Text drawn from MCP elicitation requests, WebSocket/MCP tool descriptors, Bash commands, skill names and descriptions, notebook cell ids, and select-artifact labels now passes through display sanitizers with a size cap before being rendered, instead of being shown raw; MCP server and tool names fall back to placeholders like "(unnamed server)" when needed. In the elicitation URL dialog, the accept option is only pre-selected when both the message and URL round-trip cleanly with no newlines, and the accept handler refuses to open the URL otherwise. Option previews that are too long are marked withheld and their always-allow rows are suppressed.
Untrusted text in permission and elicitation dialogs is truncated and sanitized
Untrusted text shown in permission and elicitation dialogs is now sanitized and length-capped
You'll noticeTier: how much it should matter to you
1Useful: my rating, 1 to 5
1Signal: worth watching, 1 to 5
PermissionsArea: what it touches
ImprovementsKind: in v2.1.235,
ImprovementsSection of the release