service-area-pages

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses local Python scripts to manage brand context and perform quality assurance checks. It executes brand_context.py to synchronize marketing data and uses a provided overlap.py script to detect sentence-level similarity across generated pages. Additionally, curl commands are documented for technical audits of image sizes and server redirect status codes.\n- [INDIRECT_PROMPT_INJECTION]: The skill research phase involves scraping external city and competitor websites. This creates a surface where the agent processes untrusted data from the web. \n
  • Ingestion points: Data from third-party websites (e.g., city permit pages, water reports) is fetched via Firecrawl or WebFetch.\n
  • Boundary markers: There are no explicit instructions to use delimiters around external data in the prompt construction phase.\n
  • Capability inventory: The skill writes markdown files to the local file system and performs network requests for research and auditing.\n
  • Sanitization: The skill relies on factual verification and a 30% overlap threshold for duplicate detection to maintain content integrity, though it does not provide specific security sanitization for adversarial instructions embedded in the scraped data.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 04:24 PM