Agent Skills

Quickstart • Skills • Personas • Hooks • Install • Contributing
한국어

What is this?

A collection of skills, personas, and hooks for AI coding agents (Claude Code, Codex CLI, Gemini CLI). Each skill is a self-contained markdown module that gives an agent domain-specific capabilities.

CLI tool: Install and manage skills with agt — npm install -g @open330/agt or just npx @open330/agt

Quickstart for Agents

Install the agent-skills toolkit into this project:

1. Install the "core" profile (essential skills for every workspace):
   npx @open330/agt skill install --profile core

2. Install all available personas globally:
   npx @open330/agt persona install --global --from jiunbae/agent-skills

3. Verify installation:
   npx @open330/agt skill list
   npx @open330/agt persona list

Note: Use `npm install -g @open330/agt` first if you prefer running `agt` directly.

Slowstart for Humans

With npx (No Install)

npx @open330/agt skill install --profile core          # Install core skills
npx @open330/agt skill install -g --from jiunbae/agent-skills  # Browse & install from repo
npx @open330/agt persona install -g --from jiunbae/agent-skills

With agt CLI (Global Install)

npm install -g @open330/agt
agt skill install --profile core
agt skill install -g git-commit-pr     # Install a skill globally
agt persona install -g --all           # Install all personas globally

With install.sh

git clone https://github.com/jiunbae/agent-skills ~/.agent-skills
cd ~/.agent-skills

./install.sh --core                    # Core skills only
./install.sh --core --hooks            # Core + hooks
./install.sh all --link-static --codex # Everything
./install.sh --list                    # List available options

Install Options

Core Skills

Installed by default with --core:

• development/git-commit-pr — Git commit & PR guide
• context/context-manager — Project context auto-loader
• context/static-index — Global static context index
• security/security-auditor — Repository security audit
• agents/background-implementer — Background parallel implementation
• agents/background-planner — Background parallel planning
• agents/background-reviewer — Multi-LLM parallel code review

Skills Catalog

🤖 agents/ — AI Agents

🛠 development/ — Dev Tools

📊 business/ — Business

🔗 integrations/ — Integrations

🧠 ml/ — ML/AI

🔐 security/ — Security

📁 context/ — Context Management

🔧 meta/ — Meta Skills

Personas

Expert identities as markdown files — usable with any AI agent.

Using with agt CLI

agt persona review security-reviewer --codex
agt persona review security-reviewer --codex "is this architecture scalable?"
agt persona install -g --all
agt persona show security-reviewer

Using Directly

Personas are just .md files. Any agent that can read a file can adopt a persona:

cat personas/security-reviewer.md | codex -q "Review this code"
cat personas/security-reviewer.md   # pipe to any AI agent

.agents/personas/security-reviewer.md    ← project local (highest priority)
~/.agents/personas/security-reviewer.md  ← user global
personas/security-reviewer.md            ← library (bundled)

Hooks

Event-driven automation for Claude Code.

./install.sh --hooks

Creating Your Own Skills Repo

You can create your own skills repository with the same structure as this one. agt will discover it automatically.

Directory Structure

my-skills/
├── agents/                    # Group: AI agent skills
│   └── my-agent/
│       └── SKILL.md           # Required
├── development/               # Group: dev tools
│   └── my-tool/
│       ├── SKILL.md           # Required
│       ├── scripts/           # Optional: helper scripts
│       └── templates/         # Optional: templates
├── personas/                  # Personas directory
│   ├── my-reviewer/
│   │   └── PERSONA.md
│   └── my-expert.md           # Single-file persona also works
├── profiles.yml               # Optional: installation profiles
└── README.md

SKILL.md Format

---
name: my-skill-name
description: "What this skill does. Use for 'keyword1', 'keyword2' requests."
trigger-keywords: keyword1, keyword2, 키워드
allowed-tools: Read, Write, Edit, Bash, Grep, Glob
priority: high
tags: [category, tool-name]
---

# Skill Title

## Quick Start

\`\`\`bash
command --example
\`\`\`

## Workflow

Step-by-step instructions for the AI agent...

## Best Practices

**DO:**
- Follow this pattern

**DON'T:**
- Avoid this anti-pattern

PERSONA.md Format

---
name: my-reviewer
role: "Senior Engineer — Domain Expert"
domain: security
type: review
tags: [security, owasp, auth]
---

# My Reviewer

## Identity

Background, expertise, and review philosophy...

## Review Lens

When reviewing, evaluate:
1. Security vulnerabilities
2. Best practice adherence
...

