The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous templates that use shell commands for CI/CD tasks, including curl for triggering deployment webhooks (e.g., Laravel Forge) and ssh for remote server management and deployment execution.
[PRIVILEGE_ESCALATION]: Several deployment examples involve the use of sudo to reload system services on remote servers (e.g., sudo systemctl reload nginx), which is an expected administrative requirement for deployment automation.
[CREDENTIALS_UNSAFE]: The documentation includes examples of accessing sensitive local paths such as ~/.ssh/id_rsa and ~/.ssh/config to set up repository secrets and SSH access. One GitLab CI template demonstrates echo $DB_PASSWORD, a practice that would expose sensitive credentials in pipeline logs.
[INDIRECT_PROMPT_INJECTION]: The skill enables the agent to generate workflow files based on user prompts. These templates interpolate secrets and environment variables into shell scripts, creating a potential surface for injection if user-controlled input is not sanitized before being incorporated into the generated CI/CD pipelines.
Ingestion points: User prompts for creating or configuring GitHub Actions/GitLab CI workflows (SKILL.md).
Boundary markers: No explicit delimiters are present in the provided templates to isolate user-supplied data from code.
Capability inventory: Templates generate workflows with capabilities for ssh, curl, sudo, docker, and kubectl execution.
Sanitization: No input sanitization or validation logic is included in the provided deployment templates.

ci-cd