[prellr]

What & why

The Drafts annotation thread / side conversation (the right "Annotations" rail) rendered every thread message as raw plain text in a cramped ~320px column. A structured multi-point Aye reply — bold, numbered lists, headings, code — collapsed into an unreadable wall. This is a presentation pass on the thread display itself; no behavior, backend, or schema change.

Changes (frontend only — packages/views/drafts/)

1. Markdown rendering. Expanded-thread message bodies now render through the app's shared markdown renderer — Markdown from @multica/views/common/markdown with mode="minimal" (the mode whose docstring is literally "Best for: Chat messages"). Same renderer chat + channel messages use. Bold/lists/headings/inline-code/links render with tasteful, token-driven typography (no new markdown lib, no hardcoded colors).

2. Sender distinction (human vs Aye). Each message in the expanded thread is now a conversation row: an avatar + sender name + relative timestamp + the rendered body.

• Aye (author_type: "agent") reuses the base ActorAvatar (@multica/ui/components/common/actor-avatar, the Bot robot affordance) and is set apart with a primary-accent avatar/name plus a Bot icon. Labeled from the existing drafts.turn.agent_name ("Aye").
• Humans resolve their member name via the shared useActorName() hook (falling back to a new annotations.you string).

3. Room to breathe. Rail widened w-80 → w-96 (320 → 384px); the editor column (mx-auto max-w-3xl) is unaffected. When a pin is expanded the full conversation renders (including the opener, which was previously only shown truncated in the collapsed card), with comfortable padding/line-height and message grouping. Collapsed cards keep a compact one-line plain-text preview.

4. Mechanics unchanged. Typed annotation header (comment/question/suggestion/approve/block/highlight), reply box, Resolve/Dismiss/Reopen, open-count badge, orphaned tray, and re-anchoring all behave exactly as before.

Before → after (structure)

• Before: quote line → messages[0] raw text → replies as raw <p> text in tiny bg-muted boxes, no sender, no avatar, everything plain-text in 320px.
• After: quote line → expanded thread of conversation rows, each with avatar + sender (Aye/member) + timestamp + markdown body, in a 384px rail.

Reused components

• Markdown: Markdown (@multica/views/common/markdown, mode="minimal") — wraps @multica/ui/markdown.
• Agent styling/avatar: ActorAvatar base (@multica/ui/components/common/actor-avatar) — the Bot robot affordance. Accent uses the --primary semantic token (see note below).
• Name resolution: useActorName (@multica/core/workspace/hooks).

Note on "purple": the brief asked for a purple agent accent, but there is no purple/agent semantic token in the theme and CLAUDE.md forbids hardcoded Tailwind colors (the channels' bg-purple-100 is a pre-existing one-off I did not propagate). I used the --primary accent + the Bot icon as the token-correct distinct agent affordance. Easy to swap to a dedicated --agent token if one is introduced.

Verification

• pnpm --filter @multica/views typecheck ✅ and full monorepo pnpm typecheck ✅ (6/6)
• pnpm --filter @multica/views exec vitest run drafts/ ✅ (59) — extended the thread-panel test (markdown renders as <strong>/<li>; Aye + member sender labels show)
• pnpm --filter @multica/views lint ✅ (0 errors; target file clean)
• Full pnpm test ✅ (857 views tests, all packages green)

i18n: added annotations.you to en + zh-Hans.

🤖 Generated with Claude Code