The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill consumes external data via scripts/read_prospect.py from Google Sheets (sheet_url). The ingested records are mapped directly to properties like description, keywords, and company_name. This untrusted context flows directly into scripts/generate_website.py to populate localized textual fields. If an attacker controls the source sheet rows, they could inject malicious instructions intended to manipulate the agent processing this workflow or compromise the output context.
[COMMAND_EXECUTION]: The execution path uses user-supplied strings inside simple command line piping (python3 read_prospect.py --url "SHEET_URL" --row 1 | python3 generate_website.py). The parameters are handled safely via standard Python argparse arguments, rather than subshells, which restricts direct command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The script connects to the external Google Sheets API via gspread and dynamically requests images from api.unsplash.com and picsum.photos. These are well-known, trusted domains for standard stock image curation and OAuth flow authentication, presenting low inherent malicious risk.

design-website