CVE-2026-123498 Details

A prompt injection vulnerability has been identified in Vikunja veans versions prior to 2.7.1. This vulnerability allows project members with write access to inject instructions by manipulating project titles or task descriptions. The injected content is rendered verbatim in the coding agent's authoritative instruction block. Exploitation occurs when the prime hook loads the crafted metadata during SessionStart and PreCompact, potentially leading the agent to alter the developer's checkout or execution commands.

To address this vulnerability, two key measures are recommended: First, implement a trust gate that prevents metadata from untrusted servers from reaching the prompt, ensuring that 'prime' fails closed when not interactive. Second, establish fencing to render remote metadata and task content from project writers on trusted servers as a labeled untrusted-data block, outside the authority section.