The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to execute a remote installation script directly in a bash shell using the command curl -fsSL https://raw.githubusercontent.com/ZupIT/horusec/main/deployments/scripts/install.sh | bash in SKILL.md. This allows for arbitrary code execution from an unverified third-party source.
[REMOTE_CODE_EXECUTION]: In assets/ci-config-template.yml, the skill uses curl -s https://raw.githubusercontent.com/aquasecurity/tfsec/master/scripts/install_linux.sh | bash to install security tools, which executes remote code without prior verification.
[COMMAND_EXECUTION]: The troubleshooting section of SKILL.md recommends sudo chmod 666 /var/run/docker.sock. This command grants every user on the host system full control over the Docker daemon, which is a significant privilege escalation risk equivalent to root access.
[COMMAND_EXECUTION]: Instructions in SKILL.md and CI templates recommend mounting the host's Docker socket inside containers using -v /var/run/docker.sock:/var/run/docker.sock. This configuration is a known security risk that enables container escapes to the host system.
[EXTERNAL_DOWNLOADS]: The skill fetches code and configuration from several external sources, including github.com/ZupIT and github.com/aquasecurity.
[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (ingestion points: project source code and git history) while maintaining high-privilege capabilities such as Docker and shell access (capability inventory). It lacks boundary markers for scanned content and does not specify input sanitization (sanitization: absent), creating an attack surface for indirect prompt injection via malicious code comments or documentation within a scanned repository.

sast-horusec