Multi-agent flows don't inherit permission approvals, causing repetitive prompts

When using multi-agent orchestration (via the Task tool), spawned agents don't inherit the parent agent's runtime approvals or respect pre-approved patterns in settings.local.json consistently. This results in getting bombarded with permission prompts for commands that have already been approved.

Steps to Reproduce

1. Have extensive pre-approvals in .claude/settings.local.json including patterns like "Bash(find:*)", "Bash(grep:*)", etc.
2. Run a multi-agent flow that spawns specialized agents
3. Watch as each spawned agent asks for permission to run commands like:

``bash
find ~/workspace/my_project/tests -name "*.test.ts" -type f < /dev/null | xargs grep -L "afterAll\|afterEach" | head -10
``

Expected Behavior

• Spawned agents should inherit parent's runtime approvals
• Pre-approved patterns in settings.local.json should be respected by all agents
• Complex piped commands that match component patterns should be auto-approved

Current Behavior

• Each spawned agent starts with fresh approval state
• Users get prompted repeatedly for the same types of commands
• Workflow is significantly disrupted by constant approval requests

Suggested Solutions

1. Add a flag to Task tool to inherit parent approvals
2. Improve pattern matching for complex piped commands
3. Add a "trust mode" for multi-agent workflows in secure environments
4. Allow wildcard approval patterns like "Bash(*)" for trusted workspaces

Impact

This makes multi-agent workflows extremely frustrating to use, defeating much of their productivity benefit.