The Agent Skills Directory

[SAFE]: The core instructions in SKILL.md provide guidance for skill lifecycle management, including scoping, behavioral testing, and delivery. There are no attempts to override safety filters or perform prompt injection.
[DYNAMIC_EXECUTION]: The Python scripts scripts/package_metadata.py and scripts/validate-package.py utilize yaml.load(). A manual audit confirms that the UniqueKeyLoader class used in these calls inherits from yaml.SafeLoader, which prevents arbitrary code execution during deserialization. The static detector flags these as a precaution, but the implementation is secure.
[DATA_EXFILTRATION]: The skill does not perform network operations. In scripts/validate-package.py, a list of REMOTE_SCHEMES (including http and https) is explicitly used to filter out and skip external targets during link validation, preventing accidental network requests.
[COMMAND_EXECUTION]: There is no evidence of subprocess spawning, shell execution, or privilege escalation. The scripts are limited to file system read operations for validation purposes.
[INDIRECT_PROMPT_INJECTION]: Although the skill is designed to process and validate other skill files, it implements robust path validation. The validate_skill function in scripts/validate-package.py uses relative_to checks to ensure that file access remains within the specific skill directory, mitigating potential directory traversal attacks from malicious input files.

ko-skill