[Feature Request] Cascade Teams: Multi-agent collaboration for cross-domain tasks

[Neizan93]

Problem Statement:
While Cascade excels at deep context awareness, executing comprehensive cross-domain workflows (e.g., developing a full-stack feature with frontend implementation, backend logic, and CI/CD integration) within a single Cascade instance leads to context saturation. Currently, there is no native segregation of duties or isolated context windows for different technical domains, which dilutes the specialization of the output.

Proposed Solution: Cascade Teams
Evolve Cascade into an orchestration engine capable of deploying Cascade Teams: predefined groups of domain-specific agents that pass context and artifacts sequentially or run concurrently to achieve a unified objective.

Workflow Execution:

1. Scope Definition: The developer defines a cross-domain objective in Cascade (e.g., "Implement a new feature: a Kotlin Spring Boot backend API, a Next.js frontend using TSX and Zustand for state management, and full test coverage using Vitest").
2. Team Provisioning: Cascade suggests a multi-agent roster tailored to the stack (e.g., Backend Engineer Agent, Frontend Engineer Agent, QA Agent).
3. Pipeline Configuration & Sign-off: The user audits the assigned roles, modifies parameters or system prompts, and executes the team workflow (Go/No-Go).
4. Telemetry & Orchestration: Cascade displays a visual execution pipeline. As the Backend Engineer Agent generates the data models and endpoints, the Frontend Engineer Agent concurrently scaffolds the UI components, passing artifacts internally.
5. Governance: Total control to pause the Cascade Team at any pipeline stage, inspect agent-to-agent communication, validate pnpm test-ci or pnpm build outputs, and approve the final codebase generation.

Value Proposition:

• Role-Specific Accuracy: Prevents model hallucination and enforces clean code practices by constraining Cascade's focus to one technical domain at a time.
• Automated Peer Review: Allows one agent to act as a reviewer/QA for another's output, automatically iterating on failing tests before presenting the final result.
• Optimized Context Window: Distributes context load across specialized sub-agents, significantly improving performance on large codebases or monorepos.

Acceptance Criteria:

• Submitting a complex objective triggers a Cascade Team roster proposal.
• The UI displays specific agent roles and pipeline phases.
• Users can modify the team structure before runtime execution.
• Inline telemetry shows real-time progress for each concurrent agent.
• Global execution controls (Pause/Stop/Rollback) are available.