social-intent-monitoring
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a system for monitoring and processing untrusted public social media data (LinkedIn, X, Reddit) to trigger outreach actions. This architectural design is vulnerable to indirect prompt injection, where malicious actors could embed instructions in public posts to manipulate the agent's output or behavior during lead qualification.
- Ingestion points: Public social conversations and mentions monitored via external tools (SKILL.md).
- Boundary markers: The instructions lack explicit delimitation or instructions for the agent to distinguish between data and instructions within the social signals.
- Capability inventory: The skill routes signals to outreach tools and CRMs, requiring the agent to generate personalized messages based on untrusted content (SKILL.md).
- Sanitization: No sanitization or safety-filtering logic is described for the content being processed.
Audit Metadata