Roo-Code CVE-2026-81833

Exploitation requires a low level of privileges (PR:L per CVSS 4.0), meaning the attacker must have some authenticated or credentialed access to the Roo-Code environment rather than being fully unauthenticated. Passive user interaction (UI:P) is also required, suggesting a victim user must perform a routine action - such as opening a file, triggering indexing, or initiating a query - to activate the vulnerable optimizeQuery code path in the CodeIndexManager component. The vulnerability is limited to products still running Roo-Code version 3.51.1 or earlier; environments that have already discontinued use of this archived tool are not affected. The fact that no attack requirements (AT:N) are specified means no unusual system state or race condition is needed beyond the above prerequisites.

Despite the RCE tag, the official CVSS 4.0 score is 2.0, reflecting low-level impacts (VC:L/VI:L/VA:L) and exploitation constraints (PR:L/UI:P). The low CIA impact scores suggest the injection may be constrained in scope rather than achieving full system compromise. Key risk signals: a public POC exists (E:P in the CVSS 4.0 threat metric), raising the likelihood of opportunistic use; the product is NOT in CISA KEV, indicating no confirmed widespread active exploitation. No EPSS score was provided in the input data, so exploitation probability cannot be quantified from that signal. The most significant risk amplifier is the permanent no-patch status - the vendor has archived the project and explicitly disclaimed responsibility. Any organization still running Roo-Code faces a vulnerability with no remediation path from the vendor. The low deployment footprint of an archived niche tool materially limits aggregate exposure.

A low-privileged remote attacker with access to the Roo-Code interface crafts a malicious input designed to manipulate the optimizeQuery function in CodeIndexManager, injecting executable code into the query-processing pipeline. With a public POC already available on GitHub, an attacker can adapt the demonstrated technique with minimal effort. Passive user interaction is required - for example, a developer opening or indexing a malicious project could trigger the vulnerable code path.