pricing-quarterly-review

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that ingests untrusted data from external sources, specifically customer quotes and CRM records.
  • Ingestion points: CRM data exports, including "lost-to-price verbatim quotes" and billing system distributions (referenced in the 'Core data pull' and 'Workflow' sections).
  • Boundary markers: The instructions lack explicit delimiters or "ignore embedded instructions" directives to prevent the agent from following malicious commands hidden in the CRM quotes.
  • Capability inventory: The skill uses Read and Write tools to modify project configuration files (brand-config.json).
  • Sanitization: There is no evidence of sanitization, filtering, or escaping for the ingested external strings before they are logged or included in the review agenda.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 10:28 PM
Security Audit — agent-trust-hub — pricing-quarterly-review