The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a factory pattern that generates new skill files using user-provided text from an interactive Q&A. This behavior introduces a surface for indirect prompt injection where untrusted input is interpolated into the frontmatter and instructions of generated skills. \n- Ingestion points: Interactive user responses collected during the 11-step setup process in SKILL.md. \n- Boundary markers: Absent; the template engine in references/skill-template.md uses standard placeholders (e.g., {{SKILL_NAME}}) without delimiters to separate untrusted user content from system instructions. \n- Capability inventory: The skill possesses extensive file-write capabilities (creating new skill directories and markdown files) and performs network operations via an external script during its trial phase. \n- Sanitization: No escaping, validation, or filtering of user-provided strings is performed before they are written to the new skill's SKILL.md or style-guide.md. \n- [DATA_EXFILTRATION]: The skill's template file, references/skill-template.md, contains a hardcoded local file path starting with '/Users/ugreen/Documents/obsidian/09image/'. This constitutes a minor data exposure, revealing the author's local system username ('ugreen') and specific directory structure.

image-skill-builder