, Llama3, Mistral, 100+ models. Self-host (Docker) or cloud (200 free GPT-4 calls). USE FOR:
Open-source LLM application development platform.
git clone https://github.com/langgenius/dify
cd dify/docker
cp .env.example .env
docker compose up -d
# Access: http://localhost/install
Google Search, Bing, DuckDuckGo, Wikipedia, DALL-E, Stable Diffusion, WolframAlpha, Weather API, News API, Code execution, Web scraping, + custom tools
name: open-webui description: > Open WebUI - self-hosted, offline-capable AI platform. Ollama + OpenAI-compatible backends. RAG with 9 vector DBs, web search (15+ providers), image gen (DALL-E/ComfyUI), voice/video chat, Python function calling, enterprise auth (LDAP/OAuth/SCIM). Docker install. Privacy-first local AI deployment. USE FOR:
Extensible self-hosted AI platform — runs fully offline.
# With Ollama bundled
docker run -d -p 3000:8080 --gpus=all -v ollama:/root/.ollama -v open-webui:/app/backend/data --name open-webui ghcr.io/open-webui/open-webui:ollama
# Existing Ollama
docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui ghcr.io/open-webui/open-webui:main
# Access: http://localhost:3000
| Feature | Open WebUI | ChatGPT |
|---|---|---|
| Self-hosted | Yes | No |
| Offline | Yes | No |
| Local models | Ollama | No |
| RAG | 9 vector DBs | Limited |
| Web search | 15+ providers | Yes |
| Image gen | DALL-E/ComfyUI/A1111 | DALL-E only |
| Python tools | Native | Sandboxed |
| Cost | Free | $20/mo |
500+ MCP servers across 40+ categories. Key ones by domain:
| Server | What It Does |
|---|---|
| GitHub MCP | Repos, PRs, issues, code search |
| GitLab MCP | GitLab API integration |
| Filesystem MCP | Local file read/write/search |
| Git MCP | git log, diff, branch operations |
| Docker MCP | Container management |
| Kubernetes MCP | Cluster insights, kubectl |
| Server | What It Does |
|---|---|
| Memory MCP | Persistent knowledge graph |
| Sequential Thinking | Chain-of-thought reasoning |
| Fetch/Browser | Web content retrieval |
| Playwright MCP | Browser automation |
| AgentShield | Security vulnerability scanning |
| Server | What It Does |
|---|---|
| PostgreSQL MCP | Schema inspection + queries |
| MongoDB MCP | Document DB queries |
| Elasticsearch MCP | Search and analytics |
| Snowflake MCP | Data warehouse queries |
| SQLite MCP | Local database |
| Server | What It Does |
|---|---|
| Gmail MCP | Email read/send/search |
| Slack MCP | Channel messages, search |
| Telegram MCP | Bot messages |
| Discord MCP | Server interaction |
| Server | What It Does |
|---|---|
| Notion MCP | Pages, databases, blocks |
| Jira MCP | Issues, sprints, projects |
| Google Calendar MCP | Events, scheduling |
| Airtable MCP | Base/table operations |
| Server | What It Does |
|---|---|
| Crypto APIs | Price feeds, portfolio |
| Payment MCP | Stripe, payment processing |
| Multi-cloud cost | Cloud cost analysis |
| Server | What It Does |
|---|---|
| Brave Search | Web search |
| Firecrawl | Deep web scraping |
| EXA MCP | AI-powered search |
| 75+ data extractors | Specialized data sources |
Full list: https://github.com/punkpeye/awesome-mcp-servers
Complete system for maximizing Claude Code performance (10+ months production-tested).
| Component | Count | Purpose |
|---|---|---|
| Agents | 21-25 | Specialized subagents (planner, architect, reviewer, security, language-specific) |
| Skills | 102+ | Domain workflows (TDD, security review, frontend/backend patterns) |
| Commands | 52-57 | Slash commands: /tdd, /plan, /code-review, /build-fix, /e2e |
| Rules | 29-34 | Universal + language-specific (TS, Python, Go, Swift, PHP, Java) |
| Hooks | 8-20+ | PreToolUse, PostToolUse, Stop, SessionStart automations |
| MCP | 14+ | Pre-configured: GitHub, Supabase, Vercel, Railway |
/tdd - Test-driven development workflow
/plan - Architecture planning
/code-review - Security + quality review
/build-fix - Build error diagnosis
/e2e - End-to-end test generation
/multi-plan - Multi-agent orchestration
/instinct-status - Check learned patterns
/evolve - Extract patterns from session into skills
{
"hooks": {
"PreToolUse": ["secret-detector", "format-checker"],
"PostToolUse": ["session-persist", "pattern-extractor"],
"Stop": ["summary-generator"],
"SessionStart": ["context-loader", "instinct-injector"]
}
}
name: cherry-studio description: > Cherry Studio - desktop AI client for Windows/Mac/Linux. Unifies 20+ AI providers: OpenAI, Anthropic Claude, Gemini, Ollama (local), LM Studio, Qwen, Kimi. 300+ pre-configured assistants, multi-model simultaneous chat, MCP server support, document processing (PDF/Office/images), no environment setup needed. USE FOR:
Unified desktop AI client — no setup required.
