Expand description
§skm-select
Multi-strategy cascading skill selection engine.
This crate provides:
SelectionStrategytrait for pluggable selection algorithmsTriggerStrategyfor fast regex/keyword matching (µs)SemanticStrategyfor embedding-based similarity (ms)LlmStrategyfor LLM classification fallback (s)FewShotEnhancedwrapper for dynamic few-shot injectionCascadeSelectorfor composing strategies with early-exit
Structs§
- Cascade
Config - Configuration for the cascade selector.
- Cascade
Selector - Cascading skill selector.
- Cascade
Selector Builder - Builder for ergonomic cascade construction.
- FewShot
Enhanced - Wraps any strategy with dynamic few-shot example injection.
- FewShot
Example - A few-shot example for training.
- LlmStrategy
- LLM-based intent classification for ambiguous queries.
- LlmStrategy
Config - Configuration for LLM strategy.
- Selection
Context - Context available during selection.
- Selection
Outcome - Complete outcome of a selection, with full audit trail.
- Selection
Result - Result of a selection operation.
- Semantic
Config - Configuration for semantic selection.
- Semantic
Strategy - Embedding-based semantic similarity search.
- Trigger
Strategy - Fast-path matching using patterns defined in skill metadata.
Enums§
- Confidence
- Confidence level for a selection result.
- Latency
Class - Latency class for a strategy.
- LlmError
- LLM-specific errors.
- Merge
Strategy - How to merge results from multiple strategies.
- Select
Error - Errors from selection operations.
Traits§
- LlmClient
- LLM client trait — users implement for their LLM provider.
- Selection
Strategy - A skill selection strategy.