The Agent Skills Directory

[PROMPT_INJECTION]: No prompt injection or behavior override instructions were found in the skill metadata or body.
[CREDENTIALS_UNSAFE]: The skill correctly demonstrates the use of CI/CD environment variables (e.g., $CI_REGISTRY_PASSWORD) rather than hardcoding credentials. It explicitly advises against storing plaintext passwords in configurations.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns, such as piping external scripts to a shell, were detected. The CI/CD examples use standard tooling (npm, cargo, docker, kubectl).
[EXTERNAL_DOWNLOADS]: References and docker images (e.g., node, alpine, docker) target official or well-known registries. Links point to the official GitLab documentation.
[DATA_EXFILTRATION]: No suspicious network operations or exfiltration patterns were identified. The network calls demonstrated are standard for container registry logins and package management.
[OBFUSCATION]: No obfuscated strings, hidden characters, or encoding techniques were found in the file content.
[PRIVILEGE_ESCALATION]: The skill does not contain instructions for privilege escalation. In fact, it recommends best practices like using Kaniko for rootless builds instead of privileged Docker-in-Docker containers.
[INDIRECT_PROMPT_INJECTION]: The skill does not process untrusted external data in a way that creates an indirect prompt injection surface.
[DYNAMIC_CONTEXT_INJECTION]: No dynamic context injection patterns using the exclamation-backtick syntax were found.

gitlab-ci