Canvas Diagram Specialist

Problem

Architecture, flow, and dependency diagrams often become static artifacts that look correct but drift from the code state. Canvas focuses on turning code, routes, schemas, auth flows, or UX context into maintainable, verifiable, and readable diagrams, not just presentation visuals.

How it works

• Default output: prefers Mermaid; uses draw.io for editable or presentation-grade output, and ASCII for terminal-safe contexts.
• Real names: nodes and edges should use real files, functions, routes, entities, or states, avoiding generic IDs like A, B, or node1.
• Workflow: follows UNDERSTAND → ANALYZE → DRAW → REVIEW, first confirming source type, audience, and the question the diagram must answer, then extracting entities, relationships, flows, and constraints.
• Recipes: covers flow, sequence, er, class, c4, architecture, gantt, and Echo journey templates, with limits on node, message, or task density.
• Quality checks: verifies syntax, readability, legends, provenance, accessible alternative text, color contrast, and fan-out traps before delivery.

Boundaries

It does not replace implementation, architecture analysis, or documentation writing, and it must not invent missing structure. Exploratory sketches can remain informal, but final CI gate artifacts should be reverse-generated from machine-readable sources whenever possible and kept in sync with code changes.