Feature Request: Expose Agent ID or Type in PreToolUse Hook context

Problem

When using PreToolUse hooks to enforce security policies (e.g., blocking git commands for subagents), there is currently no way to distinguish whether a tool call originates from the main agent or a native subagent (such as general-purpose, Explore, or Plan).

Current Behavior

The hook input JSON contains:

{
  "session_id": "...",
  "transcript_path": "/path/to/session.jsonl",
  "tool_name": "Bash",
  "tool_input": { "command": "git status" },
  ...
}

Both the main agent and native subagents report the same session_id and transcript_path. There is no field to identify the caller.

Expected Behavior

The hook input should include additional context to identify the agent, such as:

{
  "agent_id": "a1b2c3d",
  "agent_type": "general-purpose",  // or "main", "Explore", etc.
  "parent_agent_id": null,          // null for main, ID for subagents
  ...
}

Use Case

Users may want to:

• Allow the main agent to run git commands while blocking subagents.
• Apply different permission policies based on agent type.
• Audit tool usage per agent for compliance or debugging.

Environment

• Claude Code CLI
• Hooks: PreToolUse
• Native subagents tested: general-purpose, Explore

Workaround Attempted

Checking for agent- prefix in transcript_path filename (as suggested in docs) does not work for native subagents, as they share the main agent's transcript.