propensity-scoring-realtime

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input from conversation messages to calculate propensity scores using regex-based pattern matching. This creates a surface where an attacker can craft specific phrases to manipulate their score, potentially triggering high-priority routing or alerting sales representatives.
  • Ingestion points: The update_score_from_message and analyze_message_signals functions in SKILL.md process raw message content.
  • Boundary markers: No boundary markers or 'ignore instructions' delimiters are used when processing message content.
  • Capability inventory: The resulting scores directly influence notify_rep, accelerate_outreach, and route_prospect actions.
  • Sanitization: The implementation relies on basic regex matching without semantic verification or filtering for manipulation attempts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 04:05 PM
Security Audit — agent-trust-hub — propensity-scoring-realtime