OpenAI (GPT-4o) · Anthropic (Claude) · Google (Gemini) · Mistral Ollama (local) · LM Studio (local) · Qwen · Kimi · Baidu · iFlytek OpenRouter · Perplexity · Poe · + more
name: aionui-desktop-agent description: > AionUi - free open-source multi-agent desktop platform (macOS/Windows/Linux). Auto-detects Claude Code, Codex, Qwen Code CLI tools. 20+ AI providers. Cron scheduling for 24/7 unattended automation. File management, Excel AI, PowerPoint generation, image recognition. Remote access via Telegram/Lark/DingTalk. USE FOR:
Multi-agent desktop platform — AI agents operate autonomously on your computer.
Gemini · Anthropic Claude · OpenAI · Qwen · Kimi · Baidu · Ollama · LM Studio
| Feature | Description |
|---|---|
| Cron scheduling | Natural language task descriptions, runs 24/7 |
| File management | Batch rename, intelligent classification |
| Excel processing | AI analysis, formatting, report generation |
| Document generation | PowerPoint, Word, Markdown automated creation |
| Image operations | Text-to-image, editing, recognition |
Real-world agent skills from 40+ organizations. Works with Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot.
Source: https://github.com/VoltAgent/awesome-agent-skills
Official / Anthropic
anthropics/docx — Create, edit, analyze Word documentsInfrastructure & DevOps
Databases & Data
Dev Frameworks
AI/ML
Security (Trail of Bits — 23 skills)
Product & SaaS
Web3/Crypto
Security warning: Skills are curated, not audited. Review before install. Scanners: Snyk Skill Security Scanner, Agent Trust Hub.
Production Claude Code plugin system: 112 agents, 146 skills, 79 tools, 72 plugins.
/plugin marketplace add wshobson/agents
/plugin install python-development
/plugin install security-audit
name: ruflo-agent-orchestration description: > Ruflo (formerly Claude Flow) - enterprise multi-agent orchestration for Claude Code. 60+ specialized agents in swarms (hierarchical/mesh/ring/star topologies). Self-learning SONA architecture, HNSW vector search (150x-12500x faster retrieval), EWC++ anti-forgetting, Mixture of Experts routing. MCP server. npx ruflo@latest init USE FOR:
Enterprise AI agent orchestration platform built on Claude Code.
npx ruflo@latest init --wizardclaude mcp add ruflo -- npx ruflo@latest mcp start| Feature | Ruflo | CrewAI | LangGraph |
|---|---|---|---|
| Self-learning | SONA (0.05ms adapt) | No | No |
| Anti-forgetting | EWC++ | No | No |
| Vector retrieval | HNSW (12500x faster) | Basic | Basic |
| Agent topologies | 4 (hier/mesh/ring/star) | Hierarchical | Graph |
| Swarm size | Unlimited | Limited | Limited |
curl -fsSL https://cdn.jsdelivr.net/gh/ruvnet/claude-flow@main/scripts/install.sh | bash
# or
npx ruflo@latest init --wizard
npx ruflo@latest --agent coder --task "Implement user authentication"
npx ruflo@latest --list # list all 60+ agents
npx ruflo@latest mcp start # start MCP server
Hierarchical: Queen agent → Worker agents (tree structure)
Mesh: All agents peer-to-peer (collaborative)
Ring: Sequential pipeline (A→B→C→A)
Star: Central coordinator → spoke agents
Most comprehensive open-source Claude Code skills library (204 skills, 266 Python CLI tools). Compatible with 11 AI coding platforms.
| Domain | Skills | Notable |
|---|---|---|
| Engineering Core | 25 | Architecture, DevOps, security, AI/ML |
| Engineering POWERFUL | 30 | Advanced tier skills |
| Playwright Testing | 9+ | Test generation, migration |
| Product Management | 13 | Strategy, UX research, analytics |
| Marketing | 43 | 7 specialized pods |
| Project Management | 6 | PM, scrum, Jira/Confluence |
| Regulatory/Quality | 12 | FDA, ISO, MDR, GDPR |
| C-Level Advisory | 28 | Full executive suite |
| Business & Growth | 4 | Customer success, sales |
| Finance | 2 | Financial analysis, SaaS metrics |
skill-name/
├── SKILL.md # structured instructions + workflow
├── tools/ # Python CLI tools (standard library only)
│ └── tool_name.py
├── scripts/ # optional automation scripts
└── references/ # templates + domain knowledge
# Claude Code plugin marketplace
/plugin install alirezarezvani/claude-skills/engineering
# Manual: copy skill folder to ~/.claude/skills/
cp -r marketing-pod ~/.claude/skills/
name: aris-auto-research description: > ARIS - Autonomous ML research methodology using Claude Code + GPT adversarial reviewer. 4 workflows: idea discovery, experiment bridge, auto-review loop, paper writing. 20 composable skills. State file recovery (no daemon needed). GPU automation via SSH. Proven: borderline reject (5/10) -> submission-ready (7.5/10) in 4 overnight rounds. USE FOR:
Autonomous ML research methodology — Claude Code executes, GPT-5.4 reviews adversarially.
