attending-physician
Warn
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: MEDIUMPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [PERSISTENCE]: The skill provides instructions in SKILL.md and references/platform.md to establish persistence by appending activation commands to tool-specific configuration files, specifically targeting ~/.claude/CLAUDE.md, ~/.cursor/rules/, and .clinerules.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted clinical data provided by users, creating a surface for indirect prompt injection attacks.
- Ingestion points: User-provided patient histories, laboratory results, and clinical scenarios mentioned in the role definition (SKILL.md §1.1) and example scenarios (references/scenarios.md).
- Boundary markers: None identified; there are no explicit delimiters or instructions to the agent to treat patient data as untrusted content.
- Capability inventory: The persona generates diagnostic reasoning, treatment plans, and documentation-ready statements based on input data (SKILL.md §1.1).
- Sanitization: No evidence of input validation or sanitization for user-provided clinical scenarios.
Audit Metadata