Agent Skills

<div align="center">

76 AI agent skills · TOON Format · Cross-platform

Quick Start · Skills List · Installation · 한국어

</div>

---

💡 What is Agent Skills?

76 AI agent skills · TOON Format · Cross-platform

Agent Skills is a curated collection of 76 AI agent skills for LLM-based development workflows. Built around the jeo orchestration protocol, it provides:

• Unified orchestration across Claude Code, Gemini CLI, OpenAI Codex, and OpenCode
• Plan → Execute → Verify → Cleanup automated pipelines
• Multi-agent team coordination with parallel execution

---

🚀 Quick Start

Prerequisite: Install skills CLI before running npx skills add.

npm install -g skills

# Send to your LLM agent — it will read and install automatically
curl -s https://raw.githubusercontent.com/supercent-io/skills-template/main/setup-all-skills-prompt.md

Platform	First Command
Claude Code	jeo "task description" or /omc:team "task"
Gemini CLI	/jeo "task description"
Codex CLI	/jeo "task description"
OpenCode	/jeo "task description"

---

🏗 Architecture

graph TD
    JEO["🎯 JEO\nCore Orchestration"] --> PLAN["📋 PLAN\nralph + plannotator"]
    JEO --> EXEC["⚡ EXECUTE\nteam / bmad"]
    JEO --> VERIFY["🔍 VERIFY\nagent-browser"]
    JEO --> UI["🎨 VERIFY_UI\nagentation"]
    JEO --> CLEAN["🧹 CLEANUP\nworktree"]

    PLAN --> OMC["omc\nClaude Code"]
    PLAN --> OHMG["ohmg\nGemini CLI"]
    PLAN --> OMX["omx\nCodex CLI"]

    SURVEY["🔭 survey"] -.-> JEO
    RALPH["🔄 ralph"] -.-> EXEC
    AUTORESEARCH["🔬 autoresearch"] -.-> EXEC

---

🆕 What's New in v2026-03-16

Change	Details
jeo v1.3.1: plannotator feedback loop accept fix	Bug fix: When plannotator returned feedback_required and the plan hash was unchanged, the SKILL.md pre-flight guard incorrectly exited 0 (approved), causing JEO to enter EXECUTE without real approval. Fixed: same hash + feedback_required now correctly exits 1 (not approved) on Codex/Gemini/OpenCode. Updated FLOW.md with recovery procedure and feedback_required state semantics.
firebase-ai-logic: Firebase AI Logic (Gemini in Firebase)	Firebase AI SDK integration — text generation, image analysis, multimodal prompts, streaming responses. Works with Claude, GPT-4o, Gemini. Total: 76 skills.
bmad-orchestrator v1.3.0: SSD + TEA cycles	SSD (Structured System Design) meta-framework embeds TEA (Task→Execute→Architect) cycles in each BMAD phase. Automated architect validation before plannotator human review. Cross-phase traceability. ssd-state.json task-level state. New commands: /workflow-init --ssd, /ssd-cycle, /ssd-decompose, /ssd-execute, /ssd-validate, /ssd-advance, /ssd-status. Fabric pattern bmad_ssd_phase_review bundled. Platform expanded to All (Claude · Gemini · Codex · OpenCode).

What's New in v2026-03-13

Change	Details
langsmith: LLM observability, tracing & evaluation	Framework-agnostic LangSmith skill — @traceable decorator, wrap_openai/wrap_anthropic, evaluate() with LLM-as-judge (openevals), Prompt Hub versioning, dataset management, online/offline evaluation, distributed tracing. Python + TypeScript. Total: 75 skills.
firebase-cli: complete Firebase CLI skill	Full Firebase CLI reference — Hosting, Cloud Functions, Firestore, Realtime DB, Auth import/export, Emulator Suite, Remote Config, App Distribution, Extensions, App Hosting. scripts/install.sh + deploy.sh + emulators.sh. Total: 75 skills.
react-grab: browser element context for AI agents	Hover over any React UI element → Cmd+C → copy component name + file path + HTML to clipboard for Claude Code / Cursor / Copilot / Gemini. MCP server, plugin API, primitives API. Total: 74 skills.
Removed deprecated agent-development skills	Removed agent-configuration, agent-evaluation, agentic-development-principles, agentic-principles, agentic-workflow.
Removed deprecated image/media skills	Removed image-generation, image-generation-mcp, pollinations-ai. Use remotion-video-production / video-production for media.
autoresearch: Karpathy autonomous ML experiment skill	AI agent modifies train.py, runs 5-min GPU experiments, evaluates with val_bpb, ratchets improvements via git. Includes scripts/ and references/.
jeo v1.2.3: plannotator-plan-loop.sh all-platform hardening	Cross-platform temp dir fallback, dedicated port PLANNOTATOR_PORT=47291, probe_plannotator_port() + wait_for_listen(), browser-crash retry up to 3 times, structured jeo-blocked.json output.
survey: cross-platform landscape scan	4-lane discovery flow, artifacts to .survey/{slug}/, Claude/Codex/Gemini abstraction as settings/rules/hooks.
presentation-builder: slides-grab workflow	HTML-first deck creation, visual editing, PPTX/PDF export. Removed duplicate pptx-presentation-builder.

