The Agent Skills Directory

[COMMAND_EXECUTION]: The deployment script scripts/deploy.sh uses eval to execute shell commands built from variables (e.g., $COMPOSE_FILE, $STACK_NAME, $NAMESPACE). While the script uses single quotes to wrap these variables within the evaluated string, this pattern can be a risk if the agent or user provides input containing shell metacharacters.
[EXTERNAL_DOWNLOADS]: scripts/deploy.sh suggests downloading the kompose binary from the official Kubernetes GitHub organization. This is a well-known and trusted source.
[CREDENTIALS_UNSAFE]: Multiple files, including assets/compose_template.yml and assets/example_app_architectures.md, contain placeholder credentials like POSTGRES_PASSWORD=password and SECRET_KEY=YOUR_SECRET_KEY. These are illustrative markers for user configuration and do not represent the exposure of sensitive data.
[PROMPT_INJECTION]: The skill instructions in SKILL.md involve scanning existing project files for configuration details, which presents a surface for indirect prompt injection if those files contain instructions designed to influence the agent.
Ingestion points: SKILL.md (Step 1: Scan project for existing Dockerfiles and compose files).
Boundary markers: Absent.
Capability inventory: Write, Edit, and Bash access for docker and kubectl (specified in SKILL.md).
Sanitization: No explicit sanitization of ingested file content is documented.

generating-docker-compose-files