Workflow 1: Idea Discovery
literature survey → 8-12 ideas → novelty check → GPU pilots → ranked report
Workflow 1.5: Experiment Bridge
reads plan → implements code → validates small-scale → deploys full GPU suite
Workflow 2: Auto Review Loop (key workflow)
GPT-5.4 reviews paper → identifies gaps → Claude Code writes experiments
→ SSH deploys to GPU → monitors → rewrites sections → repeat (max 4 rounds)
Workflow 3: Paper Writing
narrative → claims-evidence matrix → figures/tables → LaTeX → PDF → 2-round auto-improve
/research-pipeline "your direction" # full pipeline
/idea-discovery "topic" # workflow 1
/experiment-bridge # workflow 1.5
/auto-review-loop "paper scope" # workflow 2 (overnight key)
/paper-writing "NARRATIVE_REPORT.md" # workflow 3
# With overrides:
/research-pipeline "topic" -- AUTO_PROCEED: false, wandb: true
Each skill saves progress to JSON after every round:
REVIEW_STATE.json
AUTO_REVIEW.md
If session terminates mid-loop:
Restart Claude Code → skill reads state → resumes from last checkpoint
Configure in project CLAUDE.md:
SSH_HOST: your-gpu-server.com
SSH_USER: ubuntu
CONDA_ENV: research
SLURM: true # if using SLURM cluster
Skills handle: ssh → rsync code → conda activate → run experiments → fetch results
name: ai-maestro-orchestration description: > AI Maestro - centralized dashboard for managing 35+ AI agents across machines. Peer mesh network (no central server), agent-to-agent messaging (AMP protocol), Code Graph visualization, Kanban task tracking, persistent memory, tmux auto-discovery. Gateway: Slack/Discord/Email/WhatsApp. curl install, Node.js 18+, tmux required. USE FOR:
Centralized dashboard for managing distributed AI agent workforces.
curl -fsSL https://raw.githubusercontent.com/23blocks-OS/ai-maestro/main/scripts/remote-install.sh | sh
# Requirements: Node.js 18+, tmux
| Feature | Description |
|---|---|
| Auto-discovery | Finds existing tmux sessions automatically |
| Peer mesh | No central server — equal-status machines |
| AMP protocol | Agent-to-Agent Messaging with priority + crypto signatures |
| Code Graph | Interactive codebase visualization, delta indexing |
| Kanban | Task dependencies + status tracking |
| Persistent memory | Cross-session contextual continuity |
| War rooms | Multi-agent team assembly |
| Gateways | Slack, Discord, Email, WhatsApp routing |
Message types: command, query, response, broadcast
Priority: critical > high > normal > low
Signature: cryptographic for auth
Push notifications: real-time delivery
name: babysitter-agent-control description: > Babysitter - deterministic agent workflow framework. Process-as-code (JS functions), mandatory stops between steps, enforced quality gates, event-sourced audit journal. 50-67% token compression. 2000+ pre-built process templates. 4 modes: interactive, autonomous (yolo), planning, continuous (forever monitoring). USE FOR:
Deterministic agent workflow framework — agents can only execute what the process permits.
Process as Code → JS functions define exactly what agents may do
Mandatory Stops → every step halts; process logic decides next action
Enforced Gates → quality thresholds block progression
Event Journal → immutable audit trail, replay from any checkpoint
/babysitter:call # Interactive — pauses for human approval
/babysitter:yolo # Autonomous — fully automatic
/babysitter:plan # Planning — review before executing
/babysitter:forever # Continuous — indefinite monitoring
// Automated refinement until threshold met
process.qualityGate({
check: () => runTests(),
threshold: 0.95, // 95% pass rate required
maxRounds: 5, // max refinement attempts
onFail: "refine-code" // action if threshold not met
})
50-67% context reduction built-in — summarizes completed steps, retains only active context.
/babysitter:observe # live dashboard
/babysitter:resume # continue interrupted workflow from checkpoint
2,000+ templates covering:
name: skills-manager-desktop description: > Skills Manager - desktop app (Tauri 2.0 + React 19) for managing AI skills across multiple coding assistants. Write a skill once, sync via symlinks to Claude Code, Codex, Opencode simultaneously. Granular enable/disable per tool. Cross-platform (macOS/Windows/Linux). Monaco Editor for in-app skill editing. USE FOR:
Centralized desktop app to manage AI assistant skills across Claude Code, Codex, Opencode, etc.
