feat(agent-gui): render Mermaid diagrams

[chovy-ai]

Summary

• render explicit fenced mermaid blocks in AgentGUI transcript Markdown
• lazy-load Mermaid and wait until streaming completes before diagram layout
• use strict rendering settings with bounded text/edge limits and source fallback
• preserve the existing fenced-code copy affordance for non-Mermaid blocks
• document the supported syntax and rendering boundary

Why

AgentGUI previously displayed Mermaid flowchart source as an ordinary code block, so architecture and delivery flows were difficult to read in conversation transcripts.

User impact

Agent responses can now include diagrams using a fenced mermaid block. Invalid or incomplete diagrams remain readable as source instead of leaving an empty transcript row.

Validation

• pnpm check:full
• pnpm --filter @tutti-os/agent-gui test (129 files, 2244 tests)
• pnpm --filter @tutti-os/agent-gui typecheck
• pnpm --filter @tutti-os/agent-gui build
• pnpm --filter @tutti-os/desktop build
• pnpm lint:ts
• pnpm check:i18n
• pnpm check:agent-activity-runtime-boundaries

---

Summary by cubic

Render Mermaid diagrams in AgentGUI transcripts to make architecture and flow visuals readable. Fenced mermaid code blocks now render after streaming completes, with strict settings and a safe source fallback.

• New Features

  • Render fenced mermaid blocks as diagrams in transcript Markdown.
  • Lazy-load mermaid and defer layout until streaming ends.
  • Strict mode (securityLevel: "strict", HTML labels off, bounded text/edge limits) with source shown on errors.
  • Keep existing copy affordance for non-mermaid code blocks; add i18n for rendering status and failure.

• Dependencies

  • Added mermaid@11.16.0.

^{Written for commit 9324c18. Summary will update on new commits.}