/agents

/agents

Find and install specialist modes for your project.

user-invocable: true description: Without arguments, scans your project and recommends specialist modes. With a search query, searches for specific modes. Usage: /agents [search query] allowed-tools: [Read, Glob, Grep, Bash, WebFetch, WebSearch, Write, Skill, AskUserQuestion] argument-description: Optional search query (e.g., "react", "devops", "testing"). If omitted, scans the project automatically.

Instructions

Check if the user provided an argument.

• No argument → go to Mode A: Project Scan
• Has argument → go to Mode B: Keyword Search

---

Mode A: Project Scan (no argument)

Step 1: Gather Context (MANDATORY - DO NOT SKIP)

You MUST perform ALL of the following in parallel. Do NOT assume the project is empty without checking.

1a. Read project files (MUST attempt all): Run Glob for these patterns and Read any that exist:

• package.json — extract dependencies and devDependencies
• tsconfig.json or jsconfig.json
• pyproject.toml or requirements.txt or setup.py
• go.mod
• Cargo.toml
• Gemfile
• docker-compose.yml or Dockerfile
• .github/workflows/*.yml or .gitlab-ci.yml
• terraform/ or *.tf
• pubspec.yaml (Flutter)
• Podfile or *.xcodeproj (iOS)

1b. Detect file structure: Run ls on project root and src/ (if exists) to identify framework patterns:

• .jsx/.tsx files → React
• .vue files → Vue
• .svelte files → Svelte
• .py files → Python
• .go files → Go
• .rs files → Rust
• .swift files → Swift
• __tests__/ or *.test.* → Testing

1c. Build tech profile: Combine ALL detected technologies into a single list.

Step 2: Check Installed Agents

Glob: .claude/agents/*.md
Glob: ~/.claude/agents/*.md

Read each found file to determine what it covers.

Step 3: Fetch & Match

Fetch the agency-agents repository:

WebFetch: https://raw.githubusercontent.com/msitarzewski/agency-agents/main/README.md
WebFetch: https://api.github.com/repos/msitarzewski/agency-agents/git/trees/main?recursive=1

Score each agent against the project's tech profile:

Tech detected in project	Agent id	Score boost
React, Vue, Angular, Svelte, Next.js, Vite, Webpack	engineering-frontend-developer	+3 per match
TypeScript, CSS, Tailwind CSS	engineering-frontend-developer	+2 per match
Express, Django, FastAPI, Go, Node.js, PostgreSQL, Redis	engineering-backend-architect	+3 per match
React Native, Flutter, Swift, SwiftUI, Kotlin	engineering-mobile-app-builder	+3 per match
TensorFlow, PyTorch, LangChain, OpenAI, Anthropic, Hugging Face	engineering-ai-engineer	+3 per match
Docker, Kubernetes, Terraform, GitHub Actions, Jenkins	engineering-devops-automator	+3 per match
OAuth, OWASP, security headers, auth libraries	engineering-security-engineer	+3 per match
Jest, Pytest, Playwright, test files	testing-evidence-collector	+2 per match
API test files, REST	testing-api-tester	+2 per match
Few files, no package manager	engineering-rapid-prototyper	+3
CSS frameworks, UI libraries, design tokens	design-ui-designer	+2 per match
Accessibility configs, ARIA	testing-accessibility-auditor	+2 per match
Laravel, PHP, Livewire	engineering-senior-developer	+3 per match

For agents not in this table, try keyword matching against the README descriptions.

Exclude already-installed agents and 0-score agents. Sort by score, take top 3-5.

Then go to Step: Recommend.

---

Mode B: Keyword Search (with argument)

Step 1: Check Installed Agents

Glob: .claude/agents/*.md
Glob: ~/.claude/agents/*.md
Grep: search for query keywords in matched files

Step 2: Search GitHub

WebFetch: https://raw.githubusercontent.com/msitarzewski/agency-agents/main/README.md
WebFetch: https://api.github.com/repos/msitarzewski/agency-agents/git/trees/main?recursive=1

Match the query against:

• Agent filenames (e.g., query "react" matches engineering-frontend-developer)
• Division names (e.g., query "testing" matches all agents in testing/)
• README descriptions

If fewer than 3 results, fallback:

WebSearch: site:github.com "claude agent" {query} filetype:md

Then go to Step: Recommend.

---

Step: Recommend (shared)

Present the results:

감지된 기술: React 19, Next.js, Tailwind CSS, Jest (Mode A only) 이미 설치된 전문가 모드: (없음)

추천 전문가 모드:

#	Name	Division	Score	Description
1	Frontend Developer	engineering	9	React, Vue, Angular 전문 개발
2	Evidence Collector	testing	4	스크린샷 기반 QA 테스팅
3	UI Designer	design	3	디자인 시스템, 컴포넌트 설계

설치할 번호를 선택하세요 (여러 개: 1,2,3 / 건너뛰기: skip)

Mode A only — always append the category guide after the recommendation table:

다른 분야의 전문가 모드도 검색할 수 있습니다: /agents marketing · /agents product · /agents design · /agents project · /agents support · /agents data · /agents testing · /agents spatial

This MUST always be shown in Mode A, even when there are scored recommendations. It helps users discover the 49+ non-code specialist modes that can't be detected by project scanning.

For Mode B, omit the "감지된 기술" line and replace "Score" with "Relevance". Do NOT append the category guide.

If no results:

"검색 결과가 없습니다. 다른 키워드로 시도해 보세요."

If all technologies already covered:

필요한 전문가 모드가 모두 설치되어 있습니다.

다른 분야의 전문가 모드도 검색할 수 있습니다: /agents marketing · /agents product · /agents design · /agents project · /agents support · /agents data · /agents testing · /agents spatial

Step: Install (shared)

When the user selects:

1. Download from GitHub:

   WebFetch: https://raw.githubusercontent.com/msitarzewski/agency-agents/main/{division}/{agent-id}.md
   

2. Ask: project local (.claude/agents/) or global (~/.claude/agents/)?
3. Save to chosen location
4. Add @agents/{agent-id}.md include to CLAUDE.md
5. Verify installation

If user says "skip", "건너뛰기", or "pass", skip.

After install, briefly summarize what was installed and how it helps.

Language Rules

• Respond in the same language the user used
• NEVER use "에이전트", "agent", "persona" — use "전문가 모드", "specialist mode" instead
• Maximum 3 specialist modes recommended per project