CVE-2026-27955: Coolify Command Injection via Docker Compose Configuration

Technical summary

The vulnerability resides in the executeInDocker() helper function, which constructs bash commands using the pattern bash -c '{$command}'. User-supplied values from docker_compose_custom_build_command and docker_compose_custom_start_command parameters are interpolated directly into this string without quote escaping. Because single quotes in bash mark literal string boundaries, an attacker can close the intended quoted argument with a single quote, inject a semicolon or pipe operator, and append arbitrary shell syntax. The bash shell then interprets the injected code in the host execution context, not within the Docker environment. Proper remediation requires either parameterized command construction (avoiding shell interpretation) or comprehensive escaping of all shell metacharacters before interpolation.

Business impact

For organizations using Coolify to manage multi-tenant or multi-application infrastructure, this vulnerability means that any administrator account compromise—or intentionally malicious admin—can escape container isolation and gain host-level command execution. This breaks the security boundary that containerization is supposed to provide. An attacker could exfiltrate data across all hosted applications, modify infrastructure configurations, install persistence mechanisms, or disrupt service availability. The blast radius extends beyond a single application to the entire managed infrastructure. Organizations relying on role-based separation between application and infrastructure tiers lose that protection.

Affected systems

Coolify versions prior to 4.0.0-beta.464 are affected. The vulnerability requires authenticated access at the administrator level to exploit, since the vulnerable fields (custom build and start commands) are typically available only to admins who configure the Docker Compose orchestration. Self-hosted instances are the primary deployment model for Coolify, meaning affected users operate their own infrastructure and bear the full impact of compromise.

Exploitability

Exploitation requires valid administrator credentials; there is no unauthenticated or low-privilege attack path. However, once an attacker has admin access—whether through credential theft, insider threat, or social engineering—the vulnerability is trivial to exploit: a simple payload like ' || malicious_command # ' injected into a build or start command field will execute immediately during the next container operation. The attack is reliable and requires no special conditions (CVSS AC:L reflects this low barrier). Detection by the target organization may be delayed if command execution logs are not actively monitored.

Remediation

Upgrade to Coolify version 4.0.0-beta.464 or later. This release patches the executeInDocker() function to properly escape single quotes and other shell metacharacters before interpolating user input into bash commands. Organizations should prioritize this patch for instances accessible to untrusted administrators or shared infrastructure environments. After patching, verify that custom Docker Compose commands continue to function as expected, as the escaping may alter how special characters in legitimate commands are interpreted.

Patch guidance

Apply the upgrade to version 4.0.0-beta.464 as soon as practical. The patch is focused on the command execution pathway and carries minimal risk of regression. Before deployment to production, test any custom Docker build or start commands you have configured to ensure they still behave correctly after the escaping changes. Consider this a critical update for instances with multiple administrators or shared access models. Coolify's self-hosted architecture means patches must be manually deployed; there is no automatic or server-side remediation.

Detection guidance

Monitor Coolify's application logs for admin users creating or modifying custom Docker Compose commands with suspicious content (unmatched quotes, semicolons, pipes, or command substitution syntax). Endpoint Detection and Response (EDR) tools on the host server can detect unexpected child processes spawned from the Coolify container runtime or the Coolify application process itself—a sign that command injection succeeded. In container environments, watch for host-level syscalls initiated from the Coolify process tree. Review audit logs of who modified Docker Compose settings and when. Forensic indicators include: shell syntax in the custom_build_command or custom_start_command fields, unexpected process parents, and outbound connections from the Coolify process.

Why prioritize this

This is a medium-severity issue due to its authentication requirement (PR:H in CVSS), but it directly enables host compromise in environments that explicitly trust admins to configure containers. The risk is concentrated in organizations with: (1) multiple administrators, (2) admins from different teams with separate responsibilities, (3) any risk of admin credential compromise, or (4) shared Coolify instances serving multiple customers or business units. For single-admin, air-gapped, or tightly controlled Coolify deployments, the risk is lower but still present if that admin account is ever compromised. The scope impact (S:C) reflects that host-level code execution affects infrastructure beyond the application.

Risk score, explained

CVSS 6.6 (MEDIUM) reflects: network-accessible attack surface (AV:N), no complex conditions to exploit (AC:L), requirement for high-privilege credentials to trigger (PR:H), no user interaction needed (UI:N), and scope change from container isolation to host execution (S:C). The integrity, confidentiality, and availability impacts are all rated LOW rather than HIGH because the vulnerability is bounded by the administrator's legitimate permissions to manage infrastructure—the attack does not grant new logical capabilities, only escapes the technical control that was supposed to confine execution. Organizations with strict admin identity verification and short-lived credentials may experience lower practical risk than the base score suggests.

Frequently asked questions