LobeChat CVE-2026-58578

Exploitation requires an authenticated LobeChat account (CVSS PR:L) with access to the GitHub skill-import feature; the attacker supplies a malicious basePath in the GitHub repository URL path containing unescaped regex metacharacters (catastrophic-backtracking patterns) that reach the dynamically constructed regular expression in findSkillMd. The specific trigger is the skill-import code path that compiles the URL-derived basePath into a regex and matches it synchronously against archive entries. Limiting factors: valid authentication is mandatory, the skill-import feature must be enabled/reachable for the account, and impact is confined to availability (no data disclosure or modification); on single-user self-hosted instances the practical blast radius is limited to the attacker's own service.

The vendor CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:N/VA:H) scores 7.1 and correctly isolates impact to availability only - no confidentiality or integrity loss. The PR:L requirement is the key limiter: exploitation needs a valid authenticated account and access to the skill-import feature, so this is not an anonymous internet-wide DoS. That said, network attack vector, low complexity, no user interaction, and publicly available exploit code make it a realistic and low-effort disruption for any tenant on a multi-user instance, since one request degrades service for everyone. No EPSS or KEV data was provided; absence of KEV means there is no confirmation of active exploitation, and the real-world priority is moderate - high for shared/multi-tenant LobeChat deployments, lower for single-user self-hosted instances where the attacker would only be harming themselves.

An authenticated LobeChat user navigates to the skill-import feature and submits a crafted GitHub repository URL whose path contains a catastrophic-backtracking regex pattern as the basePath. When findSkillMd compiles and runs that pattern synchronously against archive entries, the Node.js event loop freezes for tens of seconds, making the application unresponsive for every other user; the attacker can repeat the request to sustain the outage. Public exploit code and a reproducing issue (issue #16494) are available, and the low attack complexity means no special tooling is required.