competitive-positioning

Pass

Audited by Gen Agent Trust Hub on Mar 8, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface through its research functionality.
  • Ingestion points: Data enters the agent context via WebSearch and WebFetch when analyzing competitor websites and public review forums like G2 or Capterra.
  • Boundary markers: The skill does not implement delimiters or safety instructions to separate untrusted external data from its own core logic, which could lead to the agent following instructions found on a competitor's site.
  • Capability inventory: The skill is granted access to file-writing tools (Write, Edit) and network tools (WebFetch), increasing the potential risk if a successful injection occurs.
  • Sanitization: There is no evidence of content filtering or validation for external data before it is processed by the model.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 8, 2026, 04:30 PM