customer-research

Pass

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill connects to an external Model Context Protocol (MCP) server at https://mcp.unifapi.com to fetch data.
  • [EXTERNAL_DOWNLOADS]: Instructions for installation involve using npx to add packages from the unifapi-agent repository, which is the official source for this skill.
  • [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection because it ingests and processes untrusted content from public forums (Reddit, TikTok comments).
  • Ingestion points: Specifically fetches data using reddit/posts/{id}/comments and tiktok/videos/{id}/comments (referenced in SKILL.md).
  • Boundary markers: The skill uses structured output templates but lacks specific instructions to ignore embedded commands within the fetched social media text.
  • Capability inventory: The skill is restricted to read-only operations and does not possess capabilities for file writing, arbitrary command execution, or network exfiltration beyond the vendor API.
  • Sanitization: No explicit sanitization or filtering of the external social media content is described in the provided scripts.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 2, 2026, 10:28 AM
Security Audit — agent-trust-hub — customer-research