Lean and directive throughout; it assumes Claude knows what diagrams and Mermaid are and spends tokens only on operational constraints, routing, and failure handling rather than concept explanation.

Highly actionable for an instruction/routing skill: a concrete routing table, exact universal constraints with syntax examples ('A["Process (main)"]', '-->|"O(1) lookup"|'), explicit reserved words, and named tool parameters (name, mermaidSyntax, fileKey, useArchitectureLayoutCode).

A clear 7-step sequence with explicit decision checkpoints (Step 1 right-tool check, Step 5 hybrid judgment with yes/no signals) and a feedback loop in Step 7 ('If the user is dissatisfied after 2 attempts... stop regenerating. Ask what specifically is wrong').

The body is an overview that routes to one-level-deep, clearly signaled reference files (architecture.md, flowchart.md, sequence.md, erd.md, state.md, gantt.md, workflow.md), all of which exist in ./references/, with type-specific detail appropriately split out.

Lists multiple concrete actions — 'Routes to type-specific guidance, sets universal Mermaid constraints, and tells you when to use a different diagram type or skip the tool entirely' — matching the multiple-specific-actions anchor.

Explicitly answers both what (routes, sets constraints, tells when to skip) and when ('Trigger whenever the user asks to create, generate, draw, render, sketch, or build a diagram'), satisfying the explicit-trigger requirement.

Excellent coverage of natural user terms: 'create, generate, draw, render, sketch, or build a diagram', 'flowchart, architecture diagram, sequence diagram, ERD...state diagram...gantt chart, or timeline', plus 'Mermaid syntax', 'decision tree', 'dependency graph', 'API call flow', 'auth handshake'.

A clear niche (Mermaid/FigJam diagram generation via the `generate_diagram` tool) with distinct, diagram-specific triggers and a mandatory-prerequisite framing that is unlikely to fire for unrelated skills.