~/.claude/skills/~/.codex/skills/~/.opencode/skills/Requires Administrator privileges or Developer Mode enabled for symlink creation.
name: claude-code-karma description: > Claude Code Karma - local-first analytics dashboard for Claude Code sessions. Reads ~/.claude/ JSONL files, serves via FastAPI (port 8000), displays in SvelteKit UI (port 5173). No cloud, no accounts, no telemetry. Shows token usage, costs, cache hit rates, tool/agent distribution, file operations, plugin/skill/hook inventory, live session monitoring. USE FOR:
Local analytics dashboard for your Claude Code sessions.
git clone https://github.com/JayantDevkar/claude-code-karma.git
cd claude-code-karma
# Terminal 1 — backend
cd api && pip install -e ".[dev]" && pip install -r requirements.txt
uvicorn main:app --reload --port 8000
# Terminal 2 — frontend
cd frontend && npm install && npm run dev
# Access: http://localhost:5173
| View | What It Shows |
|---|---|
| Session Browser | All sessions with search/filter, real-time status |
| Analytics | Token usage, costs, cache hit rates, tool distribution |
| Project Organization | Workspaces by git repo + activity tracking |
| Tool & Agent Tracking | Built-in + MCP tools + custom agents + usage stats |
| File & Task Management | File operations, task creation/completion |
| Live Monitoring | Real-time via Claude Code hooks (optional) |
| Plugin Ecosystem | Installed plugins, skills, commands, hooks |
name: swing-trade-psychology-coach-firewall description: > Swing - 6-skill AI trade-psychology-coach firewall suite. Prevents systematic AI reasoning failures: ambiguous execution, unverified claims, anchoring, sycophancy, hidden reasoning, optimism bias. Skills: swing-clarify (5W1H), swing-research (4-stage verification), swing-options (5 alternatives), swing-review (steel-man + 3-vector critique), swing-trace (assumption mapping), swing-mortem (5-category failure projection). npx skills add whynowlab/swing-skills --all. USE FOR:
6-skill trade-psychology-coach firewall that addresses systematic AI reasoning failures.
npx skills add whynowlab/swing-skills --all # all 6 skills
# Individual:
npx skills add whynowlab/swing-skills/swing-clarify
npx skills add whynowlab/swing-skills/swing-research
# Manual:
cp -r swing-skills/skills/* ~/.claude/skills/
| Skill | Problem Solved | Method |
|---|---|---|
| swing-clarify | Ambiguous requests rushed | 5W1H decomposition |
| swing-research | Unverified claims | 4-stage verification, S/A/B/C source grading |
| swing-options | Anchoring on obvious answer | 5 probability-weighted alternatives |
| swing-review | Sycophancy / no critique | Steel-man then 3-vector critical analysis |
| swing-trace | Hidden reasoning | Assumption map, decision forks, weakest-link |
| swing-mortem | Optimism bias | 5-category failure projection + leading indicators |
/swing-clarify Build me an auth system
/swing-research Is gRPC better than REST for mobile?
/swing-review We chose Kubernetes for our 3-person startup
/swing-options Which database for real-time leaderboard?
/swing-trace Why do you recommend microservices?
/swing-mortem We are migrating to microservices in Q3
clarify -> (research decision) -> options -> research -> review
-> (risk analysis) -> mortem
name: spec-flow description: > Spec-Flow - structured feature development workflow for AI coding agents. 5 sequential phases with approval gates: Proposal -> Requirements (EARS syntax) -> Design -> Tasks -> Implementation. Generates living markdown docs in .spec-flow/ directory. 3 modes: Step, Batch, Phase. Triggers: "spec-flow", "spec mode", "need a plan", "structured development". Install: git clone into ~/.claude/skills. USE FOR:
Phase-gated structured development workflow for AI coding agents.
cd ~/.claude/skills && git clone https://github.com/echoVic/spec-flow.git1. Proposal → overview, scope, success criteria
2. Requirements → EARS syntax specs (SHALL/WHEN/WHERE/IF clauses)
3. Design → architecture, components, data models, interfaces
4. Tasks → numbered implementation checklist, dependencies
5. Implementation → execute tasks one by one, verify each
# Default (step-by-step, confirmation at each phase)
spec-flow: Build user authentication system
# Fast (skip confirmations)
spec-flow --fast: Add dark mode toggle
# Simple (skip design phase)
spec-flow --skip-design: Fix the login bug
.spec-flow/
steering/ # project context, conventions
active/ # in-progress feature docs
feature-name/
proposal.md
requirements.md
design.md
tasks.md
archive/ # completed features
WHEN user submits login form
IF credentials are valid
THE SYSTEM SHALL authenticate the user and redirect to dashboard
WHERE the user is not authenticated
THE SYSTEM SHALL redirect to login page
name: ham-hierarchical-memory description: > HAM (Hierarchical Agent Memory) - reduces Claude Code token consumption 50% by replacing one massive CLAUDE.md with small scoped CLAUDE.md files at each directory level. Agent loads only 2-3 relevant files per session. Auto-detects project stack, self-maintaining (updates decision/pattern files as work progresses). Analytics dashboard at localhost:7777. Install: git clone into ~/.claude/skills/ham. USE FOR:
Hierarchical Agent Memory — cuts Claude Code starting context from thousands of tokens to hundreds.
