progressive-discovery-skill

To reduce LLM token overhead, use progressive tool discovery to load capabilities only when needed instead of upfront. This approach checks which tools are actually connected before taking action, avoiding context pollution from enumerating all available tools at once.

Progressive capability loading is a context management technique where an agent starts with minimal context and discovers connected tools dynamically. It routes tasks by checking available capabilities first, delegating to specialized agents only when direct tools are missing.

Delegate tasks by checking if the required tool set is directly connected to the current agent. If the capability is missing, route the task to a specialized agent while passing only the focused execution context needed, reducing noise and improving decision quality.

Yes, capability-aware delegation works for device control, coding, web access, scheduling, messaging, and travel planning. It applies conditional routing to find the correct direct tool or specialized agent without loading unrelated tools into the context.

You should avoid upfront tool enumeration when managing complex agent workflows with diverse capabilities. Loading everything at once causes context pollution and wasted tokens; instead, use conditional delegation to execute with focused context only for the required capability.

The best way to route agent tasks is through capability checking of connected tools before execution. This method applies conditional delegation to the correct agent and maintains a focused execution context, avoiding the token optimization issues of upfront loading.