Bayesian-Agent Updates LLM Skills via Posterior Over Prompt-Context Pairs

Bayesian-Agent treats reusable LLM skills and SOPs as hypotheses with feature-conditioned categorical posteriors, updating them via verified trajectory evidence rather than heuristic reflection. Actions like patch, split, compress, and retire are triggered by posterior state, tested with deepseek-v4-flash.