The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides an automatic validation feature that calls a peer script (subsystem-validate) upon completion. This call is implemented using safe process invocation methods (PowerShell -File or Python subprocess.run with an argument list) and resolves the script path relative to the skill's own directory, ensuring execution is confined to the vendor's local tooling.
[INDIRECT_PROMPT_INJECTION]: The skill processes external JSON definitions to generate metadata files. It effectively handles this attack surface by sanitizing all user-contributed strings before inclusion in the XML output.
Ingestion points: External JSON data supplied via the -Value or -DefinitionFile arguments in the PowerShell and Python scripts.
Boundary markers: The skill uses a dedicated Esc-XmlText function to escape XML special characters (e.g., &, <, >) in all generated tags.
Capability inventory: The skill is authorized to write metadata files to the project directory and execute a local validation tool.
Sanitization: The skill validates that the input is a correctly formatted JSON object and ensures that all dynamic values inserted into XML properties are properly entity-encoded.

subsystem-compile