Render Mermaid diagrams inline

[rumpl]

[docker-agent]

Assessment: 🔴 CRITICAL

1 high-severity finding in new Mermaid rendering code.

Lower-confidence findings (not posted inline)

• [low] pkg/mermaid/parser.go:33 — Sequence diagrams with only notes (no messages) incorrectly rejected as ErrInvalidDiagram (confidence: weak 35/100)
• [low] pkg/mermaid/parser.go:69 — Escape check in splitMermaidStatements uses simple look-behind, mishandling double-escaped backslashes (confidence: negligible 25/100 — dropped)
• [low] pkg/mermaid/parser.go:144 — readDelimited byte-level escape check can misfire on multi-byte UTF-8 content (confidence: negligible 25/100 — dropped)
• [low] pkg/tui/components/mermaid/render_sequence.go:115 — Self-message label may overflow canvas width silently (confidence: negligible 20/100 — dropped)
• [low] pkg/tui/components/mermaid/render_graph.go:14 — append(roots, order...) may mutate roots backing array if it has spare capacity (confidence: negligible 20/100 — dropped)

[docker-agent]

[high] writeMermaidCanvas skips a column for every wide Unicode character, corrupting diagram layout

The loop advances start by 1 for every rune, but double-width characters (CJK, emoji) occupy 2 terminal columns. After writing a wide rune to canvas[start], the next character is placed at start+1 — the second visual cell of the wide character — causing characters to visually overlap and misalign.

The rest of the file correctly uses runewidth.StringWidth() and runewidth.RuneWidth() for width accounting; this function is the odd one out. The fix is to advance by the rune's display width and, for the two-column case, also blank out the adjacent cell so it is not filled with stale content:

Suggested change

	func writeMermaidCanvas(canvas []rune, start int, value string) {
	for _, r := range value {
	if start >= 0 && start < len(canvas) {
	canvas[start] = r
	}
	start++
	}
	}
	func writeMermaidCanvas(canvas []rune, start int, value string) {
	for _, r := range value {
	if start >= 0 && start < len(canvas) {
	canvas[start] = r
	}
	rw := runewidth.RuneWidth(r)
	if rw == 2 && start+1 >= 0 && start+1 < len(canvas) {
	canvas[start+1] = ' '
	}
	start += rw
	}
	}

Confidence	Score
🟡 moderate	67/100

[rumpl]

Fixed in aba4fe7. I changed the canvas from []rune to display-cell-aware []string storage: wide runes reserve continuation cells that serialize to zero additional width, and zero-width combining runes are appended to the preceding cell. This avoids the extra terminal column that a literal space in the continuation cell would introduce. Added regression coverage for CJK and combining characters in render_primitives_test.go.

[docker-agent]

Excellent fix! The cell-aware storage approach is much cleaner than my suggestion. Storing wide-rune continuation cells as empty strings and appending zero-width combining characters to the preceding cell correctly handles both categories of non-trivial Unicode without introducing phantom width. The test coverage for both CJK and combining characters confirms it works as designed.

Thank you for the thorough implementation!

[rumpl]

Review follow-up in aba4fe758:

• Fixed display-cell accounting for wide and combining Unicode characters and added regression tests.
• Allowed sequence diagrams containing notes without messages.
• Made quote escape detection handle runs of backslashes correctly.
• Avoided mutating the roots slice backing array during graph layout.

Validation passes with task build, task test, and task lint.