Coolify EUVD-2026-41998

Exploitation requires an authenticated Coolify account with permission to edit application settings (CVSS PR:L), and specifically the ability to set the docker_compose_custom_build_command or docker_compose_custom_start_command fields with a single-quote-containing payload. The injected command only fires when a deployment of that application executes, so a build/deploy must be triggered. Exploitation is limited to Coolify instances prior to 4.0.0-beta.469 where the attacker can reach and modify these deployment command fields; it does not require special OS-level configuration, but it does require the application-edit privilege, which bounds the attacker population to existing operators/tenants rather than fully unauthenticated internet users.

The provided CVSS 3.1 vector (AV:N/AC:L/PR:L/UI:N, C:H/I:H/A:H, base 8.8 HIGH) indicates network-reachable, low-complexity exploitation requiring only low privileges (an account able to edit application settings) and no user interaction, with full confidentiality/integrity/availability impact. The real-world severity is amplified because the injected commands run on the host and escape Docker container confinement, effectively granting host-level command execution to any tenant/operator who can configure an application. No EPSS score, KEV listing, POC, or SSVC decision data was supplied, so exploitation-probability signals are absent; the assessment therefore rests on the CVSS vector and the vendor advisory description alone. The main mitigating factor is the PR:L requirement - an attacker must already hold an authenticated account with application-edit rights - which makes this most dangerous in multi-user/multi-tenant Coolify deployments or where such accounts are loosely provisioned. Note: the vendor description of host command execution escaping container confinement arguably implies a scope change (S:C), whereas the supplied vector uses S:U - verify with the vendor.

An attacker holding a low-privileged Coolify account with rights to edit an application sets docker_compose_custom_build_command (or the custom start command) to a value containing a single quote followed by injected shell syntax, e.g. closing the quoted context and appending a reverse-shell or data-exfiltration command. When a deployment runs, executeInDocker() builds the unescaped command string and the injected payload executes on the managed host outside container confinement. No user interaction is required beyond triggering the deployment; no public exploit code was identified at time of analysis.