skeleton-ad
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process external content from URLs and user-provided links to research products and brands. This data ingestion creates a surface for indirect prompt injection, where malicious instructions could be embedded in the retrieved web content. Evidence includes the instruction to fetch product pages for research and the inclusion of
references/agent-security.mdwhich provides rules to treat untrusted content as data and not as instruction. Capability inventory includes writing to the local.agents/directory and performing network requests to fetch content. Sanitization requirements include HTML escaping for interpolated values and redaction of sensitive tokens. - [PROMPT_INJECTION]: The instructions include a specific directive to avoid returning drafts with warnings or caveats, stating "Do not return a draft with a warning on it." Although intended to ensure high-quality, professional output, this instruction explicitly mandates the suppression of cautionary messages that might otherwise inform the user of missing context, errors, or potential risks identified during the generation process.
Audit Metadata