Agent Skills

Design repeatable agent evaluation loops on Google Cloud, from datasets and metrics to regression analysis and monitored improvement.

Checks

Metadata only

Evidence

1 source

Design agent memory: episodic memory, semantic memory, working memory, memory consolidation, Mem0, and persistent context strategies.

Checks

Metadata only

Evidence

1 source

Portable SKILL.md format and validation guidance for agent capabilities.

Checks

Metadata only

Evidence

1 source

Design robust agent tool use: tool schemas, parallel tool calls, error recovery, tool result injection, and multi-step agent loops.

Checks

Metadata only

Evidence

1 source

Design agentic workflows: planner-executor loops, reflection/self-critique, tool routing, and ReAct/Plan-and-Execute patterns.

Checks

Metadata only

Evidence

1 source

Apply AI safety: constitutional AI, RLHF, red-teaming, guardrails, jailbreak detection, harm scoring, and model cards.

Checks

Metadata only

Evidence

1 source

Expert guidance for integrating the Claude API including messages, streaming, tool use, vision, and prompt caching.

Checks

Metadata only

Evidence

1 source

Reusable instructions for producing polished frontend experiences.

Checks

Metadata only

Evidence

1 source

Generate Model Context Protocol servers from a description — scaffolds code, tools, and handlers for a working MCP implementation.

Checks

Metadata only

Evidence

1 source

Guidance for designing and packaging reusable agent skills.

Checks

Metadata only

Evidence

1 source

Build interactive web components and rich artifacts (charts, forms, games, demos) that render inside Claude or a browser.

Checks

Metadata only

Evidence

1 source

Orchestrate AI crews: role-based agents, task delegation, hierarchical process, memory systems, and tool integration in CrewAI.

Checks

Metadata only

Evidence

1 source

Author, organize, and maintain .cursor/rules/ files for consistent agent behavior across a codebase.

Checks

Metadata only

Evidence

1 source

Program LLM pipelines declaratively with DSPy: signatures, modules, optimizers (MIPRO/BootstrapFewShot), and automatic prompt tuning.

Checks

Metadata only

Evidence

1 source

Optimize vector embeddings: model selection, fine-tuning, matryoshka embeddings, quantization, HNSW indexing, and hybrid search.

Checks

Metadata only

Evidence

1 source

Combine knowledge graphs with RAG: entity extraction, graph traversal retrieval, GraphRAG, and hybrid graph+vector search.

Checks

Metadata only

Evidence

1 source

Design LangChain chains: LCEL composition, retrieval chains, extraction chains, structured output, LangGraph agents, and streaming.

Checks

Metadata only

Evidence

1 source

Build stateful agent graphs with LangGraph: state channels, conditional edges, tool nodes, human-in-the-loop, and persistence.

Checks

Metadata only

Evidence

1 source

Build LlamaIndex pipelines: ingestion pipelines, vector stores, retrieval modes, routing, sub-question engines, and agentic RAG.

Checks

Metadata only

Evidence

1 source

Reduce LLM inference costs: model routing by task complexity, prompt caching, batching, and token usage monitoring.

Checks

Metadata only

Evidence

1 source

Design LLM evaluation frameworks: define metrics, build test sets, run evals with LLM-as-judge, and track regressions.

Checks

Metadata only

Evidence

1 source

Fine-tune LLMs with LoRA/QLoRA, full fine-tuning, PEFT, DPO, GRPO, dataset curation, and evaluation with lm-eval-harness.

Checks

Metadata only

Evidence

1 source

Reduce LLM costs and latency with prompt caching: cache breakpoints, prefix caching, cache-aware prompt design, and TTL management.

Checks

Metadata only

Evidence

1 source

Extract structured data with LLMs: JSON mode, Instructor library, function calling, Pydantic validation, and confidence scoring.

Checks

Metadata only

Evidence

1 source