define-customer-profile
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill collects untrusted user input during the Discovery phase and interpolates it directly into prompts for secondary research agents (Phase 2) without sanitization or boundary markers.
- Ingestion points: User-provided responses to discovery questions (e.g., product type, problem space) in Phase 1.
- Boundary markers: Absent. The user input is concatenated directly into the task string for sub-agents.
- Capability inventory: The skill launches general-purpose agents (Claude 3 Opus) equipped with
WebSearchcapabilities. - Sanitization: None detected. Malicious user input could potentially influence the behavior of the research agents or their search queries.
- [DATA_EXPOSURE]: The skill writes detailed logs of the discovery process to a world-readable directory, which could expose proprietary business logic or product ideas on multi-user systems.
- Evidence: Logs are saved to
/tmp/customer-discovery-{YYYYMMDD-HHMMSS}.md. - [DYNAMIC_EXECUTION]: The skill dynamically constructs tasks for sub-agents using string interpolation of user data.
- Evidence: Prompts for 'ICP Pattern Researcher' and 'Competitor/Market Researcher' in Phase 2 are built using the
[product type]and[problem space]variables derived from user input.
Audit Metadata