The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/validate_compose.py uses subprocess.run to execute the docker compose config command for rendering deployment models. This implementation is secure as it uses a fixed argument list (no shell injection risk) and only processes files with specific, hardcoded filenames (e.g., compose.yaml, docker-compose.yaml) discovered within the provided root directory.
[INDIRECT_PROMPT_INJECTION]: As a tool designed to analyze external deployment artifacts, the skill inherently possesses a surface for indirect prompt injection if those files contain malicious instructions in comments or metadata intended to mislead the AI agent.
Ingestion points: The skill reads and processes all deployment-related files (Compose files, Dockerfiles, etc.) within the directory specified by the user in the --root argument.
Boundary markers: There are no explicit instructions or delimiters provided to the agent to distinguish between the configuration data it is validating and potentially malicious natural language instructions embedded within the files.
Capability inventory: The skill has the capability to perform filesystem reads and execute docker commands via subprocess to audit the environment.
Sanitization: While the skill performs robust structural validation and enforces security policies on the configuration data, it does not filter or sanitize natural language content (like comments) that the agent may observe during the validation process.

validate-deployment