Activate at session start when you are the primary Claude instance. Establishes orchestrator role with delegation protocols, subagent launch templates, token efficiency rules, and parallel safety constraints. You orchestrate; subagents implement.
Guide for implementing Google Gemini API image understanding - analyze images with captioning, classification, visual QA, object detection, segmentation, and multi-image comparison. Use when analyzing images, answering visual questions, detecting objects, or processing documents with vision.
Secure coding with Claude Code - vulnerability prevention, secrets, security review
创建有效技能的指南。当用户想要创建新技能（或更新现有技能）来扩展Claude在专门知识、工作流程或工具集成方面的能力时，应该使用此技能。
SEMO 패키지 구조 및 Sub-Agent 최적화 규칙 검증. Use when (1) Agent/Skill 추가/수정/삭제 후 검증, (2) CLAUDE.md 업데이트 후 일관성 확인, (3) 버저닝 전 품질 게이트, (4) Sub-Agent 최적화 규칙 준수 감사.
Handoff protocol for Rust multi-agent development system. Use when working as rust-architect, rust-developer, rust-testing-engineer, rust-performance-engineer, rust-security-maintenance, rust-code-reviewer, rust-cicd-devops, or rust-debugger. ALWAYS read on agent startup.
Explains how to use Synapse RuntimeContext API. Use when the user asks about "RuntimeContext", "ctx.", "logging", "progress tracking", "set_progress", "set_metrics", "log_message", "BaseStepContext", "TrainContext", "ExportContext", "UploadContext", "InferenceContext", "AddTaskDataContext", or needs help with synapse plugin context and logging.
Load, activate, and optionally install an AI skill from a .skill file. Use this skill when the user wants to load, activate, or use a skill file. The user can invoke this skill with slash command `/load-skill [ --install | -i ] SKILLFILE`, where SKILLFILE is the path to a .skill file.
Use PROACTIVELY when developing Claude Code hooks to implement content-based deduplication and prevent duplicate insight storage across sessions
Test AI systems for code injection vulnerabilities including prompt-to-code attacks and agent exploitation
基于 RAG 的知识库检索技能。从本地文档中检索相关信息，支持语义搜索和精确匹配。适用于问答、信息查询、文档理解等场景。
Design multi-agent architectures for complex tasks. Use when single-agent context limits are exceeded, when tasks decompose naturally into subtasks, or when specializing agents improves quality.
Methodology for identifying LLM-suited tasks, designing pipelines, and iterating with agent-assisted development.
Review and propose improvements for Claude skills based on usage patterns.
Use when user explicitly requests to coordinate with other Claude Code agents, join an agent chat, or communicate across multiple repositories/projects
开始任何对话时使用 - 确立如何查找和使用技能。在任何响应之前（包括澄清问题）都需要调用 Skill 工具。Use when starting any conversation - establishes how to find and use skills. If a skill might apply (even 1% chance), you MUST invoke it.
Guide for creating Agent Skills: structure, best practices, and SKILL.md format for Claude Code, Codex, Gemini CLI, and other AI agents.
Understand where FAF works - Claude Code (CLI + Skills), claude-faf-mcp (Desktop + MCP), vs claude.ai (web - file upload only). Explains faf-cli vs MCP differences, when to use each platform. Use when user asks "does FAF work with", "CLI vs MCP", "Claude Desktop vs claude.ai", or platform compatibility questions.
Techniques to test and bypass AI safety filters, content moderation systems, and guardrails for security assessment
Clean ~/.claude.json when it becomes too large (>25K tokens). Use when Claude cannot read the file, user mentions "claude.json trop gros", "nettoyer claude.json", or "cleanup claude.json".
Advanced prompt manipulation including direct attacks, indirect injection, and multi-turn exploitation
ユーザープロンプトを分析し、topic_type（instruction/question/context）を判定する。
