Problem
Code blocks are plain monospace with no token coloring, so diffs and code the agent emits are hard to scan on a small screen.
Suggested fix
Add lightweight, dependency-conscious highlighting (e.g. a small highlighter or precomputed tokens) keyed off the (currently discarded) fence language. Keep it XSS-safe like the current renderer.
Reference
Claude / ChatGPT / Copilot all syntax-highlight code blocks.
Filed from a UI research sweep of the Helm mobile app (landing + live demo session) using Playwright + full source review, cross-referenced against Claude, ChatGPT, GitHub Copilot Chat, and Codex/Claude Code UIs.
Problem
Code blocks are plain monospace with no token coloring, so diffs and code the agent emits are hard to scan on a small screen.
Suggested fix
Add lightweight, dependency-conscious highlighting (e.g. a small highlighter or precomputed tokens) keyed off the (currently discarded) fence language. Keep it XSS-safe like the current renderer.
Reference
Claude / ChatGPT / Copilot all syntax-highlight code blocks.
Filed from a UI research sweep of the Helm mobile app (landing + live demo session) using Playwright + full source review, cross-referenced against Claude, ChatGPT, GitHub Copilot Chat, and Codex/Claude Code UIs.