Trust Propagation in Multi-Agent Systems Using Agent Skills

Formalize a trust propagation model for multi-agent systems that incorporate Agent Skills, ensuring security and operational practicality when authority is delegated and results are exchanged between agents.

Background

The paper highlights that multi-agent pipelines can propagate adversarial instructions (prompt infection) from a compromised agent to downstream agents, even those without malicious Skills installed.

There is currently no standardized guidance for how authority and trust should be granted, inherited, or evaluated across agent boundaries.

References

Formalizing a trust propagation model for multi-agent systems that is both secure and operationally practical is an open research challenge with implications beyond Agent Skills.

— Towards Secure Agent Skills: Architecture, Threat Taxonomy, and Security Analysis  (2604.02837 - Li et al., 3 Apr 2026) in Section 7.2, Open Challenges (C5: Trust Propagation in Multi-Agent Systems)

As a result it remains an open question whether trust develops similarly across virtual and physical settings in multi-robot HRT tasks, where trust must be distributed, calibrated, and maintained across multiple autonomous agents.

— Examining the Difference in Human Behavior Between Virtual and Real-World Human-Robot Teaming  (2609.18900 - Dallas et al., 16 Sep 2026) in Section 2, Related Works, subsection “Trust”