LLM & Agents

Generate five brand-aligned design concepts for validated niches using Claude creativity constrained by deterministic structure and brand voice memories.

Comprehensive Type 2 executable skill for automated WCS to Godot migration orchestration. Implements Plan-Implement-Validate lifecycle, agent lifecycle management, task state machines, and event-driven hooks with structured JSON I/O for reliable automated remediation loops. Use for coordinating complex migration workflows with deterministic execution.

Validates command frontmatter, delegation patterns, simplicity guidelines, and documentation proportionality. Use when reviewing, auditing, analyzing, evaluating, improving, or fixing commands, validating official frontmatter (description, argument-hint, allowed-tools, model), checking delegation clarity or standalone prompts, assessing simplicity guidelines (6-15 simple, 25-80 documented), validating argument handling, or assessing documentation appropriateness. Distinguishes official Anthropic requirements from custom best practices. Also triggers when user asks about command best practices, whether a command should be a skill instead, or needs help with command structure.

Use when integrating LLMs (OpenAI, Qwen, Claude), extracting structured data from text, building prompts, parsing AI responses, handling JSON output, or implementing multi-step AI workflows

Python stack configuration - uv, ruff, mypy, pytest with 96% coverage threshold

AI agent workflow patterns including ReAct agents, multi-agent systems, loop control, tool orchestration, and autonomous agent architectures. Use when building AI agents, implementing workflows, creating autonomous systems, or when user mentions agents, workflows, ReAct, multi-step reasoning, loop control, agent orchestration, or autonomous AI.

Conduct a quarterly review of your overall research mission and direction. This is a user-level review stored in ~/.researchAssistant/. Use when the user types /quarterly_review, every 3 months, after major project milestones, or when questioning research direction.

Manage beads correctly including claiming, closing, announcements, and dependencies. Use when starting work on a task, when finishing a task, when the user mentions beads or tasks, or when coordinating with other agents on task ownership.

Search for items and properties on Wikidata and retrieve external identifiers. Use when an agent needs to (1) search for Wikidata items by label or alias, (2) get entity details including labels, descriptions, aliases, (3) retrieve external identifiers (authority control IDs) for an entity, (4) look up properties or claims for items. Triggers on queries mentioning Wikidata, Q-IDs, P-IDs, authority control, external identifiers, or structured knowledge base lookups.

Guide for creating custom slash commands for Claude Code

World-class ML engineering skill for productionizing ML models, MLOps, and building scalable ML systems. Expertise in PyTorch, TensorFlow, model deployment, feature stores, model monitoring, and ML infrastructure. Includes LLM integration, fine-tuning, RAG systems, and agentic AI. Use when deploying ML models, building ML platforms, implementing MLOps, or integrating LLMs into production systems.

Creates new Agent Skills for AI Agents following best practices and documentation. Use when the user wants prompts 'create a new skill ...' or 'use your meta skill to ...'.

Expert on Shelby Protocol network infrastructure, RPC servers, storage providers, Cavalier implementation, tile architecture, performance optimization, connection management, and DoubleZero private network. Triggers on keywords Shelby RPC, storage provider, Cavalier, tile architecture, private network, DoubleZero, network performance, RPC endpoint, request hedging, connection pooling.

Build Retrieval-Augmented Generation (RAG) systems for LLM applications with vector databases and semantic search. Use when implementing knowledge-grounded AI, building document Q&A systems, or integrating LLMs with external knowledge bases.

Claude acts as manager/architect while delegating all coding to external LLM CLIs (Gemini, Codex, Qwen). Claude never writes code - only plans, delegates, and verifies. Use when user says "manage", "architect mode", "delegate to", or wants Claude to drive another LLM.

Propose structured interpretations of underspecified requests. Use when a user's request is vague or incomplete but reasonable assumptions can be made. Rapidly prototypes intent by restating, structuring, and committing to a next action.

Backtest results interpreter and strategy evaluator. Analyzes historical backtest performance, identifies strengths/weaknesses, and provides actionable recommendations for strategy improvement.

Semantic code search using vector embeddings. Find functions/classes by meaning, not exact text. Use when exploring codebases or when grep doesn't find what you need.

AI-powered PR Author Agent that transforms Observability Diff Plans into Pull Requests. Use when: (1) Generating instrumentation code from Scout Agent output, (2) Creating OTel configuration, correlation headers, lineage specs, (3) Scaffolding telemetry validation tests, (4) Creating GitHub/GitLab PRs with observability artifacts. Triggers: "generate PR from diff plan", "create instrumentation PR", "scaffold observability code", "generate OTel config".

プロジェクトの .claude/commands/ に新しいスラッシュコマンドを作成する。「コマンド作成」「新しいコマンド」「コマンドを作って」「コマンド追加」「command 作成」「コマンドを追加したい」「新規コマンド」などで起動。プロジェクト固有のコマンドファイルを生成。