comment-mining

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is designed to process untrusted user-generated content from third-party social media platforms, which constitutes a surface for indirect prompt injection. If malicious instructions are embedded in the comments being mined, they could potentially influence the agent's behavior.
  • Ingestion points: Comments and replies fetched via specified endpoints for TikTok, YouTube, Instagram, Facebook, Reddit, and Rumble.
  • Boundary markers: The instructions do not define clear delimiters or specific warnings to the agent to ignore or isolate instructions contained within the scraped content.
  • Capability inventory: The skill is configured with Bash, WebFetch, Read, and Write capabilities, increasing the potential impact of a successful injection.
  • Sanitization: No explicit prompt-level sanitization or escaping of external content is implemented; the 'clean lightly' step focuses on spam and duplicate removal rather than security filtering.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 07:55 AM
Security Audit — agent-trust-hub — comment-mining