The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and analyzes untrusted data from target Dockerfiles.
Ingestion points: The agent reads the target Dockerfile using sed and processes the output generated by the validation script in scripts/dockerfile-validate.sh.
Boundary markers: There are no explicit instructions to use delimiters or ignore instructions embedded within the Dockerfile content.
Capability inventory: The skill is capable of executing shell commands (bash, docker build) and performing file-write operations to apply fixes.
Sanitization: No sanitization or escaping of the external Dockerfile content is performed before interpolation into the agent's context.
[EXTERNAL_DOWNLOADS]: The scripts/dockerfile-validate.sh script fetches and installs the hadolint-bin and checkov packages from PyPI at runtime if they are not already available on the host system.
[DYNAMIC_EXECUTION]: The validation script creates temporary Python virtual environments in /tmp and executes binaries installed dynamically into those environments.
[COMMAND_EXECUTION]: The skill uses docker build and docker run (e.g., hadolint/hadolint) as fallbacks for validation, granting the agent significant execution privileges in environments where Docker is accessible.

dockerfile-validator