Your team has built a CI pipeline that invokes claude -p as a subprocess to auto-generate test scaffolding for pull requests. In roughly 15% of runs, Claude calls AskUserQuestion mid-session to resolve an ambiguity about the test framework. Because the pipeline runs headlessly, these calls hang indefinitely — the CI job times out after 10 minutes with no output. A teammate suggests returning permissionDecision: "defer" from a PreToolUse hook so the calling process can surface the question and resume later. Examining the pipeline logs, you notice that in the failing runs, Claude is making two or three parallel tool calls in the same turn — AskUserQuestion alongside Read calls. What is the most likely reason the defer approach will not solve the timeout in these cases?