The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The GitLab Runner registration example in references/gitlab-runner-configuration.md uses the --docker-privileged flag. This flag provides the container with nearly all the capabilities of the host machine, which is a known security risk. The skill's main SKILL.md includes a specific 'DON'T' warning regarding privileged Docker usage.
[EXTERNAL_DOWNLOADS]: The pipeline configuration in references/complete-pipeline-configuration.md fetches the Trivy security scanner from Aqua Security's official GitHub repository to perform container image scanning.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands within GitLab CI jobs, including Docker CLI operations for building and pushing images, Helm and Kubectl for cluster deployments, and npm for package management and testing.
[INDIRECT_PROMPT_INJECTION]: The skill establishes automated CI/CD workflows that process external code, commit messages, and environment variables, creating a surface for indirect prompt injection if the agent is tasked with interpreting pipeline failures or logs containing malicious content.
Ingestion points: Pipeline job logs, source code repository contents, and environment variables ($CI_COMMIT_SHA, $CI_REGISTRY_IMAGE).
Boundary markers: None identified in the provided templates; scripts directly interpolate variables into shell environments.
Capability inventory: File system writes (artifacts, kubeconfig), network operations (docker push, helm upgrade), and arbitrary command execution (npm scripts).
Sanitization: No explicit sanitization or validation of commit-based variables before execution in shell environments.

gitlab-cicd-pipeline