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
This repository is the single source of truth for skill content. The Rust CLI,
npm packages, and platform releases are maintained only in Open330/agt.
Quickstart for Agents
Install the agent-skills toolkit into this project:
1. Install the "core" profile globally for Claude Code:
npx @open330/agt skill install --profile core --from jiunbae/agent-skills --global
2. Install the same profile globally for Codex:
npx @open330/agt skill install --profile core --from jiunbae/agent-skills --global --agent codex
3. Install all available personas globally:
npx @open330/agt persona install --global --from jiunbae/agent-skills
4. Verify installation:
npx @open330/agt skill list
npx @open330/agt skill list --agent codex
npx @open330/agt persona list
5. Optional — the "core" profile does not include `agents/air-workbench`.
Install that one explicitly only if you want the local graph workbench:
npx @open330/agt skill install --from jiunbae/agent-skills/agents/air-workbench --global
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 --from jiunbae/agent-skills
npx @open330/agt skill install --profile core --from jiunbae/agent-skills --agent codex
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 --from jiunbae/agent-skills
agt skill install --profile core --from jiunbae/agent-skills --agent codex
agt skill install -g --from jiunbae/agent-skills/development/git-commit-pr
agt persona install -g --from jiunbae/agent-skills
--agent claude is the default and installs to .claude/skills/<skill>.
--agent codex installs to .agents/skills/<skill>. Add --global for the
corresponding user-level directory. Skills an older agt placed under
<group>/<skill> are not loaded by Claude Code. agt releases after 2026.7.23
add --claude-dir (or CLAUDE_CONFIG_DIR) to pick another config directory
and agt skill migrate --global to move grouped installs.
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 --codex # Core skills + Codex user skill links
./install.sh --core --hooks # Core + hooks
./install.sh all --link-static --codex # Everything
./install.sh --list # List available options
Install Options
| Option | Description |
|---|
--core | Install core skills globally (recommended) |
--cli | Install the published @open330/agt CLI plus legacy compatibility commands |
--link-static | Link each static/* item under ~/.agents while preserving ~/.agents/skills |
--codex | Link selected skills individually into ~/.agents/skills without replacing Codex system skills (also links static/* items under ~/.agents) |
--hooks | Install Claude Code hooks |
--personas | Install agent personas |
--copy | Copy instead of symlink |
--dry-run | Preview only |
--uninstall | Remove installed skills |
Core Skills
Installed by default with --core:
development/git-commit-pr — Git commit & PR guidecontext/static-index — Global static context indexsecurity/security-auditor — Repository security auditagents/background-implementer — Isolated parallel implementation with verified integrationagents/background-planner — Parallel persona planning with stance-steered synthesisagents/background-reviewer — Parallel persona review with adversarial verificationagents/rpf — Pointer-driven iterative review, plan, work, and feedback
Skill Naming
A Skill's name frontmatter must equal its directory name. A host lists a Skill
by its directory, so any other declared name is a second name nobody can type,
and catalog items are keyed on the frontmatter name, which makes the divergence
observable rather than cosmetic. agents/air-workbench enforces this in
tests/core.test.mjs.
Opt-in Skills (not in core)
Every other skill in the catalog is installed on request. agents/air-workbench
is deliberately one of them: it ships a local HTTP server, a checked-in browser
bundle, and an approval-gated native-run path, so it stays opt-in rather than
being added to every workspace by default. Nothing in --profile core or
--core installs it. Pick whichever path you already use:
# npx
npx @open330/agt skill install --from jiunbae/agent-skills/agents/air-workbench --global
# agt CLI
agt skill install -g --from jiunbae/agent-skills/agents/air-workbench
# install.sh, from a repository checkout
./install.sh agents/air-workbench
Then run it from the installed copy — it needs no npm install:
node ~/.claude/skills/air-workbench/scripts/air.mjs workbench
The same form works for any other catalog entry, for example
--from jiunbae/agent-skills/development/playwright or
./install.sh integrations/slack-skill.
Skills Catalog
🤖 agents/ — AI Agents
| Skill | Description |
|---|
background-implementer | Isolated parallel implementation (worktrees) with pre-integration verification |
background-planner | Bounded parallel planning with planner personas and stance-steered synthesis |
background-reviewer | Bounded parallel persona review with adversarial verification and root-cause merge |
incident-writer | Structured incident and status-page reports |
rpf | Explicit pointer-driven multi-agent review, plan, work, and feedback loop |
air-workbench | AIR Skill workflow editing, local Skill discovery, plan approval, and observable tracing |
🛠 development/ — Dev Tools
| Skill | Description |
|---|
appstore-screenshots | App Store screenshot capture/upload and authenticated ASC inspection |
context-worktree | Auto git worktree per task |
git-commit-pr | Git commit & PR generation guide |
grill-me | Adversarial plan questioning before implementation |
iac-deploy-prep | IaC deployment prep (K8s, Dockerfile, CI/CD) |
playwright | Playwright browser automation |
📊 business/ — Business
| Skill | Description |
|---|
bm-analyzer | Business model analysis & monetization strategy |
proposal-analyzer | Proposal / RFP analysis |
🔗 integrations/ — Integrations
| Skill | Description |
|---|
discord-skill | Discord REST API |
notion-summary | Notion page upload |
obsidian-tasks | Obsidian TaskManager (Kanban, Dataview) |
obsidian-writer | Obsidian Vault writing and docs.jiun.dev publishing |
service-manager | Docker container & service management |
slack-skill | Slack app development & API |
vault-secrets | Vaultwarden credentials & API key management |
🧠 ml/ — ML/AI
| Skill | Description |
|---|
audio-processor | ffmpeg-based audio processing |
ml-benchmark | ML model benchmarking |
model-sync | Model file server sync |
triton-deploy | Triton Inference Server deployment |
🔐 security/ — Security
| Skill | Description |
|---|
security-auditor | Repository security audit |
📁 context/ — Context Management
| Skill | Description |
|---|
static-index | Global static context index with user profile |
🔧 meta/ — Meta Skills
| Skill | Description |
|---|
skill-manager | Skill ecosystem management |
✍️ common/ — Writing
| Skill | Description |
|---|
korean-editor | Conservative Korean copy editing with fidelity checks |
Personas
Expert identities as markdown files — usable with any AI agent.
