Skill: auto-agent-discovery

Skill: auto-agent-discovery

Automatically recommend relevant specialist modes when the user works with specific frameworks or tools.

user-invocable: false description: When the user is working with a specific framework, language, or tool and no specialist mode is installed for it, this skill activates to suggest relevant expert modes. Fetches available modes from GitHub in real-time. allowed-tools: [Read, Glob, Grep, Skill, WebFetch, WebSearch]

Instructions

Activation

This skill activates when the user's request involves a specific technology, framework, or role pattern. Examples:

• Building a React application → frontend-developer
• Setting up CI/CD pipelines → devops-automator
• Writing API tests → api-tester
• Designing a UI → ui-designer
• Mobile app development → mobile-app-builder

Step 1: Detect Technology Context

From the user's message, identify:

• Programming languages (Python, TypeScript, Rust, Go, etc.)
• Frameworks (React, Next.js, Django, FastAPI, etc.)
• Tools (Docker, Kubernetes, Terraform, Ansible, etc.)
• Platforms (AWS, GCP, Azure, Vercel, etc.)
• Roles/tasks (security audit, UX research, performance testing, etc.)

Step 2: Check Installed Agents

Search for already-installed specialist modes matching the detected context:

Glob: .claude/agents/*.md
Glob: ~/.claude/agents/*.md
Grep: search for the technology name in matched files

If a relevant specialist mode is already installed, do NOT recommend anything — let the existing mode handle the task.

Step 3: Search GitHub Repository

Fetch the agency-agents repository to find matching specialist modes:

WebFetch: https://raw.githubusercontent.com/msitarzewski/agency-agents/main/README.md

Match the detected technology against agent names, descriptions, and divisions from the README.

Known mappings (use these for fast matching before fetching):

• React, Vue, Angular, Svelte, CSS, Tailwind → engineering/engineering-frontend-developer.md
• Express, Django, FastAPI, Go, Node.js → engineering/engineering-backend-architect.md
• React Native, Flutter, Swift, Kotlin → engineering/engineering-mobile-app-builder.md
• TensorFlow, PyTorch, LangChain, OpenAI → engineering/engineering-ai-engineer.md
• Docker, Kubernetes, Terraform, CI/CD → engineering/engineering-devops-automator.md
• Small/new project, MVP → engineering/engineering-rapid-prototyper.md
• Security, OWASP, auth → engineering/engineering-security-engineer.md
• UI design, design system → design/design-ui-designer.md
• Testing, QA, Jest, Pytest → testing/testing-evidence-collector.md
• API testing → testing/testing-api-tester.md
• Accessibility, WCAG → testing/testing-accessibility-auditor.md
• Performance, load testing → testing/testing-performance-benchmarker.md

If no known mapping matches, fetch the README and search for keyword matches in agent names and descriptions.

Step 4: Natural Language Recommendation

CRITICAL UX RULES:

• NEVER use the words "에이전트", "agent", "persona" when talking to the user
• Instead use: "전문가 모드", "전문 지원", "specialist mode", "expert support"
• Frame recommendations naturally, as if suggesting expertise

Example recommendation format:

참고로, Frontend 전문가 모드를 사용할 수 있습니다. React, Vue, Angular 등 프론트엔드 개발에 특화된 전문 지원을 받을 수 있어요. 설치하시겠습니까?

Or in English:

By the way, I found a Frontend specialist mode that can help with React, Vue, Angular development — including component architecture, performance optimization, and accessibility. Would you like me to set it up?

Step 5: Handle User Response

• If the user accepts: invoke the agent-installer skill with the selected agent info
• If the user declines: proceed with the task using general knowledge, do not ask again in this session
• If the user ignores: proceed normally, do not repeat the suggestion

Constraints

• Maximum 1 recommendation per conversation turn
• Do not recommend if the user is doing something trivial (e.g., a simple file edit)
• Do not recommend for technologies the user clearly already knows well
• Maximum 3 specialist modes per project recommended (to avoid context overload)