git clone https://github.com/kromahlusenii-ops/ham.git ~/.claude/skills/hamBefore HAM: One massive CLAUDE.md → loads entire project context every session
After HAM: Small CLAUDE.md at each directory level → loads only 2-3 relevant files
# Setup
go ham # auto-configures everything
ham update # refresh memory files
ham status # show current memory state
ham route # show which files will load for current directory
# Analytics (web UI at localhost:7777)
ham dashboard # open analytics dashboard
ham savings # show token savings report
ham carbon # environmental impact calculator
# Benchmarking
ham benchmark # run benchmark vs baseline
ham baseline start # start baseline measurement
ham baseline stop # stop baseline, compute metrics
# Maintenance
ham audit # check for stale/missing memory files
ham commands # list all available commands
project/
CLAUDE.md # global conventions, project overview
src/
CLAUDE.md # frontend/backend patterns
components/
CLAUDE.md # component-specific conventions
api/
CLAUDE.md # API patterns, auth flows
tests/
CLAUDE.md # testing patterns, fixtures
Read: agent reads directory CLAUDE.md before starting work
Write: agent updates CLAUDE.md after completing work
(new patterns, decisions, gotchas discovered)
name: geo-lint description: > geo-lint - first open-source linter for GEO (Generative Engine Optimization). 97 rules across 5 categories: GEO (AI citation readiness, E-E-A-T, RAG optimization), SEO (metadata/schema/keywords), content quality, technical (broken links/perf), i18n. Agent-first JSON output for auto-fix loops. Ensures content gets cited by ChatGPT, Perplexity, Google AI Overviews, Gemini. npm install -D @ijonis/geo-lint. Claude Code skill: /geo-lint audit. USE FOR:
First open-source linter for Generative Engine Optimization — ensures content gets cited by AI search engines.
npm install -D @ijonis/geo-lintnpm install -D @ijonis/geo-lint
# Claude Code users:
curl -fsSL https://raw.githubusercontent.com/IJONIS/geo-lint/main/install.sh | bash
npx geo-lint # human-readable output
npx geo-lint --format=json # machine-readable for AI agents
/geo-lint audit # Claude Code skill: full site scan + fix
/content-creator setup # create SEO/GEO-optimized content
| Category | Rules | Focus |
|---|---|---|
| GEO | 36 | AI citation readiness, E-E-A-T signals, RAG optimization |
| SEO | 34 | Metadata, schema markup, keywords |
| Content Quality | 14 | Readability, word count, jargon |
| Technical | 10 | Broken links, performance |
| i18n | 3 | Translation pairs |
geo-lint --format=json → violations with fix suggestions
→ AI agent modifies content
→ geo-lint re-runs
→ repeat until violations == 0
Each violation includes: file location, rule name, severity, plain-language fix instruction, machine-readable fix pattern.
Curated index of 400+ Claude Code extensions, skills, tools, hooks, and workflows.
| Command | Purpose |
|---|---|
| /commit | Conventional commit with emojis |
| /create-pr | PR creation workflow |
| /tdd | TDD enforcement |
| /check | Code quality + security checks |
| /context-prime | Comprehensive project priming |
| /prd-generator | Product Requirements Document generation |
| /optimize | Performance bottleneck identification |
| /mermaid | ER diagram generation from SQL |
Full index: https://github.com/hesreallyhim/awesome-claude-code
name: refly-agent-skill-builder description: > Refly - open-source platform transforming enterprise workflows into versioned agent skills. Copilot-led DSL compiler: describe workflow in natural language, compiles in <3 minutes. Intervenable runtime (pause/audit/redirect mid-execution). Exports to Claude Code, Cursor, MCP. 3000+ native integrations: Stripe, Slack, Salesforce, GitHub, MCP servers. Central skill registry, version-controlled, team-shareable. USE FOR:
Open-source platform for building versioned enterprise agent skills.
npm install -g @powerformer/refly-cliInput Layer: 3000+ tools, MCP servers, private connectors
Processing Layer: Vibe-driven DSL compiler + stateful intervenable runtime
Output Layer: Claude Code, Cursor, APIs, webhooks, agent frameworks
npm install -g @powerformer/refly-cli
# Install a skill from registry
refly skill install <skill-id>
# Publish your skill
refly skill publish <skill-id>
# Execute workflow via API
curl -X POST https://instance/api/v1/workflows/{WORKFLOW_ID}/execute \
-H "Authorization: Bearer API_KEY"
143,000+ GitHub stars. Works with ChatGPT, Claude, Gemini, Llama, Mistral. Originally launched as Awesome ChatGPT Prompts (Dec 2022).
# CLI
npx prompts.chat
# Claude Code plugin
/plugin install prompts.chat
# MCP server
# add to MCP config
# Browse online: https://prompts.chat/prompts
# CSV/Markdown/Hugging Face dataset available
Free interactive guide with 25+ chapters on techniques:
Full prompt browser: https://prompts.chat/prompts
name: lobehub-platform description: > LobeHub - open-source AI platform for human-agent co-evolution. Web app + desktop + PWA. Multi-model (OpenAI, Ollama, 100+ providers), 10,000+ MCP plugins marketplace, agent builder, agent groups, personal memory, real-time internet search, chain-of-thought visualization, voice TTS/STT, image generation, knowledge base. Self-host via Docker or Vercel one-click. USE FOR:
Open-source AI platform — web app, desktop, PWA with 10,000+ MCP plugins.
