Junglewise Threat Intelligence

CVE-2026-58578: LobeHub LobeChat ReDoS in GitHub skill import

CVE-2026-58578 · Severity: medium · CVSS 6.5 · Published 2026-07-02

Technologies: LobeHub LobeChat. Vendors: LobeHub.

Executive brief

LobeChat, an open-source AI chat framework, is vulnerable to a denial-of-service attack during the "skill import" process. An authenticated user can provide a specially crafted GitHub URL that causes the server to become unresponsive for nearly a minute per request. This can be used to block all other users from accessing the service, leading to significant operational downtime.

Technical details

A Regular Expression Denial of Service (ReDoS) vulnerability exists in the `agentSkills.importFromGitHub` tRPC endpoint. The root cause is the `findSkillMd` function in `src/server/services/skill/parser.ts`, which interpolates a user-controlled GitHub URL path segment (`basePath`) directly into a `new RegExp()` constructor without escaping. An authenticated attacker can provide a path containing catastrophic backtracking patterns (e.g., `(a+)+`). When the server processes the repository ZIP, the synchronous `.test()` call against archive entries blocks the Node.js event loop, denying service to all concurrent users. The vulnerability was fixed in version 2.2.10-canary.15 by replacing dynamic regex construction with plain string matching.

Affected products

  • LobeHub LobeChat before 2.2.10-canary.15

Timeline

  • 2026-05-18: disclosed: Initial private disclosure via GitHub Security Advisory
  • 2026-06-30: disclosed: Public issue opened on GitHub repository
  • 2026-07-01: patched: Fix merged into canary branch
  • 2026-07-02: advisory: CVE published to NVD

References

Related threats