The paper introduces SkillFuzz, an execution-free testing approach designed to discover implicit intents in open skill marketplaces where individually benign skills may interact to redirect agents toward unintended objectives. By formulating this discovery as a fuzzing problem over skill compositions, the method extracts structured contracts and uses contract-guided Monte Carlo Tree Search to prioritize potentially conflicting combinations.
- SkillFuzz is the first execution-free testing approach that extracts structured skill contracts and utilizes contract-guided Monte Carlo Tree Search.
- Across representative workloads, it discovers over 1,000 distinct implicit intents under a fixed query budget.
- The system confirms more than 80% of the highest-risk flagged compositions during execution-time validation.
- It identifies substantially more high-severity implicit intents than alternative search strategies while exploring only a fraction of the pairwise interaction space.
This approach addresses the challenge of detecting effects that emerge only through skill composition, offering a scalable way to audit agent behavior without requiring execution environments at admission time.