answer-engine-content-audit

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses curl -s "$URL" and curl -s "$SITE/sitemap.xml" within bash scripts defined in references/checks.md. While the script uses double quotes to prevent basic word splitting, a malicious user could provide a URL that triggers unintended behavior if the environment variable is not strictly validated by the agent platform.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process raw HTML from external websites using curl and grep. This creates a surface for indirect prompt injection where a target website could serve malicious content (hidden instructions in headings or meta tags) designed to influence the agent's audit findings or subsequent actions. The skill lacks explicit boundary markers to delimit this untrusted content.\n
  • Ingestion points: External URLs provided by the user are fetched via curl in references/checks.md.\n
  • Boundary markers: None identified. The raw text is piped directly into python3 and grep.\n
  • Capability inventory: The skill uses curl for network access and python3 for text processing.\n
  • Sanitization: The skill uses regex-based cleaning (re.sub('<[^<]+?>', ' ', html)) to remove tags, which provides some protection against HTML-based attacks but does not sanitize the remaining natural language instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 05:54 AM
Security Audit — agent-trust-hub — answer-engine-content-audit