The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a local Python script (validate_output_style.py) for the discovery and validation of output styles. The script follows security best practices by implementing path confinement checks (is_within logic) to ensure that file operations remain within the designated plugin or repository roots, effectively preventing directory traversal attacks.
[INDIRECT_PROMPT_INJECTION]: The skill processes markdown files that contain instructions for the agent's persona and behavior. While this represents a surface for indirect prompt injection, the skill mitigates this through a multi-layered validation process.
Ingestion points: The agent reads style definitions from local markdown files in .claude/output-styles/ or plugin directories.
Boundary markers: Instructions explicitly designate these files as system prompt replacements, providing clear context for their interpretation.
Capability inventory: The skill is limited to reading/writing local configuration files and executing specific, bundled validation tools.
Sanitization: The automated validator enforces a strict YAML schema for frontmatter (using yaml.safe_load), while Phase 5 of the workflow provides a manual checklist for the agent to verify the safety and imperative nature of the instruction body before activation.

output-style-creator