Peer usernames are remote-controlled and were interpolated into an
innerHTML string in updateLobbyUI. The server only truncates them to 30
chars, so a peer could inject markup into everyone else's popup: enough
to load a remote image (leaking viewer IPs) or spoof readiness badges.
Inline handlers were already blocked by the MV3 default CSP.
Build the peer items with the DOM API, matching the pattern the sibling
peer list already uses.
Add the two checks that would have caught this before upload:
- eslint no-unsanitized, which reproduces the AMO warning at lint time
- addons-linter on the built XPI in verify-release, with
--warnings-as-errors since it exits 0 on warnings and AMO rejects them
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Tests: 20 deterministic avatar assertions (exact match, case insensitive,
longest-match-wins, ZWJ sequences, fallback), 30 randomized username format
checks, and validation that every noun in the list has an emoji mapping.
Tests: connect, room create/join, password auth, play/pause/seek relay,
force_sync, event_ack, episode_lobby, peer leave, ping/pong, room list,
protocol version check, deduplication, health HTTP endpoint.
All 13 tests pass as part of npm run verify.
Timo