Bundle the panel script so it can import core (one source of truth for shared feature code)

[RemilioNubilio]

Summary

The VS Code panel is one template string in core (chatHtml() in chat/ui/webview.ts) whose inline script cannot import anything. So every shared behavior lands twice: the quote ref grammar ships as a byte-identical regex copy pinned to notes/quoteRefs.ts by change-BOTH comments (#213), the quote card logic exists in a panel form and a CLI form, and the hydration cap lives under two names. The copies are documented and tested, but every future shared feature pays the same tax, and the two sides drift the moment one is edited alone.

Proposal

Give the panel script a small build step of its own:

1. Move the inline script out of the template string into packages/core/src/chat/ui/panel/ as a real module.
2. Bundle it with the tsup the repo already uses (IIFE output, no new dependencies).
3. chatHtml() inlines the bundled artifact instead of carrying the source. The HTML shell stays a template; only the script becomes a module.

The panel then imports extractQuoteRefs, QUOTE_REFS_MAX, and future shared helpers directly, and the copies plus their pin comments are deleted.

Scope notes

• No behavior, message, or contract change: a refactor of how the panel script is authored, not what it does.
• Incremental migration works: extract the feed and quote section first (where the live duplication is), leave the rest of the script in place until convenient.
• Happy to build it if wanted. Parking the idea here first since it changes how the panel is authored, which is yours to call.

[zo-sol]

Approved, go for it. This is the right call: the panel script being an unimportable string is exactly why the quote grammar ships as a pinned copy and the card logic exists twice, and moving it to a real module turns the hand-kept pins into compiler-checked imports.

On scope: if you are fully confident in the approach, you do not have to stay incremental. A full migration of the panel script to a module is fine, on one condition: heavy test coverage. The inline script has never been type checked or unit tested (it was only node --check and headless renders), so porting all of it at once is only safe if the move comes with real tests around the extracted modules and the bundled artifact, plus a headless render pass proving the panel still boots and the feed, quote cards, and composer all work. Land it behind that and the byte-identical copies plus their change-BOTH comments can all be deleted.

No behavior or message change, as you noted. If the bundle step touches build order (panel bundle then chatHtml inline), call that out in the PR so the publish path stays reproducible.

[RemilioNubilio]

Landed in #218, and the engine-update banner is the first feature written against the modules. Closing.