The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill contains examples that utilize high-privilege container configurations which could lead to container escape or host compromise if misused.
Evidence: In examples/03-monitoring-stack.md, the cadvisor service is configured with privileged: true and mounts the host's Docker socket at /var/run/docker.sock:/var/run/docker.sock.
Evidence: references/01-compose-file-reference.md explicitly lists /var/run/docker.sock:/var/run/docker.sock as a common volume mount strategy for service definitions.
[CREDENTIALS_UNSAFE]: The documentation and examples demonstrate managing secrets via environment variables and .env files, which may lead to accidental exposure.
Evidence: examples/01-lamp-stack.md provides a sample .env file containing placeholder passwords such as MYSQL_ROOT_PASSWORD=rootpass123.
Mitigation: The skill includes a 'Gotchas' section in SKILL.md that explicitly warns users to add .env files to .gitignore and recommends using Docker secrets for production environments.
[INDIRECT_PROMPT_INJECTION]: The skill acts on user-supplied project files and orchestration requirements, creating a potential attack surface for indirect instructions.
Ingestion points: The skill is triggered when users provide or ask to modify docker-compose.yml or project configurations as noted in the 'When to Use' section of SKILL.md.
Boundary markers: No explicit boundary markers or 'ignore embedded instructions' warnings are defined for processing user-provided file content.
Capability inventory: The skill can generate complex orchestration files, shell commands (docker compose up), and persistence scripts (systemd units in references/04-production-compose.md).
Sanitization: No specific sanitization or validation of user-provided environment variable keys or volume paths is described before inclusion in generated files.

docker-compose