sequence-performance

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it is designed to ingest and analyze untrusted content from external email recipients.\n
  • Ingestion points: In Step 1 (Pull Campaign Data) and Step 3 (Reply Analysis), the skill fetches and reads the reply_text from external outreach tools like Smartlead and Instantly.\n
  • Boundary markers: The instructions in Step 3 for 'Reply Classification' and 'Objection Pattern Analysis' do not define boundary markers or instruct the agent to ignore potentially malicious instructions embedded within the reply text.\n
  • Capability inventory: The skill produces reports and passes data to other skills such as email-drafting, creating a chain where a malicious reply could influence downstream drafting or outreach actions.\n
  • Sanitization: No explicit sanitization, filtering, or validation of the retrieved reply content is mentioned in the sequence of operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 01:31 AM
Security Audit — agent-trust-hub — sequence-performance