CaSKG Framework for Counterfactual-Causal Agent Skill Retrieval

CaSKG improves LLM agent skill retrieval by building directed candidate graphs calibrated through textual counterfactual probes. By removing, substituting, or reordering skill pairs with Bayesian smoothing, the framework ensures more reliable procedural context than standard vector or graph retrieval.