Expose invocation mode to hooks (interactive vs -p/pipe mode)

Preflight Checklist

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

Problem Statement

Hooks currently have no way to determine whether Claude Code was invoked interactively or via -p (pipe/non-interactive mode). The hook input includes session_id, transcript_path, cwd, permission_mode, and hook_event_name, but nothing about invocation context.

Proposed Solution

Add an invocation_mode field (or similar) to the hook input JSON, e.g.:

  {
    "session_id": "abc123",
    "hook_event_name": "Stop",
    "invocation_mode": "interactive",
    ...
  }

Possible values: interactive, pipe, remote (or whatever maps to the current invocation modes).

This would let hook authors conditionally run logic based on how Claude was started, without needing external workarounds.

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

Configuration and settings

Use Case Example

I have a Stop hook that should only run during interactive sessions — it doesn't make sense when Claude is invoked programmatically with -p. Today the only workaround is wrapping the CLI call with a custom environment variable (CLAUDE_PIPE=1 claude -p "...") and checking for it in the hook script, which is fragile and easy to forget.

Ex. playing audio on stop hook

Additional Context

_No response_