[Feature Request] Implement Configurable Subagent Hierarchies with Tool/Agent Restrictions

Here’s a more detailed and polished version of your feature request formatted for Claude Code:

---

Feature Request: Subagents Calling Other Subagents with Configurable Access Controls

Description
Currently, subagents in Claude Code cannot invoke other subagents directly. This limits the flexibility of agent orchestration, especially for complex workflows where delegation across multiple specialized subagents is desirable.

I would like to request a feature enhancement that allows:

1. Subagent-to-subagent calls – A subagent should be able to call other subagents, not just tools.
2. Configurable restrictions – Each agent (including the main agent and subagents) should have configurable permissions that define:

• allowed\_tools – The specific tools the agent/subagent is allowed to use.
• allowed\_subagents – The specific subagents the agent/subagent is allowed to delegate to.

This configuration could be managed within the existing settings.json file. An example schema might look like:

{
  "main_agent": {
    "allowed_tools": ["search", "code_editor"],
    "allowed_subagents": ["summarizer", "qa_agent"]
  },
  "summarizer": {
    "allowed_tools": ["text_analyzer"],
    "allowed_subagents": []
  },
  "qa_agent": {
    "allowed_tools": ["retriever"],
    "allowed_subagents": ["summarizer"]
  }
}

This setup ensures fine-grained control over agent orchestration, prevents unwanted recursive loops, and improves security by limiting scope per agent. This also ensures that only the relevant tools and subagent definition is passed to every agent, keeping the context clean.

---

Environment Info

• Platform: darwin
• Terminal: vscode
• Version: 1.0.98
• Feedback ID: e595245f-5a7c-4268-a9a8-8ae9b8617dd5

---

Errors

[]

---

Do you want me to also draft example use cases (e.g., “main agent delegates summarization to summarizer, which then delegates fact-checking to qa_agent”) so the feature request is stronger when submitted?