The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The script scripts/init_plugin.py accepts user-provided strings for plugin metadata and writes them directly into the generated .claude-plugin/plugin.json and README.md files. This establishes a surface where malicious instructions could be embedded in metadata and later executed by other agents that process the generated plugin files.
Ingestion points: description, author_name, and author_email fields in scripts/init_plugin.py.
Boundary markers: No delimiters or safety warnings are used when writing the user-provided content into the output files.
Capability inventory: The script performs local file and directory creation (Path.mkdir, json.dump, open().write()).
Sanitization: No sanitization, escaping, or validation is performed on the content of the description or author strings.

plugin-builder