web: custom slash commands expand to full prompt in transcript, hide original shorthand

Summary

Custom slash commands (e.g. /loop ...) are expanded to their full prompt template text in the session transcript. The original shorthand I typed is lost, making history hard to scan. Prefer showing the shorthand with the expanded text collapsible.

Environment

• opencode version: 2.0.3
• OS: Linux omarchy 7.2.3-arch1-3 #1 SMP PREEMPT_DYNAMIC Sun, 06 Sep 2026 13:01:04 +0000 x86_64 GNU/Linux
• Terminal: TERM=xterm-256color
• Shell: /usr/bin/fish
• Install/channel: latest
• Active plugins: local plugins found (fileview.ts, loop-manager.ts in ~/.config/opencode/plugins/), no registry plugin entries in opencode.json

Reproduction

1. Define a custom command with $ARGUMENTS, e.g. ~/.config/opencode/commands/loop.md containing User request: $ARGUMENTS + instructions.
2. In TUI, type /loop my recurring task and submit.
3. Observe transcript message.

Expected Behavior

Transcript should show what I typed (/loop my recurring task), with the expanded full prompt available as collapsible / expandable detail, or via toggle/setting.

Actual Behavior

Transcript shows the translated/expanded full prompt text instead of /loop my recurring task. Original shorthand is not visible, so scanning history shows verbose template text rather than concise commands.

Additional Context

• Consistently reproducible.
• Custom commands affected, e.g. /loop, /advisor, /done from ~/.config/opencode/commands/.
• Example template: Load the loop-manager skill and follow it. User request: $ARGUMENTS ...
• Workaround tried: none — have to mentally map expanded text back to shorthand.
• Preference: shorthand + collapsible expanded view, or a config option in cli.json to choose display mode.