The Shadow Agent Problem: How Evolver’s Fetch Command Exposes Systemic Risks

On May 5, 2026, Evolver disclosed a HIGH-severity vulnerability (CVE-2026-42334) in its fetch command, allowing a malicious Hub to overwrite project files via path traversal. The flaw stems from insufficient sanitization of Hub-supplied skill_id and bundled_files[], enabling attackers to place .js, .json, and .sh files in arbitrary directories.

The Shadow Agent Problem: Agents installed by individuals without IT approval represent the same threat as Shadow IT, but with broader system access.

This vulnerability exemplifies the Shadow Agent Problem. Evolver’s fetch command, designed to simplify skill updates, inadvertently amplifies systemic risks by allowing Hub-supplied files to bypass traditional IT vetting. The flaw lets malicious Hubs overwrite critical files, effectively executing arbitrary code. This mirrors the Shadow IT dilemma, where unapproved tools introduce vulnerabilities. However, with agents, the stakes are higher: unchecked Hub interactions can compromise entire systems. The vulnerability highlights a critical gap in agent ecosystem governance: the assumption that Hubs are inherently trustworthy.

Critics may argue that users should take responsibility for vetting Hub-supplied updates. However, this shifts an unreasonable burden onto individual users, especially in complex environments. The flaw underscores the need for built-in safeguards, not just user vigilance.