Subagents cannot request interactive tool permissions from the parent session

Problem

When the main agent launches a subagent to verify an implementation, the subagent may need to execute tools such as Bash to run a linter, tests, TypeScript checks, etc.

The problem is that the subagent cannot request interactive permission from the user at runtime. The permission prompt belongs to the subagent context, while the user only has access to the main session UI.

For example:

Main agent → launches verification subagent → subagent runs npm run lint → Bash requires permission → subagent cannot ask the user → command fails → subagent reports the verification as failed

If the same Bash permission request appeared in the main session, the user could approve it interactively.

Expected behavior

Permission requests originating from subagents should be forwarded to the parent/main session so the user can approve or deny them interactively.

Ideally:

Main agent → subagent → Bash permission required → permission request forwarded to main session → user approves → Bash executes in the subagent

Alternatively, there could be a configurable permission mode allowing subagents to inherit the parent session's permissions while forwarding previously unauthorized tool requests to the parent.

This is especially important for verification subagents that need to run linters, tests, builds, or type checks.

Environment

OpenClaude 0.30.0

Important

Using --dangerously-skip-permissions / --yolo is not an adequate solution for users who want to retain interactive permission control.