orchestrator

Mid-task skill injection monitors evolving task context to detect domain shifts and proactively suggests relevant skills for dynamic loading. It analyzes keywords from file types, tool usage, and discovered content to identify when new capabilities are needed.

Context monitoring detects domain shifts by analyzing keywords from file types, tool usage, and discovered content during task execution. When the context evolves into a new domain, the system identifies the shift and presents appropriate skill suggestions.

Yes, mid-task skill suggestions are never auto-loaded; they are always presented as checkpoints for manual user approval. You retain full control over skill injection and can approve or reject suggestions before any new skill is loaded.

You can define owner-defined activation triggers to exercise fine-grained control over skill surfacing. These custom triggers dictate exactly when the context monitoring system should proactively suggest injecting new skills based on your specific workflow requirements.

When code refactoring uncovers a security vulnerability mid-task, the context monitoring system detects this domain shift and suggests loading a security-focused skill. This mid-task suggestion appears as a checkpoint, allowing you to approve the injection to address the new concern.

No, the dynamic skill injection system features a modular design that can be enabled or disabled independently. Disabling mid-task suggestions will not affect or degrade the core agent functionality in any way.