user-profile-interview
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted input from the user during the conversational interview and reads data from local system files without explicit sanitization or boundary markers.
- Ingestion points: User messages (Phase 1 & 2) and the
agent_file_system/USER.mdfile (read viaread_file). - Boundary markers: Absent. The instructions do not define delimiters or tags to isolate untrusted user content from agent instructions during processing.
- Capability inventory: The skill uses
stream_editto modify system configuration files (USER.mdandAGENT.md) andsend_messageto communicate with the user. - Sanitization: Absent. The agent is instructed to write user-supplied goals and derived personality observations directly into markdown files as free-form text or paragraphs without validation or escaping.
Audit Metadata