Skillset-based Dynamic Skill Reference Architecture

Skills are specialized instructions that teach AI agents how to perform specific tasks. Unlike general-purpose prompts, skills provide focused, reusable guidance that agents can discover and apply when relevant to the conversation. This architecture takes the concept further by packaging skills as installable skillsets—self-contained units that can be activated on demand to extend an agent's capabilities.

This blueprint demonstrates how to implement dynamic skill loading using ChatBotKit's skillset resources. Each skill is represented as a standalone skillset with a structured description format: a short summary followed by a separator (---) and detailed usage instructions. This convention allows the agent to quickly scan available skills while having access to comprehensive guidance when needed.

The architecture provides two core abilities:

List Skills uses the blueprint/resource/list template configured for skillsets to enumerate all available skills. The agent sees each skill's name and description, enabling it to identify which skills are relevant to the current task without loading their full context.

Install Skill uses the conversation/skillset/install[by-id] template to activate a skill by bringing its skillset into the conversation context. Once installed, any abilities defined within that skillset become available to the agent for the duration of the conversation.

The skillset-based approach offers distinct advantages over the file-based variant. When a skill is more than just instructions—when it includes abilities, secrets, or other resources—skillsets provide proper encapsulation. Installing a skillset doesn't just load text; it activates a complete capability package with its own tools and configurations.

This pattern is particularly powerful for building modular agent systems. Consider an enterprise assistant that starts with basic conversational abilities but can install specialized skillsets for HR queries, IT support, or expense management based on what the user needs. Each skillset brings not only domain knowledge but also the specific abilities required to interact with relevant systems.

The example skillsets (Skill 1 through Skill 8) use a placeholder format showing the recommended description structure. In practice, each skillset would contain domain-specific abilities and instructions. The short description helps the agent decide when to install the skill, while the detailed instructions guide its behavior once the skill is active.

Compare this to the file-based variant: files are ideal when skills are purely instructional text that should be loaded on-demand, while skillsets shine when skills need to package abilities, integrate with external services, or maintain their own configuration state. A key distinction is that when a skillset is installed, its description is directly injected into the agent's backstory—making the skill's instructions an integral part of the agent's system prompt rather than content retrieved mid- conversation.