| Persona | Role | Domain |
|---|
security-reviewer | Senior AppSec Engineer | OWASP, auth, injection |
architecture-reviewer | Principal Architect | SOLID, API design, coupling |
code-quality-reviewer | Staff Engineer | Readability, complexity, DRY |
performance-reviewer | Performance Engineer | Memory, CPU, I/O, scalability |
database-reviewer | Senior DBA | Query optimization, schema, indexing |
frontend-reviewer | Senior Frontend Engineer | React, accessibility, performance |
devops-reviewer | Senior DevOps/SRE | K8s, IaC, CI/CD |
Above is a featured subset. Planning personas (type: planning) — technical-planner, product-planner, delivery-risk-planner — drive background-planner. Full library: personas/README.md.
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
| Hook | Event | Description |
|---|
english-coach | UserPromptSubmit | Rewrites prompts in natural English with vocabulary |
prompt-logger | UserPromptSubmit | Captures each submitted prompt into omp in the background |
stop-capture | Stop | Captures each completed response into omp in the background |
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, when it should trigger, and when it should not. Use for 'keyword1' and 'keyword2' requests."
---
# 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
| Field | Required | Description |
|---|
name | Yes | Skill identifier (kebab-case) |
description | Yes | What the skill does, trigger terms, scope, and important non-trigger boundaries |
Keep the portable frontmatter baseline to name and description. Put Codex UI
metadata, tool dependencies, and implicit-invocation policy in
agents/openai.yaml. Add host-specific frontmatter only when that host's current
validator explicitly supports it.
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
| Field | Required | Description |
|---|
name | Yes | Persona identifier (kebab-case) |
role | Yes | Job title / expert role |
domain | Yes | Expertise domain |
type | Yes | Usually review |
tags | Yes | Relevant focus areas |
profiles.yml Format
core:
description: "Essential skills for every workspace"
skills: # Explicit skill paths
- development/git-commit-pr
- context/static-index
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 only `name` and `description`; include English/Korean trigger
terms and important non-trigger boundaries in the description
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 and state what the skill does, when to use it, and when not to
- 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 local Git commit preparation and PR drafting with security checks. Use for commit, PR, pull request, and 커밋 requests."
---
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: concrete patterns the reviewer flags whenever they appear
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, common
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:
- security/security-auditor
groups:
- development
- integrations
Keep profiles focused — a profile with everything is just "full" or "all".
</details>
### Create a Complete Skills Repo
<details>
<summary><b>Prompt: Bootstrap a new skills repository</b></summary>
Create a new agent-skills repository with the following structure. This repo will be
discoverable by the agt CLI tool (https://github.com/open330/agt).
Repository purpose: [YOUR_PURPOSE]
Skills to include: [LIST_OF_SKILLS]
Personas to include: [LIST_OF_PERSONAS]
Create the following structure:
- Group directories (e.g., development/, integrations/) containing skill subdirectories
- Each skill has a SKILL.md with portable YAML frontmatter (
name and description)
and concise Markdown instructions; use agents/openai.yaml for Codex UI metadata,
dependency declarations, or implicit-invocation policy
- A personas/ directory with PERSONA.md files for each persona
- A profiles.yml defining installation profiles (core, dev, full at minimum)
- A README.md documenting the repository
Requirements:
- Skills are organized by group: each group is a directory, each skill is a subdirectory
- A valid skill must have SKILL.md in its directory
- A valid persona must have PERSONA.md (in a directory or as a standalone .md file)
- profiles.yml uses "skills" (explicit paths) and/or "groups" (entire categories)
- The "all" profile is built-in to agt, no need to define it
Users install with:
git clone https://github.com/you/repo ~/.agent-skills
agt skill install --profile core -g
agt persona install --all -g
</details>
---
## Architecture
agent-skills/ open330/agt (CLI tool)
├── agents/ AI agent skills ├── agt/ Rust CLI
├── development/ Dev tool skills ├── npm/ npm packaging
├── business/ Business skills ├── setup.sh Installer
├── integrations/ Integration skills └── assets/ Branding
├── ml/ ML/AI skills
├── security/ Security skills
├── context/ Context management
├── meta/ Meta skills
├── common/ Writing skills
├── personas/ Expert personas
├── hooks/ Claude Code hooks
├── static/ Global context
├── install.sh Local installer
└── codex-support/ Codex CLI
---
## 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
For CLI tool contributions, see open330/agt.
License
MIT License. See LICENSE, which also records the third-party work
some skills are adapted from.
