claude-code - 💡(How to fix) Fix [FEATURE] PreToolUse hooks inside sub-agents cannot surface interactive output to the user terminal [1 participants]

Preflight Checklist

• I have searched existing requests and this feature hasn't been requested yet
• This is a single feature request (not multiple features)

Problem Statement

When a PreToolUse hook fires inside a sub-agent (spawned via the Agent tool), the hook process cannot display output to the user or read user input. The sub-agent's stdio is piped to the parent Claude Code process, not the user's tty. Calling open("/dev/tty") inside the hook fails silently or returns a tty disconnected from the user's session.

This means any hook that need to do the following (as examples), cannot do so from a sub-agent:

• Print a panel or message the user will actually see
• Prompt the user for a decision (allow / deny / modify)

The hook is forced to either default to Pass (allow) or auto-block without user input — neither of which is the intended behaviour.

Proposed Solution

Inherit the parent tty file descriptor into sub-agent hook processes. When Claude Code spawns a sub-agent, pass the parent's tty file descriptor through so hooks in the child can write to and read from it. This mirrors how Docker --interactive forward ttys to child processes.

Alternative Solutions

None that preserves interactive behaviour. The only current workaround is to auto-block (exit 2) unconditionally when tty is unavailable, which removes user choice entirely.

Priority

Medium - Would be very helpful

Feature Category

Interactive mode (TUI)

Use Case Example

https://github.com/dbmikeldb/tokenGate

A token analyser, developed to improve user knowledge and token efficency.

Additional Context

No response