What problem does it solve? When building agent-driven workflows, engineers repeatedly face three ambiguous tooling decisions: whether a new capability warrants an MCP server or just a CLI call, how to render Mermaid diagrams without silent syntax failures, and whether reusable behavior should ship as a skill or a dedicated agent. This Skill consolidates the decision criteria, anti-patterns, and gotchas for all three. ## Core Features & Use Cases - MCP Server Selection: Apply a five-criterion checklist (repeated need, stability, no duplication, security, performance) plus concrete anti-patterns before wiring up a new MCP server. - Mermaid Diagram Rendering: Use mmdc (mermaid-cli) as the sanctioned local renderer, with syntax rules covering <br/> vs \n, comments inside subgraphs, and edge-label special characters. - Skill vs Agent Mapping: Decide whether behavior belongs in an on-demand skill or a separately-dispatched agent with its own context window and tool scope. - Use Case: You are about to add an MCP server for a public API. The Skill's anti-pattern table tells you an unauthenticated API needs no standing connection — use WebFetch or a direct HTTP call instead. ## Quick Start Ask the agent whether a new capability should be an MCP server, a CLI call, a skill, or a dedicated agent, and follow the linked reference for that decision.
