Adding another reproduction on Cursor **3.9.16** (Mac, July 2026), with hard evidence in case it helps prioritize.
I instrumented my hook script to log every payload and confirmed AskUserQuestion produces **zero** hook events across `preToolUse`, `postToolUse`, `postToolUseFailure`, `afterAgentResponse`, `Notification`, and `stop`. The internal Cursor hook log (`~/Library/Application Support/Cursor/logs//cursor.hooks.*.log`) shows no `Hook step requested` lines during the call. The matcher field (`{“matcher”: “AskUserQuestion”}`) is silently dropped — no error, no fire.
Also worth flagging: `Notification`, `PermissionRequest`, and `Elicitation` aren’t valid hook types in Cursor IDE at all — adding them to `hooks.json` produces `ERROR: Unknown hook type` on startup. So even the permission-prompt path has no hook coverage.
Claude Code landed a fix for `preToolUse` + `AskUserQuestion` matcher in v2.1.85 ([anthropics/claude-code#10168]( Add hook for user input/question events (UserInputRequired) · Issue #10168 · anthropics/claude-code · GitHub )). Can the same fix be ported into Cursor IDE? Right now anyone building status indicators / approval gates / observability tooling is completely blind to the “agent is waiting for user input” state.
Would appreciate even a rough timeline — I’m blocked on this for a hardware integration and need to decide whether to wait or build a workaround.