Feature Request: Expose Kiro / Max routing to agents for prompt-based automatic switching

[Cassiel-girl]

Problem

CCH currently keeps the routing target transparent to the agent. When a request is handled through CCH, the agent cannot reliably know whether it is currently running through Kiro or Max, and it has no way to request a switch between them.

This matters because Kiro and Max can produce significantly different results on the same complex task. In practice, a task that is handled poorly by Kiro may succeed when retried with Max, but the agent cannot perform that fallback itself.

The current workaround is to switch between Kiro and Max manually in CCH. This prevents routing decisions from being expressed through natural language or project-level agent instructions.

For example, the following rules are currently understandable to an agent but not executable:

• Use Kiro for simple tasks.
• Use Max for complex coding, debugging, or refactoring tasks.
• If the user explicitly requests Max or Kiro, follow that request.
• If the result from Kiro is not satisfactory, retry with Max.

Expected behavior

CCH should expose the current routing target to the agent and provide an agent-callable mechanism for requesting a routing change.

Conceptually, this could be implemented as:

• Readable metadata or context indicating the current route, such as kiro or max.
• A tool or control interface that allows the agent to request a route change.
• A documented command or structured signal that CCH can intercept before the next model request.

For example, get_current_route() could return kiro, and an agent could request set_route("max").

The exact API is open for discussion. The key requirement is that the agent must be able to both observe the current route and communicate a routing decision back to CCH.

Use cases

• Simple tasks → use Kiro.
• Complex coding, debugging, or refactoring tasks → use Max.
• The user explicitly asks to use Kiro or Max → honor that request.
• Kiro produces an incomplete or poor result → automatically fall back to Max.
• A Skill, AGENTS.md, or system prompt specifies routing rules → allow those rules to be applied.

Benefits

This would make CCH routing programmable and agent-aware while preserving manual switching for users who prefer it. It would also enable task-based routing policies, model fallback, and more reliable results on complex work.

The goal is not to prescribe a particular routing algorithm, but to expose the necessary capability so users and agents can define one.