The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a bash wrapper function openscad_run in SKILL.md that executes shell commands to run the openscad binary or a Docker container.
Evidence: The function openscad_run uses "$target" "$@" or docker run ... "$target" "$@" to execute the resolved compiler with arbitrary arguments.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided parameters (like dimensions for parametric models) and interpolates them into command-line arguments via -D flags.
Ingestion points: User requirements for model dimensions and configuration in SKILL.md and references/workflows.md.
Boundary markers: None specified to separate user data from command structure.
Capability inventory: Shell command execution via openscad_run and file system access via Docker volume mounts.
Sanitization: The instructions mention basic shell quoting for strings in SKILL.md (-D 'mode="parts"'), but do not provide comprehensive sanitization rules for the agent, creating a surface for potential command injection if user input contains shell metacharacters.
[EXTERNAL_DOWNLOADS]: The skill supports running OpenSCAD inside a Docker container. While it defaults to searching for a local image (openscad-cli), the image name can be overridden via the OPENSCAD_DOCKER_IMAGE environment variable, which could lead to pulling and executing untrusted images if the environment is configured to allow it.

use-openscad