content-strategy

Pass

Audited by Gen Agent Trust Hub on Apr 2, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by instructing the agent to process and analyze content from untrusted external origins.
  • Ingestion points: The skill explicitly processes customer call transcripts, survey responses, and third-party forum content (Reddit, Quora) retrieved via web search in the 'Ideation Sources' section of SKILL.md.
  • Boundary markers: There are no instructions to use delimiters or ignore embedded commands within the ingested data, increasing the risk that malicious instructions in the source text could influence agent behavior.
  • Capability inventory: The skill is designed to work with agents that typically have web search and file system access to retrieve context from MEMORY.md and USER.md.
  • Sanitization: No sanitization, validation, or filtering of the external content is specified before the agent is asked to extract patterns and insights.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 2, 2026, 07:02 AM
Security Audit — agent-trust-hub — content-strategy