lead-routing

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through the lead scoring mechanism.
  • Ingestion points: Untrusted external data (such as company name, job title, or behavioral data) is ingested via the {lead_data} variable in the ai_scoring prompt defined in SKILL.md.
  • Boundary markers: The scoring_prompt does not use clear delimiters or boundary markers (e.g., triple quotes, XML tags) to isolate the untrusted {lead_data} from the system instructions, increasing the risk that an attacker could embed malicious instructions in lead data to manipulate the resulting score or routing decision.
  • Capability inventory: The skill has the capability to modify CRM records (hubspot_assign_owner, salesforce_route), create tasks, and send Slack notifications based on the output of the AI scoring process.
  • Sanitization: There is no evidence of input sanitization or filtering to prevent the processing of instructions hidden within the lead data fields.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 05:18 AM