# Docker
docker run -d -p 3210:3210 lobehub/lobe-chat
# Vercel one-click (or Zeabur, Sealos, Alibaba Cloud)
# Set OPENAI_API_KEY or other provider credentials
| Feature | Details |
|---|---|
| Multi-model | OpenAI, Ollama (local), Anthropic, Gemini, 100+ providers |
| MCP Marketplace | 10,000+ compatible plugins |
| Agent Builder | Custom agents with tools, memory, persona |
| Agent Groups | Team-based agent collaboration |
| Personal Memory | Adaptive learning across sessions |
| Internet Search | Real-time web integration |
| Chain of Thought | Visualize AI reasoning steps |
| Voice | TTS/STT conversational interface |
| Image Gen | Generation + visual recognition |
| Knowledge Base | File uploads, RAG |
name: claude-mem description: > claude-mem - persistent memory compression system for Claude Code. Auto-captures tool usage, generates semantic summaries, persists across sessions. Progressive disclosure (layered retrieval with token costs). mem-search skill for natural language history queries. 4 MCP tools: search, timeline, get_observations. Web viewer at localhost:37777. Private tag for sensitive content. Install via /plugin marketplace add thedotmack/claude-mem. USE FOR:
Persistent memory compression system — context survives Claude Code session restarts.
# Via Claude Code Plugin Marketplace (recommended)
/plugin marketplace add thedotmack/claude-mem
/plugin install claude-mem
# Restart Claude Code
# Via OpenClaw Gateway
curl -fsSL https://install.cmem.ai/openclaw.sh | bash
<private> tags to exclude from memory| Tool | Purpose |
|---|---|
| search | Query memory index by natural language |
| timeline | Chronological context view |
| get_observations | Fetch full details by observation ID |
Real-time memory stream visualization + settings management.
name: continue-ai-code-review description: > Continue - open-source AI-powered code review platform running agents on every PR as GitHub status checks. Review rules stored as markdown in .continue/checks/ directories (source-controlled). Returns green pass or red with suggested diffs. VS Code extension available. CLI via npm @continuedev/cli (cn command). macOS/Linux/Windows install. Custom checks for security, best practices, team conventions. USE FOR:
Open-source AI code review platform — agents run on every PR as GitHub status checks.
cn# macOS/Linux
curl -fsSL https://install.continue.dev | bash
# Windows (PowerShell)
iwr https://install.continue.dev/windows -useb | iex
# Node.js alternative (Node 20+)
npm install -g @continuedev/cli
1. Store review rules as markdown in .continue/checks/
2. Continue runs agents on every PR
3. Returns GitHub status check: green (pass) or red (with suggested diffs)
4. Teams enforce rules as required status checks
.continue/
checks/
security.md # security vulnerability rules
best-practices.md # team coding standards
custom-check.md # any custom requirement
Available through VS Code marketplace — provides IDE-level AI assistance integrated with the same agent framework.
name: memu-proactive-memory description: > memU - 24/7 always-on proactive memory framework for AI agents. Continuously captures user intent without explicit commands. Hierarchical 3-layer memory (Resource/Item/Category). Dual pathways: memorize() continuous learning + retrieve() RAG fast or LLM deep reasoning. 92.09% accuracy on benchmarks. Reduces LLM token costs via cached insights. Storage: in-memory or PostgreSQL (pgvector). pip install -e . Python 3.13+. USE FOR:
24/7 proactive memory framework — AI agents learn continuously without explicit prompts.
pip install -e .
export OPENAI_API_KEY=your_api_key
3-Layer Hierarchical Memory:
Resource → mountable data sources (docs, sessions, external feeds)
Item → individual memory facts with cross-references
Category → auto-organized topic clusters
Dual Retrieval Pathways:
memorize() → continuous learning pipeline (immediate memory update)
retrieve() → RAG fast (sub-second embedding) OR LLM deep (complex anticipation)
# In-memory (default)
python tests/test_inmemory.py
# PostgreSQL with pgvector (persistent)
docker run -d --name memu-postgres \
-e POSTGRES_PASSWORD=postgres \
-p 5432:5432 pgvector/pgvector:pg16
python tests/test_postgres.py
python examples/example_1_conversation_memory.py # session persistence
python examples/example_2_skill_extraction.py # learn user skills
python examples/example_3_multimodal_memory.py # images + text
name: langchain-framework description: > LangChain - leading framework for building LLM-powered applications and agents. Core: chains (sequential LLM pipelines), agents (tool-using reasoning loops), memory (conversation + long-term), RAG (retrieval-augmented generation), 100+ integrations (OpenAI, Anthropic, HuggingFace, Pinecone, Chroma, etc.). Model-agnostic: swap providers without code changes. Ecosystem: LangGraph (orchestration), LangSmith (debug/deploy). pip install langchain. USE FOR:
Framework for building LLM-powered applications — agents, chains, RAG, memory, tools.
