ai-agents

What it does

Describes production-ready operational patterns for designing, building, evaluating, and deploying AI agents. Centralizes procedures, checklists, decision rules, and templates used across RAG agents, tool-using agents, OS agents, and multi-agent systems.

How it works

Outlines a production workflow: choose architecture via a Decision Tree, draft agent specs using predefined templates, specify tools with JSON Schema, add retrieval when needed, include evaluation and telemetry, run a Go/No-Go gate, and plan deployment with safety controls. Provides quick-reference agent types (Workflow/FSM, Tool-Using, RAG, Planner/Executor, Multi-Agent, OS, Code/SWE) and guidance on MCP vs A2A usage, context management, and guardrails. Lists core concepts, memory options, failure handling, and Do/Avoid practices. Includes frameworks, templates, and references for implementation and governance.

When to use it

Use when designing or evaluating agent capabilities, building action loops and delegation logic, writing tool definitions or MCP schemas, creating RAG pipelines, building memory systems, establishing observability and safety gates, planning CI/CD and production specs, generating templates, implementing MCP servers, setting up agent handoffs, and performing build-vs-not decisions or ROI analyses.

What it can touch

Tools and templates referenced include: MCP integration concepts, A2A handoffs, tool definitions, RAG patterns, memory systems, evaluation/observability references, and CI/CD/safety checklists. Specific items include files and references such as assets/core/agent-template-standard.md, assets/tools/tool-definition.md, references/evaluation-and-observability.md, and references/deployment-ci-cd-and-safety.md.

Caveats

License: MIT. The skill emphasizes production patterns and templates, with stated guardrails and governance requirements; no guarantees of outcomes are provided. No claims beyond the described procedures, templates, and reference materials are made.