recognizing-deception

Pass

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill describes using curl to fetch content from defuddle.md, an external markdown extraction service. This is used for reading public security advisories and vendor reports. The author includes specific instructions to avoid routing sensitive infrastructure or engagement-related URLs through this service to prevent information leakage.
  • [PROMPT_INJECTION]: The skill instructs the agent to fetch remote content, creating a surface for indirect prompt injection.
  • Ingestion points: Remote content fetched via curl in SKILL.md.
  • Boundary markers: Absent.
  • Capability inventory: The skill specifies read-only access to remote documentation; no write or execution capabilities are defined in this skill.
  • Sanitization: The defuddle.md service is described as stripping boilerplate text, which provides a degree of implicit filtering for the fetched content.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 27, 2026, 08:24 PM
Security Audit — agent-trust-hub — recognizing-deception