持續整合/部署

Applies the Diataxis framework to create or improve technical documentation. Use when being asked to write high quality tutorials, how-to guides, reference docs, or explanations, when reviewing documentation quality, or when deciding what type of documentation to create. Helps identify documentation types using the action/cognition and acquisition/application dimensions.

Use when writing or changing tests, adding mocks, or tempted to add test-only methods to production code - prevents testing mock behaviour, production pollution with test-only methods, and mocking without understanding dependencies

Guide for creating effective Claude Skills. This skill should be used when users want to create (or update) a skill that extends Claude's capabilities with specialised knowledge, workflows, or tool integrations.

Use when working with AWS Strands Agents SDK or Amazon Bedrock AgentCore platform for building AI agents. Provides architecture guidance, implementation patterns, deployment strategies, observability, quality evaluations, multi-agent orchestration, and MCP server integration.

Practical bash scripting guidance emphasising defensive programming, ShellCheck compliance, and simplicity. Use when writing shell scripts that need to be reliable and maintainable.

Creating and maintaining CLAUDE.md project memory files that provide non-obvious codebase context. Use when (1) creating a new CLAUDE.md for a project, (2) adding architectural patterns or design decisions to existing CLAUDE.md, (3) capturing project-specific conventions that aren't obvious from code inspection.

Load before running any glab commands to ensure correct CLI syntax. Use when creating/viewing MRs, checking pipelines, managing issues, or any GitLab operations (when remote contains "gitlab").

Use narsil-mcp code intelligence tools effectively. Use when searching code, finding symbols, analyzing call graphs, scanning for security vulnerabilities, exploring dependencies, or performing static analysis on indexed repositories.

Specialized in reverse-engineering compiled binaries (JARs, DLLs). Use this when the user asks to compare versions, find security fixes, or analyze binary patches.

Use when designing or architecting Elixir/Phoenix applications, creating comprehensive project documentation, planning OTP supervision trees, defining domain models with Ash Framework, structuring multi-app projects with path-based dependencies, or preparing handoff documentation for Director/Implementor AI collaboration

Provides guidance on software architecture patterns and design decisions. Use when designing systems, choosing patterns, structuring projects, or when asked about architectural approaches.

Post-process LaTeX project prose to improve readability through varied sentence and paragraph lengths. Removes filler phrases and unnecessary transitions while preserving all citations and semantic meaning.

Write LaTeX ML/AI review articles for arXiv using the IEEEtran template and verified BibTeX citations.

Language-independent tokenizer treating text as raw Unicode. Supports BPE and Unigram algorithms. Fast (50k sentences/sec), lightweight (6MB memory), deterministic vocabulary. Used by T5, ALBERT, XLNet, mBART. Train on raw text without pre-tokenization. Use when you need multilingual support, CJK languages, or reproducible tokenization.

Data framework for building LLM applications with RAG. Specializes in document ingestion (300+ connectors), indexing, and querying. Features vector indices, query engines, agents, and multi-modal support. Use for document Q&A, chatbots, knowledge retrieval, or building RAG pipelines. Best for data-centric LLM applications.

Scalable data processing for ML workloads. Streaming execution across CPU/GPU, supports Parquet/CSV/JSON/images. Integrates with Ray Train, PyTorch, TensorFlow. Scales from single machine to 100s of nodes. Use for batch inference, data preprocessing, multi-modal data loading, or distributed ETL pipelines.

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

Expert guidance for Fully Sharded Data Parallel training with PyTorch FSDP - parameter sharding, mixed precision, CPU offloading, FSDP2

Framework for building LLM-powered applications with agents, chains, and RAG. Supports multiple providers (OpenAI, Anthropic, Google), 500+ integrations, ReAct agents, tool calling, memory management, and vector store retrieval. Use for building chatbots, question-answering systems, autonomous agents, or RAG applications. Best for rapid prototyping and production deployments.

Meta's 7-8B specialized moderation model for LLM input/output filtering. 6 safety categories - violence/hate, sexual content, weapons, substances, self-harm, criminal planning. 94-95% accuracy. Deploy with vLLM, HuggingFace, Sagemaker. Integrates with NeMo Guardrails.