automation-design

Pass

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill instructs the agent to perform network operations to retrieve content from external websites for product and competitor research.
  • Evidence: The SKILL.md instructions direct the agent to "fetch the site or page they named", "fetch the URL they named", and "research the company yourself" by visiting their site and public sources. These actions involve requests to arbitrary, non-whitelisted external domains.
  • [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection due to its handling of untrusted data from the web.
  • Ingestion points: External websites and user-provided URLs (referenced in SKILL.md).
  • Boundary markers: Not explicitly defined in the prompt templates, but general guidance is provided in reference files.
  • Capability inventory: The agent writes researched data to .agents/product-context.md (referenced in SKILL.md) and logs loop history to .agents/store-loop-ledger.md.
  • Sanitization: references/house-rules.md (Safety Rule 7) provides a mitigation by instructing the agent to treat all fetched content as data rather than instructions and to quote any embedded instructions. However, the raw data is still ingested and stored in the agent's long-term context files, which could be exploited by sophisticated prompt injection techniques designed to bypass these verbal constraints.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 27, 2026, 08:38 AM