facebook-marketplace-seller-communication
Pass
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text from Facebook Marketplace listings and seller replies, creating a potential vector for malicious instructions to influence agent behavior.\n- Ingestion points: External text is ingested when opening listing URLs and reading seller messages (Procedure steps 1 and 7 in SKILL.md).\n- Boundary markers: Explicit delimiters to isolate untrusted seller content from agent instructions are absent.\n- Capability inventory: Capabilities include browser interaction via
browser-harnessand terminal command execution.\n- Sanitization: No content sanitization or validation logic is specified for ingested data.\n- Mitigation: The design requires manual user approval for every drafted message, providing a critical human-in-the-loop control.
Audit Metadata