SkillFuzz: Discovering Implicit Intents in Open Skill Marketplaces by Fuzzing Skill Compositions

The researchers propose SkillFuzz, a fuzzing method that discovers compositional implicit intents in open skill marketplaces without execution. The method treats skill compositions as the unit under test, extracts structured skill contracts, and uses contract-guided Monte Carlo tree search to prioritize combinations likely to conflict. On representative skill marketplace workloads, it discovers more than 1,000 distinct implicit intents under a fixed query budget, and confirms over 80% of the combinations flagged as highest risk in execution-time validation, while the explored pairwise interaction space is only a small fraction of what alternative search strategies require.