## Output Format

### Findings
#### [SEVERITY] Finding Title
- **File**: path:line
- **Impact**: Description
- **Recommendation**: Fix

profiles.yml Format

core:
  description: "Essential skills for every workspace"
  skills:                          # Explicit skill paths
    - development/git-commit-pr
    - context/context-manager
  groups:                          # Include entire groups
    - agents

dev:
  description: "Development workflow tools"
  groups:
    - development

Test & Install

# Point agt to your repo
export AGT_DIR=~/my-skills
# Or clone to the default location
git clone https://github.com/you/my-skills ~/.agent-skills

agt skill list                     # Verify skills discovered
agt skill install my-skill -g      # Install globally
agt skill install --profile core   # Install a profile
agt persona install my-reviewer    # Install a persona

# Others can install from your repo directly
agt skill install --from you/my-skills
agt persona install --from you/my-skills

Create with AI (Prompts)

Copy-paste these prompts to create skills and personas with any AI agent.

Create a Skill

Create a new skill for the agent-skills repository.

Skill name: [YOUR_SKILL_NAME]
Group: [GROUP_NAME] (e.g., development, integrations, agents, ml, security)
Description: [WHAT_THE_SKILL_DOES]

Create the file at [GROUP]/[SKILL_NAME]/SKILL.md with this structure:

1. YAML frontmatter with: name, description (include trigger keywords and Korean keywords
   if applicable), trigger-keywords, allowed-tools, priority, tags
2. A "Quick Start" section with 2-3 essential commands/examples
3. A "Workflow" section with step-by-step instructions
4. A "Best Practices" section with DO/DON'T guidelines

Rules:
- Description should be third person, under 200 chars, end with trigger conditions
- Keep under 300 lines total
- Code examples should be practical and production-ready
- Include both English and Korean trigger keywords where relevant
- Focus on what the AI agent should DO, not theory

Reference this existing skill for style:
---
name: git-commit-pr
description: "Guides git commit and PR creation with security validation. Activates on
  commit, PR, pull request requests."
trigger-keywords: commit, pr, pull request, 커밋
allowed-tools: Read, Bash, Grep, Glob
priority: high
tags: [git, commit, pr, security]
---

Create a Persona

Create a new code review persona for the agent-skills repository.

Persona name: [YOUR_PERSONA_NAME]-reviewer
Domain: [DOMAIN] (e.g., security, architecture, performance, database, frontend, devops)
Role: [JOB_TITLE]

Create the file at personas/[PERSONA_NAME]/PERSONA.md with this structure:

1. YAML frontmatter with: name, role, domain, type (review), tags
2. "Identity" section: background, expertise (5+ bullet points), attitude (2-3 sentences)
3. "Review Lens" section: numbered list of what they evaluate
4. "Evaluation Framework" section: table with Category, Severity (CRITICAL/HIGH/MEDIUM/LOW), Criteria
5. "Output Format" section: markdown template for review output
6. "Red Flags" section: patterns that MUST always be flagged
7. "Key Principles" section: numbered guiding principles

Rules:
- The persona should be opinionated and have a clear point of view
- Include specific technologies/frameworks in their expertise
- Red flags should be concrete, not generic
- Output format should include: Summary, Findings (with severity), Positive Observations
- Keep under 200 lines

Reference this existing persona for style:
---
name: security-reviewer
role: "Senior Application Security Engineer"
domain: security
type: review
tags: [security, owasp, auth, injection, xss, ssrf]
---

Create a Profile

Create a new installation profile for the agent-skills repository.

Profile name: [PROFILE_NAME]
Purpose: [WHO_IS_THIS_FOR]
Available groups: agents, development, business, integrations, ml, security, context, meta

Add the profile to profiles.yml with:
- description: one-line description of the profile
- skills: list of specific group/skill-name paths to include (optional)
- groups: list of group names to include all skills from (optional)

You can mix both "skills" (specific picks) and "groups" (entire categories).

Example:
```yaml
backend:
  description: "Backend development essentials"
  skills:
    - context/context-manager
    - security/security-auditor
  groups:
    - development
    - integrations

</details>

### Create a Complete Skills Repo

<details>
<summary><b>Prompt: Bootstrap a new skills repository</b></summary>

</details>

---

## Architecture

---

## Contributing

1. **Add a skill** — Create a new skill in the appropriate category
2. **Add a persona** — Create a domain expert persona
3. **Improve docs** — Fix typos, add examples, translate
4. **Report issues** — Bug reports and feature requests welcome

```bash
git clone https://github.com/jiunbae/agent-skills ~/.agent-skills
cd ~/.agent-skills
./install.sh --core