Namespace LMKit.Agents.Skills
Classes
- AgentSkill
Represents an Agent Skill - a modular capability that extends an agent's functionality with specialized knowledge, workflows, and resources.
Agent Skills follow the Agent Skills specification, packaging instructions, metadata, and optional resources (scripts, templates) that agents use automatically when relevant.
- SkillActivatedEventArgs
Event arguments for the SkillActivated event.
- SkillActivator
Handles skill activation and context injection for agent conversations.
The activator prepares skill instructions for injection into the agent's context, managing the transformation from raw skill content to properly formatted prompts.
- SkillBuilder
Fluent builder for creating AgentSkill instances programmatically.
Use this builder when you need to create skills in code rather than loading them from SKILL.md files on disk.
- SkillContext
Represents the context of an activated skill.
This class encapsulates all relevant information about a skill activation, including the skill itself, user context, and available resources.
- SkillDiscoveredEventArgs
Event arguments for the SkillDiscovered event.
- SkillMatch
Represents a match result from skill discovery or semantic matching.
When an agent evaluates which skills might be relevant to a user request, it produces SkillMatch instances that rank skills by relevance.
- SkillMetadata
Represents the YAML frontmatter metadata from a SKILL.md file.
This class captures both required and optional fields as defined in the Agent Skills specification.
- SkillParseException
The exception that is thrown when a SKILL.md file cannot be parsed.
- SkillParser
Parses SKILL.md files according to the Agent Skills specification.
This parser extracts YAML frontmatter (metadata) and Markdown content (instructions) from SKILL.md files.
- SkillRegisteredEventArgs
Event arguments for the SkillRegistered event.
- SkillRegistry
Registry for AgentSkill instances that can be discovered and activated by agents.
The registry provides skill discovery, matching, and lifecycle management. It supports progressive disclosure by loading only metadata during discovery and full instructions on activation.
- SkillReloadedEventArgs
Event arguments for the SkillReloaded event.
- SkillRemoteLoader
Provides functionality to load Agent Skills from remote URLs.
Supports loading skills from:
- Direct SKILL.md file URLs
- ZIP archives containing skill directories
- GitHub repositories (raw content)
- SkillRemovedEventArgs
Event arguments for the SkillRemoved event.
- SkillResource
Represents a resource file bundled with an Agent Skill.
Resources include scripts, reference documentation, templates, and other supporting files that agents can access on demand. Content is loaded lazily to optimize memory usage.
- SkillResourceLoadedEventArgs
Event arguments for the ResourceLoaded event.
- SkillResourceTool
An ITool implementation that allows models to load skill resources on demand.
This tool enables LLMs to access templates, references, checklists, and other resource files bundled with skills during a conversation, without loading all resources upfront.
- SkillTool
An ITool implementation that exposes Agent Skills to models.
When invoked, this tool activates a specified skill and returns its instructions for the agent to follow. The tool description dynamically lists all available skills, implementing progressive disclosure.
- SkillWatcher
Monitors skill directories for changes and automatically reloads modified skills.
Provides hot-reload functionality for development scenarios where skills are frequently edited and should be reloaded without restarting the application.
- SkillWatcherErrorEventArgs
Event arguments for the Error event.
Structs
- SkillCacheInfo
Contains information about the skill cache.
Enums
- SkillInjectionMode
Specifies how skill instructions should be injected into the agent context.
- SkillReloadReason
Specifies the reason a skill was reloaded.
- SkillResourceType
Specifies the type of resource contained within an Agent Skill directory.
Resources are categorized by their purpose to help agents determine how to use them.