dsh-mermaid-comm

In AI-assisted development conversations, topics like architecture, data flow, call chains, and state machines are often buried under walls of prose. Even when the model does emit Mermaid code, a stray Unicode arrow, an unpaired quote, or a minor syntax slip is enough to break rendering. dsh-mermaid-comm steps in across three layers—prompt guidance, syntax validation, and output gating—so that structured, renderable Mermaid diagrams become the default form of expression rather than an afterthought.

Four capabilities work together: behavior guidance maps six topic types to their appropriate diagram kind and enforces a diagram-first, briefly-explained style; syntax validation scans for dangerous characters and then performs a real parse via the same-version Mermaid runtime shipped with dsh-mermaid; the output gate intercepts assistant message events, auto-fixes recoverable issues, and replaces still-broken blocks with a safe notice so nothing unrenderable reaches the renderer; Mermaid Vault persists validated diagrams as versioned local assets with evolution history, injects the vault index into the system prompt, and steers later conversations to read and evolve existing diagrams instead of redrawing from scratch.

This plugin fits developers working within the dsh / dsh-mermaid rendering pipeline who want AI conversations to consistently produce renderable, traceable Mermaid diagrams. It is especially useful for teams iterating on system architecture, data models, or core flows across multiple sessions.