The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The scripts/init_skill.py utility applies executable permissions using chmod 0o755 to placeholder scripts generated during the initialization process. While this is a common step for developer bootstrapping tools, it is a privilege escalation pattern involving dynamically created files.
[INDIRECT_PROMPT_INJECTION]: The skill generates new SKILL.md and metadata files by interpolating user-supplied arguments (like the skill name) into templates.
Ingestion points: The skill_name parameter in scripts/init_skill.py and scripts/generate_openai_yaml.py.
Boundary markers: None are present in the internal SKILL_TEMPLATE string to delimit interpolated data.
Capability inventory: The skill has the capability to write to the file system, create directories, and modify file permissions.
Sanitization: The normalize_skill_name function uses regular expressions to enforce lowercase hyphen-case, which mitigates many common injection vectors despite the lack of boundary markers.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute included Python scripts (init_skill.py, generate_openai_yaml.py, and quick_validate.py) and shell commands to manage the skill development lifecycle. These actions are consistent with the skill's stated purpose as a creator tool.

skill-creator