pip install langchain
pip install langchain-openai # OpenAI provider
pip install langchain-anthropic # Anthropic/Claude
pip install langchain-community # community integrations
# or
uv add langchain
from langchain.chat_models import init_chat_model
model = init_chat_model("openai:gpt-4o")
result = model.invoke("Explain quantum computing in one sentence")
# Or Claude:
model = init_chat_model("anthropic:claude-sonnet-4-6")
Chains — Sequential LLM pipelines:
from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI
prompt = ChatPromptTemplate.from_template("Translate to French: {text}")
chain = prompt | ChatOpenAI(model="gpt-4o")
result = chain.invoke({"text": "Hello world"})
Agents — Tool-using reasoning loops:
from langchain.agents import create_react_agent, AgentExecutor
from langchain_community.tools import DuckDuckGoSearchRun
tools = [DuckDuckGoSearchRun()]
agent = create_react_agent(llm, tools, prompt)
executor = AgentExecutor(agent=agent, tools=tools)
executor.invoke({"input": "What happened in AI this week?"})
Memory — Conversation persistence:
from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationChain
memory = ConversationBufferMemory()
chain = ConversationChain(llm=llm, memory=memory)
RAG Pipeline:
from langchain_community.vectorstores import Chroma
from langchain_openai import OpenAIEmbeddings
from langchain.chains import RetrievalQA
vectorstore = Chroma.from_documents(docs, OpenAIEmbeddings())
retriever = vectorstore.as_retriever(search_kwargs={"k": 4})
qa_chain = RetrievalQA.from_chain_type(llm=llm, retriever=retriever)
qa_chain.invoke({"query": "What does the document say about X?"})
| Tool | Purpose |
|---|---|
| LangGraph | Stateful multi-agent orchestration (cycles + branching) |
| LangSmith | LLM app monitoring, debugging, evaluation, deployment |
| LangServe | Deploy chains as REST APIs |
Models: OpenAI, Anthropic, HuggingFace, Cohere, Google, Ollama, Mistral Vector Stores: Pinecone, Chroma, FAISS, Weaviate, Qdrant, pgvector Tools: DuckDuckGo, SerpAPI, Wikipedia, Python REPL, SQL, Playwright Document Loaders: PDF, Word, HTML, CSV, YouTube, Notion, GitHub
name: vercel-ai-sdk description: > Vercel AI SDK - TypeScript toolkit for building AI apps with React/Next.js/Vue/Svelte/Node.js. Unified API across 20+ providers (OpenAI, Anthropic, Google, Groq, Mistral, DeepSeek, etc.). Core: generateText, streamText, generateObject, streamObject, tool calling, embeddings. UI hooks: useChat, useCompletion. npm install ai. Two libraries: AI SDK Core + AI SDK UI. USE FOR:
TypeScript toolkit for building AI apps — unified API across 20+ LLM providers.
npm install ai
npm install @ai-sdk/openai # OpenAI provider
npm install @ai-sdk/anthropic # Anthropic/Claude
npm install @ai-sdk/google # Google Gemini
Text Generation:
import { generateText, streamText } from "ai";
import { anthropic } from "@ai-sdk/anthropic";
// Single response
const { text } = await generateText({
model: anthropic("claude-sonnet-4-6"),
prompt: "Explain quantum computing",
});
// Streaming
const result = streamText({
model: anthropic("claude-sonnet-4-6"),
prompt: "Write a short story",
});
for await (const chunk of result.textStream) {
process.stdout.write(chunk);
}
Structured Output:
import { generateObject } from "ai";
import { z } from "zod";
const { object } = await generateObject({
model: openai("gpt-4o"),
schema: z.object({
name: z.string(),
age: z.number(),
skills: z.array(z.string()),
}),
prompt: "Generate a fictional developer profile",
});
Tool Calling:
import { tool } from "ai";
const result = await generateText({
model: openai("gpt-4o"),
tools: {
getWeather: tool({
description: "Get weather for a city",
parameters: z.object({ city: z.string() }),
execute: async ({ city }) => fetchWeather(city),
}),
},
prompt: "What is the weather in Paris?",
});
useChat:
import { useChat } from "ai/react";
export default function ChatPage() {
const { messages, input, handleInputChange, handleSubmit } = useChat({
api: "/api/chat",
});
return (
<div>
{messages.map(m => <div key={m.id}>{m.role}: {m.content}</div>)}
<form onSubmit={handleSubmit}>
<input value={input} onChange={handleInputChange} />
<button type="submit">Send</button>
</form>
</div>
);
}
OpenAI · Anthropic · Google · Azure · Amazon Bedrock · Groq · Mistral Cohere · DeepSeek · xAI Grok · Together.ai · Fireworks · Perplexity · + more
| Function | Purpose |
|---|---|
| generateText | Single text response |
| streamText | Streaming text + tool calls |
| generateObject | Typed JSON (Zod schema) |
| streamObject | Streaming structured data |
| embed | Generate embeddings |
| embedMany | Batch embeddings |
| useChat | React hook for chat UI |
| useCompletion | React hook for text completion |
| useObject | React hook for streaming objects |
name: tensorzero-llm-gateway description: > TensorZero - open-source LLM gateway + optimization platform. Sub-millisecond routing to 18+ providers (Anthropic, OpenAI, Google, AWS, etc.). Observability (PostgreSQL storage + OpenTelemetry), optimization (fine-tuning, GEPA prompt engineering, dynamic in-context learning), evaluation (heuristic + LLM judge), A/B testing + adaptive routing + fallbacks. Docker deploy. OpenAI-compatible API. Fortune 50 production use. USE FOR:
Production-grade LLM gateway with optimization, observability, evaluation, and experimentation.
