Skip to content

Building a Claude Code Plugin Marketplace

SeungAh Hong8min read

Overview

This document describes how to register, install, and use the Claude Code plugin marketplace.

Plugins are shared through the seungah-claude-plugins marketplace. It currently consists of the following four team-specific plugins.

Marketplace version: 1.4.0 · Repository: https://github.com/seungahhong/seungah-claude-plugins

PluginDescriptionVersion
frontend-harnessA collection of multi-agent skills, commands, and hooks that support the entire frontend development process (Research → PRD → Develop → Review → Verify). Includes parallel execution of user-selected items among 5 static analysis perspectives (simplification, PR review, security, performance, consistency) plus a re-review loop, along with E2E and type/build verification1.3.0
harness-generatorA meta-plugin that automatically generates a domain-agnostic harness (agent team + skills + orchestrator) through a 7-step meta process0.1.0
git-harnessGit workflow — writing Korean commit messages (issue-number type: subject) and multi-perspective PR reviews (review → commit → PR, all in one)0.2.0
meta-harnessMeta-harness engineering based on a full-trace experience store. Causal diagnosis + 4-axis Pareto non-regression patches, with a user approval gate (based on arXiv 2603.28052v1)0.1.0

Installation

Installation supports three methods. For sharing within a team, Method A (install after registering the marketplace) is recommended.

This proceeds in two steps.

Step 1. Add the marketplace

Register the marketplace registry in your local environment. This step only registers the list of available plugins; it does not activate any skills.

# seugnah-claude-plugins
/plugin marketplace add https://github.com/seungahhong/seungah-claude-plugins

(Note) If the marketplace cannot be added because the github clone fails,

gh auth login

Step 2. Install a plugin

Select and install the plugin you want from the registered marketplace.

# Plugins in seugnah-claude-plugins
/plugin install frontend-harness@seungah-claude-plugins
/plugin install frontend-harness

You can choose from the following three installation scopes.

ScopeDescription
user scopeInstalls globally for the user
project scopeInstalls for all collaborators of the current repository
local scopeInstalls only for yourself in the current repository

Method B - Install directly from GitHub

This method adds the marketplace directly from the GitHub repository via the CLI.

claude plugin add seungahhong/seungah-claude-plugins

Method C - Install from a local path

This method installs by specifying the path of a repository cloned locally. It is useful during development and debugging.

claude --plugin-dir /path/to/frontend-harness

Marketplace Registration Screens

Below is the process of registering the seungah-claude-plugins marketplace and installing plugins.

Step 1. Enter the /plugin command

Launch Claude Code and enter the /plugin command.

/plugin command autocomplete screen

Step 2. Add Marketplace - Enter the marketplace URL

Select Add Marketplace and enter the marketplace repository URL.

seungah-claude-plugins URL input screen

Step 3. Discover - Check the list of installable plugins

Check the list of installable plugins in the Discover tab.

  • Discover tab plugin list
  • Press space to select → press enter to proceed with installation

Step 4. Plugin Details - Plugin details and scope selection

When you select a plugin to install, its details are displayed. Choose the installation scope (user/project/local scope).

frontend-harness Plugin Details screen

Step 5. Confirm installation - settings.local.json

Once installation is complete, the enabledPlugins setting is added to .claude/settings.local.json.

settings.local.json confirmation screen

Step 6. Run a skill

Run a skill from the installed plugin using the /skill-name format.

  • /a11y skill execution screen
  • If it does not appear, run /reload-plugins and retry the /(slash) command

Plugin Components

Each team plugin is used by combining the components below.

ComponentFormatHow to runDescription
skillsskills/<name>/SKILL.md/<plugin>:<skill>Performs tasks based on specific domain knowledge
commandscommands/<name>.md/<plugin>:<command>Tasks run via slash commands
agentsagents/<name>.mdAutomatic executionComplex multi-step workflows
hookshooks.jsonEvent triggerAutomation that reacts to Claude Code events

There is no need to modify marketplace.json when adding components. Since paths are declared in plugin.json, you only need to add files to the corresponding directory.


Skill Configuration

frontend-harness plugin (12 skills)

SkillCommandDescriptionTrigger keywords
Planner/planner4-step PRD creation: interview → research → plan → approval계획 세워줘, plan, PRD 작성, 구현 계획
Architecture/architectureDesigns system structure, APIs, data flow, and technology choices설계해줘, 아키텍처, API 설계, 시스템 설계
Critic/criticAnalyzes design weaknesses, risks, and omissions검토해줘, 비평, 리스크 분석, 약점 분석
Grill Me/grill-mePersistent questioning to clarify requirements질문해줘, grill me, 인터뷰, 요구사항 정리
TDD/tddTest-driven development based on the Red-Green-Refactor loopTDD, 테스트 주도 개발, 테스트 먼저
A11y/a11yWAI-ARIA-based web accessibility inspection and improvement웹접근성, a11y, aria, 스크린리더, WCAG
Semantic HTML/semantic-htmlInspects and improves semantic tag usage and heading hierarchy시맨틱, semantic, HTML 구조, heading
SEO/GEO Optimizer/seo-geo-optimizerOptimization for traditional search + AI search (GEO), JSON-LDSEO, GEO, 검색 최적화, 메타태그, JSON-LD
E2E Verifier/e2e-verifierBrowser verification based on Chrome MCP / Playwright MCP / Agent-BrowserE2E 검증, 브라우저 검증, 화면 테스트, 동작 확인
Lighthouse Performance/lighthouse-performanceMeasures and improves Core Web Vitals (LCP·CLS·INP·TTFB·FCP)성능 검사, lighthouse, Core Web Vitals, 로딩 속도
QA Inspector/qa-inspectorDetects interface mismatches by cross-comparing API↔hook types, routing, and state transitionsQA, 정합성 검사, 경계면 검증, 버그 찾기
Security Audit/security-auditOWASP Top 10 code analysis, npm audit, security header and secret detection보안 감사, OWASP, XSS, 취약점 스캔, CVE

