T03 · Remote Payload Retrieval and Execution
- Location
SKILL.md:42- Finding
Unpinned Remote Scripts Are Executed with Administrative Privileges
- Content
View full analysis
/dev/null 2>&1 systemctl enable docker > /dev/null 2>&1 systemctl start docker > /dev/null 2>&1 ``` ### Technical Analysis Both installation paths pipe an HTTP response directly into a shell. The downloaded scripts are not pinned to a specific version or immutable artifact, and their contents are not validated with a cryptographic digest or signature before execution. This creates a time-of-check/time-of-use supply-chain boundary: reviewing the Skill does not establish what code will execute later because the external server can return different content on every invocation. The Docker installation occurs in a workflow that performs system-level operations and enables services, while the OpenClaw command is explicitly intended to be run on the target VPS. In the expected deployment context, the commands therefore execute with administrative or root-equivalent authority. Suppressing the Docker installer output with `> /dev/null 2>&1` also reduces the operator's ability to detect unexpected actions or diagnose compromise. ### Attack Path 1. A user follows the Skill instructions or invokes the Traefik and Portainer installation workflow. 2. The VPS requests a script from `get.openclaw.ai` or `get.docker.com`. 3. The remote endpoint, its hosting infrastructure, DNS resolution, or another delivery component is compromised. 4. The attacker returns a modified shell script. 5. The response is immediately interpreted by `sh` or `bash` without inspection or integrity validation. 6. The payload execute ...[truncated 874 chars]- Remediation
View remediation