---

📦 Installation

Step 0: Install skills CLI

npm install -g skills
skills --version

For LLM Agents

curl -s https://raw.githubusercontent.com/supercent-io/skills-template/main/setup-all-skills-prompt.md

Choose by Platform

Claude Code

npx skills add https://github.com/supercent-io/skills-template \
  --skill jeo --skill omc --skill plannotator --skill agentation \
  --skill ralph --skill ralphmode --skill vibe-kanban

Gemini CLI

npx skills add https://github.com/supercent-io/skills-template \
  --skill jeo --skill ohmg --skill ralph --skill ralphmode --skill vibe-kanban
gemini extensions install https://github.com/supercent-io/skills-template

Codex CLI

npx skills add https://github.com/supercent-io/skills-template \
  --skill jeo --skill omx --skill ralph --skill ralphmode

Platform-Specific Setup

# Claude Code — jeo hook setup
bash ~/.agent-skills/jeo/scripts/setup-claude.sh

# Gemini CLI — jeo hook setup
bash ~/.agent-skills/jeo/scripts/setup-gemini.sh

# oh-my-claudecode
/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode
/omc:omc-setup

setup-claude.sh repairs legacy Claude hook entries that still point at ~/.agent-skills/omg/... and installs compatibility shims so stale UserPromptSubmit paths do not keep breaking prompt submission.

---

📚 Skills List (76 total)

Full manifest: .agent-skills/skills.json · each folder's SKILL.md

🎯 Core Orchestration (10)

Skill	Keyword	Platform	Description
jeo	jeo, annotate	All	Integrated orchestration: PLAN→EXECUTE→VERIFY→CLEANUP
omc	omc, autopilot	Claude	32-agent orchestration layer with model routing
omx	omx	Codex	Multi-agent orchestration for Codex CLI
ohmg	ohmg	Gemini	Antigravity multi-agent framework
ralph	ralph, ooo	All	Ouroboros specification-first + persistent completion loop
ralphmode	ralphmode	All	Automation permission profiles (sandbox-first, repo boundary)
bmad-orchestrator	bmad	All	Structured phase-based AI development with SSD + TEA (Task→Execute→Architect) cycles
bmad-gds	bmad-gds	All	BMAD Game Development Studio (Unity · Unreal · Godot)
bmad-idea	bmad-idea	All	Creative intelligence — 5 specialist ideation agents
survey	survey	All	Pre-implementation landscape scan

📋 Planning & Review (5)

Skill	Keyword	Description
plannotator	plan	Visual browser plan/diff review — approve or send feedback
agentation	annotate	UI annotation → targeted agent code fixes
agent-browser	agent-browser	Headless browser verification for AI agents
playwriter	playwriter	Playwright automation connecting to live browser
vibe-kanban	kanbanview	Visual Kanban board with git worktree isolation

🤖 Agent Development (2)

Skill	Description	Platforms
prompt-repetition	LLM accuracy via prompt repetition technique	All
skill-standardization	SKILL.md validation against Agent Skills spec	All

⚙️ Backend (5)

Skill	Description	Platforms
api-design	REST/GraphQL API design	All
api-documentation	OpenAPI/Swagger docs generation	All
authentication-setup	JWT, OAuth, session management	All
backend-testing	Unit/integration/API test strategies	All
database-schema-design	SQL/NoSQL schema design	All

🎨 Frontend (10)

