CVE-2026-58578: LobeChat ReDoS Vulnerability in Skill Import – CVSS 6.5

Technical summary

The vulnerability exists in the findSkillMd function, which constructs a regular expression dynamically using an unsanitized basePath parameter derived from a GitHub repository URL. When an authenticated user imports a skill, attacker-controlled input containing catastrophic-backtracking regex metacharacters (such as alternations with overlapping branches like (a+)+b) is embedded directly into the regex pattern and executed synchronously against archive entries. This causes exponential backtracking behavior that blocks the Node.js event loop, denying service to concurrent users. The synchronous execution amplifies the impact by preventing other requests from being processed during the backtracking phase.

Business impact

While exploitability requires authentication, the impact is severe for multi-tenant or shared LobeChat deployments. A single malicious authenticated user can render the entire service unavailable for all other users for extended periods by triggering skill imports with crafted URLs. This affects availability SLAs and user productivity. In environments where user accounts are easy to obtain or where internal users may be compromised, the attack surface is broader.

Affected systems

LobeChat versions before 2.2.10-canary.15 are affected. The vulnerability is specific to the skill import functionality that processes GitHub repository URLs. Any deployment allowing authenticated users to import skills is at risk.

Exploitability

Exploitation requires valid authentication credentials, which limits immediate external attack surface. However, the attack itself is trivial to execute—an attacker simply supplies a malicious basePath value during normal skill import workflow. No user interaction is needed beyond the import action, and the payload can be crafted with minimal technical knowledge. For organizations with permissive user provisioning or where compromised credentials are possible, the practical exploitability is high. The vulnerability is not currently in CISA's Known Exploited Vulnerabilities (KEV) catalog.

Remediation

Upgrade LobeChat to version 2.2.10-canary.15 or later. This version addresses the vulnerability by properly escaping or validating the basePath parameter before regex construction. Verify the patch against the vendor advisory to ensure the exact version matches the fix. As an interim mitigation, restrict the skill import feature to trusted administrators or disable it entirely until patching is complete.

Patch guidance

Apply the LobeChat update to 2.2.10-canary.15 or newer. Verify the upgrade in your deployment environment and confirm that skill import functionality behaves normally after patching. If you are on a stable release track, monitor vendor announcements for the availability of this fix in a non-canary release. Test the patch in a staging environment before production rollout to ensure compatibility with your deployment.

Detection guidance

Monitor application logs for unusual skill import requests with complex or suspicious basePath values, particularly those containing regex metacharacters or repetitive patterns. Look for periods of high CPU utilization on Node.js processes coinciding with skill import operations. Network-level monitoring may reveal prolonged request hangs followed by timeout events. Endpoint detection and response (EDR) tools can flag sustained CPU consumption by the Node.js process without corresponding I/O activity, which is indicative of regex backtracking.

Why prioritize this

Although this is a MEDIUM severity vulnerability requiring authentication, it should be prioritized for patching in organizations where skill import is actively used or where user account compromise is a realistic threat model. The ability to cause complete service denial with minimal effort makes it more urgent than the CVSS score alone suggests. Prioritize patching for internet-facing or multi-tenant LobeChat instances ahead of isolated or internal deployments.

Risk score, explained

The CVSS 3.1 score of 6.5 (MEDIUM) reflects the authentication requirement (PR:L) and purely availability impact (A:H), with no confidentiality or integrity compromise. The score appropriately penalizes the lower likelihood of exploitation due to the need for valid credentials. However, security leaders should recognize that within authenticated user populations or in scenarios of credential compromise, the practical impact approaches HIGH severity. The synchronous event-loop blocking creates a denial of service that affects all concurrent users, which is a severe business impact even if the vulnerability requires a foothold.

Frequently asked questions