The Agent Skills Directory

[SAFE]: No malicious patterns such as credential theft, remote code execution, or unauthorized data exfiltration were detected.
[PROMPT_INJECTION]: The skill includes a nudge for the agent to use tool read_file to ensure instructions are followed accurately; this is a benign instructional technique.
[PROMPT_INJECTION]: The skill is designed to ingest data from user workspace files. Evidence: 1. Ingestion points: User workspace files; 2. Boundary markers: None; 3. Capability inventory: tool create_website, tool read_file; 4. Sanitization: Not mentioned. This surface is part of the core functionality of referencing workspace content.

create-site