The Agent Skills Directory

[SAFE]: The skill performs localized file operations consistent with its stated purpose as a package manager. No malicious behavior such as data exfiltration, unauthorized command execution, or persistence mechanisms was detected.\n- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting documentation from SKILL.md files referenced in the marketplace manifest. While inherent to the skill's purpose, the following factors were identified:\n
Ingestion points: Data is ingested from SKILL.md files located at paths defined in the local marketplace.json file.\n
Boundary markers: Documentation content is read and returned to the agent without specific delimiters or instructions to ignore embedded commands.\n
Capability inventory: The skill is restricted to reading local files and writing to a specific JSON file (installed_skills.json) within its own directory.\n
Sanitization: No sanitization or validation of documentation content is performed before presentation to the agent.

skill-installer