The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the download of installation scripts, binaries, and Helm charts from official GitLab domains, including packages.gitlab.com and gitlab-runner-downloads.s3.amazonaws.com. These are legitimate and appropriate sources for the software being managed.
[COMMAND_EXECUTION]: The instructions involve the use of sudo for administrative tasks such as package installation (apt-get, yum, dnf), service management (systemctl), and host configuration. These commands are necessary for the operation of a system-level daemon like GitLab Runner.
[REMOTE_CODE_EXECUTION]: Several installation workflows include piping scripts from remote servers directly to a shell or executing downloaded binaries. These scripts and binaries are sourced from official GitLab infrastructure, which is a well-known service provider.
[PROMPT_INJECTION]: The skill provides instructions for viewing runner logs via commands like journalctl, docker logs, and kubectl logs (found in references/operations.md and references/installation.md). These logs ingest untrusted output from CI jobs and lack boundary markers or sanitization steps. Combined with the skill's broad system capabilities (sudo, docker, systemctl), this presents a surface for indirect prompt injection if an agent processes the log output.

gitlab-runner