docs(adr): 0011 live cursors in description editor (proposed)

[Musiker15]

Summary

ADR-first for the last v0.2.0-beta-deferred feature — live cursors in the card description editor (CLAUDE.md §5.2 USP). Per the project's ADR-first convention, this lands the decision record before any implementation; the editor swap is a follow-up PR gated on this ADR merging.

This PR is docs-only (ADR 0011 + the CLAUDE.md §14 index row). No code.

The decision

Adopt CodeMirror 6 + y-codemirror.next, behind a FEATURE_LIVE_CURSORS flag (default off).

Why not the alternatives:

• Keep textarea + hand-rolled cursor mapping — mapping plain offsets through Y.Text under concurrent edits is exactly the bug-prone problem the established bindings already solved; rendering selections over a <textarea> needs a shadow-mirror hack.
• TipTap / ProseMirror — rich-text model; would push us off the Markdown-source model and force a serialise/parse layer on every read/write/export/encryption path. Big migration, big bundle.
• CodeMirror 6 — edits plain text (Markdown source stays a string → zero change to storage/encryption/export/render), has the official Yjs binding that renders remote carets from Awareness, tree-shakeable, good a11y.

Why it's safe (zero-knowledge)

• Content + cursor metadata both travel through the existing E2EE WS relay encrypted under the BoardKey — the server keeps seeing only v1.<nonce>.<ct> (ADR 0003 preserved).
• The Awareness payload stays PII-free, mirroring PR feat(presence): board-level awareness via Yjs (#43) #50's board-presence shape: { userId, color } + opaque positions. No email, no name, no text. Same deterministic HSL colour as the avatar stack.
• Lazy-imported (next/dynamic, no SSR) so the perf budget (Lighthouse ≥ 95) is untouched until a self-host operator opts in.

What the follow-up implementation PR will do

(captured in the ADR's "Implementation outline" — not in this PR)

1. FEATURE_LIVE_CURSORS env flag.
2. Dynamically-imported description-editor.tsx wrapping CodeMirror 6 + yCollab over the card's existing Y.Text + Awareness.
3. Drawer renders CodeMirror when on, else the current textarea.
4. Tests: Awareness-payload PII guard; relative-position survives a concurrent insert.
5. Graduation gate: a11y review, Lighthouse ≥ 95 with editor loaded, two-user soak test.

Review ask

This is a design decision — if you'd prefer TipTap (richer, but a rich-text migration) or want to defer live cursors entirely, this is the place to say so before any code is written. If CodeMirror 6 looks right, merging this unblocks the implementation PR.

Test plan

• CI green (docs-only; markdown link check if any)
• Decision review: CodeMirror 6 vs TipTap vs defer

🤖 Generated with Claude Code