[FEATURE] Session Teams (Make structured sessions and sessions can comunicate each others interactively)

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

I am a software architect managing a multi-subsystem information platform composed of three technically distinct components:

• APP Layer: HTML5 frontend + Application Server
• Data Processing: C# + Microsoft SQL Server
• Billing System: COBOL (legacy core)

What I Am Trying to Do

I designed a hierarchical session architecture:

• 1 Lead Session acting as PM + Architect (produces ADRs, task assignments, interface specs)
• 3 Worker Sessions, one per subsystem, each with its own technical context

The Lead Session coordinates via shared files (docs/adr/, tasks/, BACKLOG.md). Worker Sessions read task files, execute, and write back status or escalation notes.

This maps closely to the multi-agent orchestration model Anthropic announced — and I am excited to see it becoming a first-class feature (Session Teams).

Proposed Solution

Specific Feedback & Feature Requests

1. Structured Escalation Path (Worker → Lead)

Currently, a Worker Session encountering a cross-system architectural question must write to a shared file and wait for a human to relay the message to the Lead Session.

Request: A native escalation mechanism where a sub-agent can flag a blocker to the coordinator with structured metadata (blocking reason, affected systems, urgency), without requiring human relay.

2. Role-Scoped CLAUDE.md Inheritance

Request: Support hierarchical CLAUDE.md files where the root-level file defines the Lead Agent's role and constraints, and subdirectory CLAUDE.md files define Worker Agent roles — with explicit scope boundaries (e.g., "this agent may not modify files outside ./billing/").

3. Cross-Agent Interface Contract Enforcement

In heterogeneous stacks, the most fragile point is the interface between subsystems (e.g., COBOL output format consumed by C# parser).

Request: A way to declare interface contracts (data format, field definitions) that all relevant agents are aware of and can validate against, reducing silent integration failures.

4. Agent Team Observability for Non-Developers

The current Agent Teams dashboard is terminal-based. For a PM or Architect role, a higher-level view showing task status per subsystem — without raw terminal output — would lower the barrier to adoption significantly.

Alternative Solutions

_No response_

Priority

High - Significant impact on productivity

Feature Category

Other

Use Case Example

_No response_

Additional Context

Why This Use Case Matters

Many enterprises run mixed-generation systems: a COBOL billing core alongside modern APIs and frontends. These are exactly the environments where AI-assisted maintenance has the highest ROI — and the highest coordination complexity. Claude Code's multi-agent architecture is well-positioned to serve this segment, but the cross-agent communication and role boundary features need to mature.

I believe the hierarchical orchestration model (Lead + Workers with clear role separation) is the right direction. These suggestions are intended to help it reach production readiness for real enterprise heterogeneous codebases.