Zero-Trust Agents: Vendor-Neutral Agent Runtime for Autonomous Networks addresses one of the most critical barriers to production-grade Autonomous Networks: trust, security, and explainability of agentic systems at runtime. As CSPs progress toward Level-4 Autonomous Networks, the challenge is no longer whether agents can automate complex decisions, but whether operators can prove who acted, under which identity, within which policy boundaries, and why. This Catalyst proposes a reusable, vendor-neutral zero-trust runtime pattern that delivers exactly that assurance. The project defines and demonstrates a secure runtime fabric where every agent, tool, and AI model operates under a verifiable workload “birth certificate,” with signed lineage preserved across agent-to-agent interactions. Composite identity trust chains span agents, tools, LLMs, and OSS/BSS systems, while policy enforcement points apply default-deny and least-privilege controls to every interaction—regardless of whether workloads run on local GPUs, public cloud endpoints, or existing OSS platforms. Informed by the draft TM Forum IG1463 guidance on agentic threats and controls, the Catalyst moves beyond theory by testing these principles in realistic Autonomous Network scenarios such as cross-domain capacity optimization and hyper-personalized services. All runtime behavior is captured via OpenTelemetry and surfaced through an explanation view that enables CTIO and security teams to reconstruct and audit end-to-end agent workflows. The outcome is a practical, installable reference pattern that CSPs, vendors, and partners can adopt as they transition from isolated AI pilots to production-grade agent workflows. By linking zero-trust controls directly to runtime evidence and explainability, the Catalyst provides a safer, faster, and more repeatable path to scaling Autonomous Networks—while feeding real-world insights back into TM Forum standards development.