Patterns

Recurring architectural and design patterns in AI/LLM systems — RAG, tool calling, agentic workflows, routing, guardrails, and security trust boundaries. Updated for early January 2027 SOTA standards (incorporating Claude 5.1, FastMCP 3.1, GPT-5.5/5.6, Gemini 4.0 Pro, Llama 4, and Pydantic v2 validation).

Contents

• Agent Skills Best Practices — Skill authoring, trigger design, permission model, and validation checklist
• Agentic Workflows Pattern — Autonomous multi-step loops, tool execution, state management, and human-in-the-loop control
• Claude Tool Search Pattern — Dynamic tool discovery and deferred parameter resolution for massive agentic tool sets
• Data Copilot Agentic RAG — Hybrid retrieval pattern for diagnostic analytics and SQL schema navigation
• Data Copilot MCP Tooling — Standardizing data access for Text-to-SQL pipelines using Model Context Protocol (MCP 3.1)
• Date Extraction Pattern — Parsing and normalizing relative temporal references into ISO 8601 with strict Pydantic v2 schemas
• Extraction and Classification Pattern — Schema-first structured data extraction and multi-label classification
• Fallback Patterns — Architecting resilience via model failover, multi-provider redundancy, and fail-safe routing
• Filesystem-as-Interface Pattern — Filesystem as the primary interface and persistence layer for agents
• Fine-tuning Open Models — LoRA/QLoRA, Unsloth, axolotl, MLX, dataset prep, and GGUF export for local runtimes
• LLM Trust Boundaries Pattern — Zero-trust agent architecture, boundary isolation, and prompt injection defense
• n8n Error Handling Pattern — Enterprise error trapping, dead-letter queues, and automated incident recovery in workflow automation
• OpenClaw Security Operations Pattern — Operational security, identity federation, and audit logging for autonomous claw deployments
• OpenClaw Use Case Catalog — Reference architecture catalog for enterprise OpenClaw agent deployment
• OpenClaw Workflow Prompt Library Pattern — Standardized meta-prompts and execution contracts for autonomous agent pipelines
• Prompt Requests Pattern — Structured request formatting, template versioning, and prompt optimization
• RAG Pattern Architecture — Advanced hybrid retrieval-augmented generation with vector and structural indexing
• Retrieval-Augmented Generation (RAG) — Grounding LLM output with retrieved context, hybrid search, and semantic re-ranking
• Search Patterns — Multi-modal search, hybrid dense/sparse retrieval, and iterative query refinement
• Software Factories Pattern — Non-interactive development via seed, validation, and automated feedback loops
• Tool Calling & Model Context Protocol (MCP) — Universal standard (FastMCP 3.1) for connecting LLMs to external tools and data

Common Patterns

• RAG (Retrieval-Augmented Generation) — Grounding LLM output with retrieved context via vector and tree-based indices
• Agentic Workflows & Multi-Agent Collaboration — Autonomous iterative execution loops with state management and dynamic handoffs
• Tool Calling & MCP 3.1 — LLMs invoking external tools via FastMCP 3.1 structured schemas and type-safe contracts
• Fine-tuning & Open Models — Adapting open models (Llama 4, Gemma 3, Qwen 3.8) via LoRA/QLoRA for domain-specific behavior
• Skills & Capability Modules — Self-contained behavior modules with explicit triggers, system instructions, and permission constraints
• Routing & Trust Boundaries — Zero-trust query routing, boundary isolation, and security sanitization across multi-tier LLMs
• Guardrails & Structured Validation — Pydantic v2 schema-first input/output validation, safety filtering, and structured JSON output

---

Contribution Metadata

• Last reviewed: 2027-01-07

• Confidence: high

• Sandboxed Code Execution — Pattern for safely executing untrusted code in isolated container/microVM runtimes.

• System Prompt Engineering — Methodology and design patterns for system prompts in LLM and agentic workflows.