Skill	Description	Platforms
design-system	Design tokens, layout rules, motion, accessibility	All
frontend-design-system	Production-grade UI with design tokens and accessibility	All
react-best-practices	React & Next.js performance optimization	All
react-grab	Browser element context capture for AI agents (Cmd+C → component name + file path + HTML)	All
vercel-react-best-practices	Vercel Engineering React & Next.js guidelines	Claude · Gemini · Codex
responsive-design	Mobile-first layouts and breakpoints	All
state-management	Redux, Context, Zustand patterns	All
ui-component-patterns	Reusable component libraries	All
web-accessibility	WCAG 2.1 compliance	All
web-design-guidelines	Web Interface Guidelines compliance review	All

🔍 Code Quality (5)

Skill	Description	Platforms
code-refactoring	Code simplification and refactoring	All
code-review	Comprehensive code review with API contracts	All
debugging	Root cause analysis, regression isolation	All
performance-optimization	Speed, efficiency, scalability optimization	All
testing-strategies	Test pyramid, coverage, flaky-test hardening	All

🏗 Infrastructure (12)

Skill	Description	Platforms
ai-tool-compliance	Internal AI tool compliance automation (P0/P1)	All
deployment-automation	CI/CD, Docker/Kubernetes, cloud infrastructure	All
environment-setup	Dev/staging/production environment config	All
firebase-ai-logic	Firebase AI Logic (Gemini in Firebase) — text generation, image analysis, multimodal prompts	All
firebase-cli	Firebase CLI — Hosting, Functions, Firestore, Auth, Emulators, App Distribution	All
genkit	Firebase Genkit AI flows and RAG pipelines	Claude · Gemini
langsmith	LLM observability, tracing & evaluation — @traceable, evaluate(), Prompt Hub, openevals LLM-as-judge, datasets	All
looker-studio-bigquery	Looker Studio + BigQuery dashboards	All
monitoring-observability	Health checks, metrics, log aggregation	All
security-best-practices	OWASP Top 10, RBAC, API security	All
system-environment-setup	Reproducible environment configuration	All
vercel-deploy	Vercel deployment automation	All

📝 Documentation (4)

Skill	Description	Platforms
changelog-maintenance	Changelog management and versioning	All
presentation-builder	HTML slides with slides-grab, PPTX/PDF export	All
technical-writing	Technical documentation and specs	All
user-guide-writing	User guides and tutorials	All

📊 Project Management (4)

Skill	Description	Platforms
sprint-retrospective	Sprint retrospective facilitation	All
standup-meeting	Daily standup management	All
task-estimation	Story points, t-shirt sizing, planning poker	All
task-planning	Task breakdown and user stories	All

🔭 Search & Analysis (5)

Skill	Description	Platforms
autoresearch	Autonomous ML experiments (Karpathy) — AI agent runs overnight GPU experiments, ratchets improvements via git	All
codebase-search	Codebase search & navigation	All
data-analysis	Dataset analysis, visualizations, statistics	All
log-analysis	Log analysis and incident debugging	All
pattern-detection	Pattern and anomaly detection	All

🎬 Creative Media (2)

Skill	Description	Platforms
remotion-video-production	Programmable video production with Remotion	All
video-production	Produce programmable videos with Remotion — scene planning, asset orchestration	All

📢 Marketing (2)

Skill	Description	Platforms
marketing-automation	23 sub-skills: CRO, copywriting, SEO, analytics, growth	All
marketing-skills-collection	23 sub-skills: CRO, copywriting, SEO, analytics, growth	All

🔗 Integrations (1)

Skill	Keyword	Description	Platforms
google-workspace	Google Doc, Google Sheet, spreadsheet, Google Slides, Google Drive, Gmail, Google Calendar, schedule meeting, Google Chat, Google Forms, Workspace user, Apps Script, 구글 문서, 구글 시트, 구글 드라이브, 구글 캘린더, 이메일 보내기	Full Google Workspace suite: Docs · Sheets · Slides · Drive · Gmail · Calendar · Chat · Forms · Admin SDK · Apps Script	All

🔧 Utilities (9)

Skill	Description	Platforms
copilot-coding-agent	GitHub Copilot Coding Agent — issue → Draft PR automation	Claude · Codex
fabric	AI prompt patterns — YouTube summaries, document analysis via 200+ Patterns	All
file-organization	File and folder organization	All
git-submodule	Git submodule management	All
git-workflow	Commit, branch, merge, PR workflows	All
llm-monitoring-dashboard	LLM usage monitoring dashboard (cost, tokens, latency)	All
npm-git-install	Install npm packages from GitHub	All
opencontext	Persistent memory and context management for AI agents	All
workflow-automation	Automate repetitive development workflows	All

---

🧬 TOON Format Injection

TOON (Token-Oriented Object Notation) compresses the skill catalog and auto-injects it into every prompt. 40-50% token savings vs JSON/Markdown.

