LLM & Agents

AI security co-pilot for identifying, testing, and fixing vulnerabilities in LLM-powered applications.Use when: (1) Securing LLM applications or agents, (2) Generating security test suites with promptfoo,(3) Testing for prompt injection, jailbreaking, data exfiltration, (4) Hardening system prompts,(5) Compliance mapping for OWASP LLM Top 10, NIST AI RMF, CJIS, SOC2, (6) Threat modeling AI systems,(7) Analyzing security eval results, (8) Research on LLM attack/defense techniques.Triggers: "secure my LLM", "prompt injection", "jailbreak test", "AI security", "red team","system prompt hardening", "LLM vulnerability", "promptfoo", "OWASP LLM", "AI compliance".

Build and integrate production-ready RAG (Retrieval-Augmented Generation) chatbots into documentation sites using OpenAI, Qdrant Cloud, and Neon Postgres. Handles complete stack from backend API to frontend UI integration.

Expert infrastructure design including networking, compute, storage, and operations

Comprehensive templates, patterns, and best practices for creating Claude Code subagents and skills. Use when building new agents/skills or need reference examples for proper structure and formatting.

プロジェクトの .claude/settings.json を更新する。「設定を更新して」「settings を変更」「permissions を追加」「設定を変えて」「Claude の設定」「設定ファイルを編集」「allowedTools を追加」などで起動。Claude Code のプロジェクト設定を管理。

This skill should be used when the user asks about "prompt_toolkit Validator", "validation", "input validation", "ValidationError", "History", "FileHistory", "InMemoryHistory", "auto_suggest", "AutoSuggest", or needs to validate user input or implement command history in prompt_toolkit applications.

Use when decomposing complex work. Dispatch fresh subagent per task, review between tasks. Flow: Load plan → Dispatch task → Review output → Apply feedback → Mark complete → Next task. No skipping reviews, no parallel dispatch.

Systematic framework for selecting the optimal specialized agent for any task. Use when delegating to subagents via the Task tool to ensure the most appropriate specialist is chosen based on framework, domain, task type, and complexity. Applies decision tree logic to match tasks with agent expertise.

Comprehensive prompt engineering framework for designing, optimizing, and iterating LLM prompts. This skill should be used when users request prompt creation, optimization, or improvement for any LLM task, or when users need help translating vague requirements into effective prompts through collaborative dialogue and iterative refinement.

Author high-quality assessment items (questions, prompts, tasks) aligned to learning objectives with answer keys and rubrics. Use when creating test questions, writing assessment items, or building item banks. Activates on "write assessment items", "create test questions", "develop quiz", or "author performance tasks".

여러 LLM(OpenAI, Gemini, Ollama 등)을 통합하여 멀티 에이전트 협업을 수행합니다. 역할 분담, 토론/합의, 체인 파이프라인, 병렬 처리 등 다양한 협업 패턴을 지원하며, 사용 시점에 시나리오를 동적으로 구성할 수 있습니다. 복잡한 작업을 여러 LLM에게 분산하여 더 나은 결과를 얻고 싶을 때 사용하세요.

Patterns for ingesting knowledge into vector databases and RAG systems

Structured incremental development workflow that enforces explicit scoping before implementation. Use when: (1) User says 'build', 'implement', 'develop', 'create feature', or 'add functionality', (2) Task involves multiple files or could have ripple effects, (3) User wants to avoid scope creep or premature implementation, (4) Breaking down complex tasks into verifiable steps. This skill prevents AI agents from jumping into code without a validated plan.

Debug TraitorSim agent memory systems including profile.md, diary entries, trust matrices (suspects.csv), and SKILL.md files. Use when troubleshooting agent behaviors, inspecting memory contents, validating memory updates, or when asked about agent memory, profile debugging, or trust matrix issues.

Master orchestration patterns, multi-agent coordination, and effective workflow composition using the Agency plugin's 51+ specialized agents. Activate when planning complex implementations, coordinating multiple agents, or optimizing development workflows.

Integrates SQLite embedded database with Node.js using better-sqlite3 for synchronous operations or the native Node.js SQLite module. Use when building applications with local storage, embedded databases, or when user mentions SQLite, better-sqlite3, or embedded SQL.

Core prompt engineering and context engineering best practices for Claude Code prompts.

Execute Gray Swan AI Arena indirect prompt injection (IPI) and machine-in-the-middle (MITM) challenges with optimized payloads, reconnaissance workflows, and evidence collection for CTF competition success

Build Retrieval-Augmented Generation systems with vector databases

DeepSeek AI large language model API via curl. Use this skill for chat completions, reasoning, and code generation with OpenAI-compatible endpoints.