Feature request: option to propagate `--dangerously-skip-permissions` / `bypassPermissions` to dispatched agents and their subagents

Summary

Add a way to make the parent session's bypass permission mode propagate automatically to every session dispatched from agent view (claude agents) and to every subagent those sessions spawn — without having to set defaultMode: bypassPermissions per-directory or permissionMode per-subagent-frontmatter.

Motivation

When running many parallel background sessions via claude agents (intended exactly for unattended work), the current model forces a choice between:

1. Manually configuring defaultMode: bypassPermissions in every working directory's .claude/settings.json, or
2. Manually adding permissionMode: bypassPermissions to every custom subagent's frontmatter, or
3. Dispatching from the shell with claude --bg --permission-mode bypassPermissions ... (one prompt at a time, defeating the point of the agent-view dispatch input).

The dispatch input inside agent view itself has no way to express "run this dangerously" — per the docs:

Dispatching from the agent view input doesn't pass a permission mode, so the session uses the defaultMode from that directory's settings or the permissionMode from the dispatched subagent's frontmatter.

And per #40241, even when the parent session is already running with --dangerously-skip-permissions, that mode does not propagate to subagents spawned via the Agent tool. The user has to re-approve permissions for each subagent, which is exactly the friction the flag was meant to remove for unattended runs.

Proposed behavior

One or both of:

• A propagatePermissionMode setting (or similar) that, when set with bypassPermissions, applies to: (a) sessions dispatched from agent view in that scope, and (b) subagents/Agent-tool invocations spawned by those sessions, transitively.
• A dispatch-input prefix or hotkey in agent view (e.g. !!<prompt> or Shift+Ctrl+Enter) that dispatches the new session in bypassPermissions mode, mirroring how ! already prefixes Bash replies in the peek panel.

In both cases, propagation should be gated by the same one-time interactive acceptance that --bg --permission-mode bypassPermissions already requires, so the safety property of "user has affirmatively opted in to unattended bypass" is preserved.

Why this isn't an exact duplicate

• #40241 covers parent → subagent propagation via the Agent tool, but not agent-view dispatch.
• #25503 is about the interactive dialog requirement for the flag itself.
• #54898 is about per-agent permission restriction, not propagation.

Environment

• Claude Code v2.1.140 (agent view, research preview)