Platform	File	Mechanism
Claude Code	~/.claude/hooks/toon-inject.mjs	UserPromptSubmit hook — 26-37ms
Gemini CLI	~/.gemini/hooks/toon-skill-inject.sh	includeDirectories session load
Codex CLI	~/.codex/skills-toon-catalog.toon	Static catalog

• Tier 1 (always): Skill catalog index (~875-3,500 tokens) — names + descriptions + tags
• Tier 2 (on-demand): Individual SKILL.toon content (~292 tokens/skill, max 3)

---

🔮 Featured Tools

jeo — Integrated Agent Orchestration

Keyword: jeo · annotate | Platforms: Claude · Codex · Gemini · OpenCode

Complete automated pipeline: Plan (ralph+plannotator) → Execute (team/bmad) → Verify (agent-browser) → UI Feedback (agentation) → Cleanup.

Phase	Tool	Description
Plan	ralph + plannotator	Visual plan review → Approve/Feedback
Execute	omc team / bmad	Parallel agent execution
Verify	agent-browser	Browser behavior verification
Verify UI	agentation (annotate)	UI annotation → fix loop
Cleanup	worktree-cleanup.sh	Auto worktree cleanup

plannotator — Visual Plan Review

Keyword: plan | Docs | GitHub

Browser UI for annotating AI plans. Approve or send structured feedback in one click. Works with Claude Code, OpenCode, Gemini CLI, and Codex CLI.

bash scripts/install.sh --all

ralph — Specification-First Development

Keyword: ralph, ooo | Docs | GitHub

Socratic interview → immutable spec → Double Diamond execution → 3-stage verification → loop until passed.

ooo interview "I want to build a task management CLI"
ooo seed && ooo run && ooo evaluate <session_id>
ooo ralph "fix all failing tests"

vibe-kanban — AI Agent Kanban Board

Keyword: kanbanview | Docs | GitHub

Visual Kanban (To Do → In Progress → Review → Done) with parallel AI agents isolated via git worktrees.

npx vibe-kanban

---

🌐 Recommended Harness OSS

Repository	Stars	Description
AutoGPT	182k	Accessible AI platform for continuous agents
AutoGen	55.4k	Microsoft multi-agent conversation framework
CrewAI	45.7k	Role-playing autonomous AI agent orchestration
smolagents	25.9k	HuggingFace code-thinking agent library
agency-agents	21.2k	61 specialized AI agents across 9 divisions

Install & integration notes → docs/harness/README.md

---

📁 Structure

.
├── .agent-skills/          ← 76 skill folders (each with SKILL.md + SKILL.toon)
├── docs/                   ← detailed guides (bmad, omc, plannotator, ralph, ...)
├── install.sh
├── setup-all-skills-prompt.md
├── README.md               ← English (this file)
└── README.ko.md            ← 한국어

---

📖 Related Docs

Tool	Keyword	Doc
jeo	jeo, annotate	.agent-skills/jeo/SKILL.md
plannotator	plan	docs/plannotator/README.md
vibe-kanban	kanbanview	docs/vibe-kanban/README.md
ralph	ralph	docs/ralph/README.md
omc	omc	docs/omc/README.md
bmad-orchestrator	bmad	docs/bmad/README.md
Harness OSS	—	docs/harness/README.md

---

📎 References

Component	Source	License
jeo	Internal	MIT
omc	Yeachan-Heo/oh-my-claudecode	MIT
ralph	Q00/ouroboros	MIT
plannotator	plannotator.ai	MIT
bmad	bmad-dev/BMAD-METHOD	MIT
agentation	benjitaylor/agentation	MIT
fabric	danielmiessler/fabric	MIT
autoresearch	Andrej Karpathy methodology	—
Agent Skills Spec	agentskills.io	—

---

<div align="center">

⭐ If this helps you, star the repository!

gh api --silent --method PUT /user/starred/supercent-io/skills-template >/dev/null 2>&1 || true

</div>