title-variants

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted video titles and summaries from external Google Sheets or JSON files and interpolates them directly into a prompt for the Claude LLM in scripts/generate_title_variants.py. An attacker who can influence the input sheet or file could inject instructions to manipulate the LLM's output.
  • Ingestion points: Video titles and summaries are read from Google Sheets via read_sheet_data() or from local JSON files.
  • Boundary markers: None. Input text is directly formatted into the prompt string.
  • Capability inventory: The skill can perform network operations via the Google Sheets API and create/update documents in the user's Google Drive.
  • Sanitization: No sanitization or escaping is performed on the input titles or summaries before they are sent to the LLM.
  • [COMMAND_EXECUTION]: The skill uses scripts/generate_title_variants.py and scripts/update_sheet.py which are designed to be executed as subprocesses by the agent. While these scripts perform intended logic (Google Sheets integration), they require broad OAuth scopes (https://www.googleapis.com/auth/spreadsheets, https://www.googleapis.com/auth/drive) which grant the skill significant control over the user's Google workspace environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 02:18 PM