Commands (7)

CommandDescription
/orchestratorSequentially runs the 6 stages research → prd → develop → review → verify, then produces an integrated report. Finishes static analysis first and runs the expensive E2E last
/researchAnalyzes requirements and derives specifications using the grill-me sub-agent
/prdCreates a PRD via the planner → architecture → critic loop
/frontend-guidelinesGuideline-based development (Develop) via parallel spawning of a11y · semantic-html · seo-geo · tdd
/reviewRuns only user-selected items among the 5 perspectives /simplify · /review · security-audit · lighthouse · qa-inspector in parallel → fix and re-review loop (up to 3 times)
/verifierE2E browser verification based on PRD acceptance criteria via the e2e-verifier skill
/verifyIntegrates E2E browser verification + type/build checks (tsc --noEmit, npm run build)

Hooks (1)

HookTriggerDescription
stop-lintStopAutomatically runs eslint --fix → stylelint --fix → prettier --write on git-changed files when a response completes (monorepo support, skips uninstalled tools)

harness-generator plugin (1 skill)

SkillCommandDescriptionTrigger keywords
Harness Generator/harness-generatorGenerates a domain-specific harness as a single bundle through a 7-step meta process (audit → domain analysis → architecture → agent definition → skill authoring → orchestration → verification/evolution)하네스 만들어줘, 오케스트레이터 작성, 워크플로우 자동화, 에이전트 팀 구성

git-harness plugin (2 skills)

SkillCommandDescriptionTrigger keywords
Commit/commitWrites Korean commit messages in the issue-number type: subject format (imperative subject; summary/impact/test body; secrets and protected-branch checks)커밋, commit, 커밋 메시지
Review to PR/review-to-prAll-in-one review → commit → PR creation. Automatically runs and applies /simplify + /review at each stagePR 생성, PR 리뷰, 리뷰 후 커밋

meta-harness plugin (4 skills + 4 agents)

SkillCommandDescription
Meta-Harness/meta-harnessEntry orchestrator. R1 current-session redirect/reinforcement · R2 plugin improvement · R3 external .md back-tracing trigger. Phase 0~8 + R4 reporting, with a user approval gate (no automatic application)
Session Signal Capture/session-signal-captureCaptures R1 utterances, the immediately preceding artifact, the active skill, and R3 external .md as a raw trace without summarization. R3 uses a 3-tier source fallback for back-tracing
Causal Diagnosis/causal-diagnosisDirectly queries the experience-store raw trace via grep/cat to isolate confounds → single-factor verification → why-first root cause diagnosis
Pareto Refinement/pareto-refinementGenerates 4-axis (behavior-alignment · rule-body-cost · trigger-precision · generalization) Pareto non-regression patches via additive-first → compose → transfer

Four agents (all model: opus): trace-capturer (raw trace normalization) · failure-diagnostician (parallel root cause diagnosis per defect) · pareto-refiner (patch generation) · experience-historian (experience-store curation).


Using the GitHub Repository

Repository: https://github.com/seungahhong/seungah-claude-plugins

How users use it

  1. Register the marketplace — In Claude Code, run /plugin marketplace add https://github.com/seungahhong/seungah-claude-plugins (or the CLI claude plugin add seungahhong/seungah-claude-plugins)
  2. Install plugins — Install only the plugins you want, such as /plugin install frontend-harness@seungah-claude-plugins (choose user / project / local scope)
  3. Run skills — After installation, invoke them via slash commands like /planner, /review, /commit, or trigger them with natural language. If the list does not appear, retry after running /reload-plugins
  4. Update — When changes are reflected in the repository, update to the latest with /plugin marketplace update seungah-claude-plugins

Contributing and local development

git clone https://github.com/seungahhong/seungah-claude-plugins.git
cd seungah-claude-plugins
# Load directly from a local path for debugging
claude --plugin-dir ./plugins/frontend-harness

For components (skills / commands / agents / hooks), you only need to add files to each plugin directory; there is no need to modify marketplace.json (paths are declared in plugin.json).


Usage Examples

After installation is complete, you can trigger skills with natural language or slash commands in the Claude Code chat window.

Skill-by-skill execution flow

/grill-me            // 1. 요구사항 명확화
/planner             // 2. 구현 계획(PRD)
/architecture        // 3. 기술 아키텍처 설계
/critic              // 4. 설계 검증·리스크 분석
/tdd                 // 5. 테스트 주도 구현
/a11y /semantic-html /seo-geo-optimizer   // 6. 품질·접근성·SEO 점검
/review              // 7. 통합 리뷰(5개 관점 중 선택 병렬 + 재리뷰 루프)
/verify              // 8. 동작·빌드 검증(E2E + tsc + build)
/commit              // 9. 커밋 메시지 작성(git-harness)

Full automation with the orchestrator

/orchestrator        // research → prd → develop → review → verify 전체 파이프라인 자동 실행

Meta workflow — handling the harness itself

/harness-generator   // 신규 도메인용 하네스(에이전트+스킬+오케스트레이터) 한 묶음 생성
/meta-harness        // 사용 중인 하네스를 full-trace로 캡처·진단·개선(사용자 승인 게이트)

Natural language trigger examples

  • "Check this component's web accessibility" → /a11y
  • "Run a security audit and performance check before deployment" → /review (selecting security-audit + lighthouse)
  • "Commit these changes in Korean" → /commit
  • "Develop the payment feature from start to finish" → /orchestrator