The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute a local Python script (scripts/scaffold.py) that handles the file system operations and template rendering.
[REMOTE_CODE_EXECUTION]: The scaffold.py script generates executable Python files (execute.py) by populating templates with user-provided strings. The create_skill function uses Python's .format() method to insert the description variable into the execute.py.jinja template without escaping. This creates a risk where a malformed or malicious description could inject arbitrary code into the generated script if executed by a user.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface.
Ingestion points: User input for name and description in SKILL.md.
Boundary markers: None present in templates like SKILL.md.jinja or agent.md.jinja.
Capability inventory: The generated skills typically include Bash, Read, and Write capabilities.
Sanitization: Name validation via regex ^[a-z0-9-]+$ is present, but description is not sanitized.
[SAFE]: The skill performs regex-based name validation to prevent path traversal. It does not perform network operations or hardcode credentials.

create-hook