[Feature Request] Auto-load multimodal files (PNG/HTML) from CLAUDE.md directories via @-imports

Bug Description
Feature request: auto-load multimodal context files
(PNG/HTML) from CLAUDE.md folders

Today CLAUDE.md's @-imports only resolve text files. Visual
artifacts (architecture PNGs, HTML dashboards) sitting next
to a CLAUDE.md are invisible to Claude unless explicitly
opened via Read, even though Claude has full vision
capability.

Proposal: extend @-imports so @./architecture.png loads as
Vision input (not as a text path), or add a settings.json
field like autoContextGlobs: ["*.png", "*.html"] that
auto-loads matching files from each CLAUDE.md-bearing
folder.

Why it matters: in AI-native repo structures where every
subfolder carries both text context (CLAUDE.md) and visual
context (_OVERVIEW.png, _FLOW.html), this closes the
asymmetry between human navigation (instant visual parsing)
and Claude's context (text-only). Current workarounds,
on-demand Read, custom MCP servers, SessionStart hooks, all
add friction this feature should remove.

—
Kilian Riester — Founder
Web: kilianriester.com
LinkedIn: linkedin.com/in/kilian-riester
GitHub: @KRiester
Email: riesterkilian@gmail.com
Happy to expand on use cases

Environment Info

• Platform: darwin
• Terminal: Apple_Terminal
• Version: 2.1.139
• Feedback ID: e8b22cb4-121f-4426-90b9-e7629234d339