from openai import OpenAI
client = OpenAI(base_url="http://localhost:3000/openai/v1")
response = client.chat.completions.create(
model="tensorzero::my_function::anthropic::claude-sonnet-4-6",
messages=[{"role": "user", "content": "Hello!"}]
)
| Feature | Details |
|---|---|
| Gateway | 18+ providers, sub-ms overhead, tool use, structured outputs, multimodal, caching |
| Observability | Stores inferences + feedback in PostgreSQL, OpenTelemetry, Prometheus |
| Optimization | Fine-tuning, GEPA automated prompt engineering, dynamic in-context learning |
| Evaluation | Heuristic benchmarks + LLM-as-judge for individual inferences |
| Experimentation | Built-in A/B testing, adaptive routing, fallbacks, retries |
Anthropic · OpenAI · Google Vertex/Gemini · AWS Bedrock/SageMaker · Azure DeepSeek · Groq · Mistral · Together AI · vLLM · OpenAI-compatible APIs
name: pentagi-security-agents description: > PentAGI - autonomous AI penetration testing platform. Multi-agent architecture: Orchestrator + Researcher + Developer + Executor + Searcher/Enricher/Memorist/Reporter. 20+ built-in security tools (nmap, metasploit, sqlmap). Long-term vector memory (PostgreSQL). Knowledge graph (Neo4j). Sandboxed Docker execution. 10+ LLM providers. REST+GraphQL API. Grafana/Prometheus monitoring. For authorized security testing and research only. USE FOR:
For authorized security testing only. Autonomous AI penetration testing platform.
Orchestrator
Researcher → target analysis, intelligence gathering
Developer → attack strategy planning
Executor → implements attack plan
Searcher → information retrieval
Enricher → context enrichment
Memorist → long-term memory management
Reporter → results documentation
Adviser → strategy recommendations
Reflector → execution review
Planner → task decomposition (3-7 steps)
nmap · metasploit · sqlmap · + 17 others
OpenAI · Anthropic · Google AI · AWS Bedrock · Ollama · + 5 others
name: arize-phoenix-observability description: > Arize Phoenix - open-source AI observability platform. OpenTelemetry-based LLM tracing, LLM-powered evaluation (response + retrieval quality), version-controlled datasets, experiment tracking (prompts/models/retrieval). Integrates LangGraph, LlamaIndex, CrewAI, DSPy, Claude Agent SDK, OpenAI, Anthropic, Bedrock. Prompt playground + management. pip install arize-phoenix. Local/notebook/Docker/cloud. USE FOR:
Open-source AI observability — trace, evaluate, and experiment on LLM applications.
pip install arize-phoenix
# Docker
docker pull arizephoenix/phoenix
docker run -p 6006:6006 arizephoenix/phoenix
# UI: http://localhost:6006
| Capability | What It Does |
|---|---|
| Tracing | OpenTelemetry-based runtime tracing of LLM calls, chains, agents |
| Evaluation | LLM-powered benchmarks: response quality, retrieval accuracy |
| Datasets | Version-controlled example collections for testing + fine-tuning |
| Experiments | Track changes to prompts, models, retrieval — compare results |
import phoenix as px
from openinference.instrumentation.langchain import LangChainInstrumentor
# Start local Phoenix
session = px.launch_app()
# Auto-instrument LangChain
LangChainInstrumentor().instrument()
# All LangChain calls now traced at http://localhost:6006
LangGraph · LlamaIndex · CrewAI · DSPy · Claude Agent SDK OpenAI · Anthropic · Google GenAI · AWS Bedrock
Real-world agent skills from 40+ organizations. Works with Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot.
Source: https://github.com/VoltAgent/awesome-agent-skills
Official / Anthropic
anthropics/docx — Create, edit, analyze Word documentsInfrastructure & DevOps
Databases & Data
Dev Frameworks
AI/ML
Security (Trail of Bits — 23 skills)
Product & SaaS
Web3/Crypto
Security warning: Skills are curated, not audited. Review before install. Scanners: Snyk Skill Security Scanner, Agent Trust Hub.
Production Claude Code plugin system: 112 agents, 146 skills, 79 tools, 72 plugins.
/plugin marketplace add wshobson/agents
/plugin install python-development
/plugin install security-audit
name: ruflo-agent-orchestration description: > Ruflo (formerly Claude Flow) - enterprise multi-agent orchestration for Claude Code. 60+ specialized agents in swarms (hierarchical/mesh/ring/star topologies). Self-learning SONA architecture, HNSW vector search (150x-12500x faster retrieval), EWC++ anti-forgetting, Mixture of Experts routing. MCP server. npx ruflo@latest init USE FOR: