The CVSS 7.8 / EPSS 0.00456 gap for this vulnerability isn't a scoring anomaly — it's a structural mismatch between how vulnerability models price attack complexity and how VS Code's extension architecture actually creates exposure.
A CVSS 'Local' designation implies an attacker needs shell access or physical proximity. But VS Code extensions can be authored by anyone and execute arbitrary code on install, with access to terminals, file systems, AI-powered completions that proxy secrets, and settings sync that distributes credentials across machines. The 'Local' qualifier is technically accurate — the proximate trigger is user action or a local extension — but it obscures the blast radius: a compromised developer environment means lateral movement into production cloud credentials, internal repos, package registries, and CI/CD pipelines.
This isn't a novel vulnerability — CVE-2023-41069 (extension host privilege escalation), CVE-2024-28956 (unsandboxed extension execution), and earlier CVEs share the same genetic fingerprint. Each patch hardens one node in the extension trust chain; the attack surface migrates one step upstream. The original trust model was designed for a local editor with convenience extensions, not for the terminal access, credential handling, and cross-machine sync VS Code now provides.
The critical question the CVE description doesn't answer: is this a gate failure (a design assumption that was violated) or a bypass of secondary hardening that users are expected to configure? That distinction determines remediation strategy — architectural redesign versus user education — but either way, treat this as a supply chain adjacency risk, not a standalone local issue. Review extension permissions, restrict settings sync to minimal credentials, and assume that any extension compromise grants access to everything your developer context touches.