The Agent Skills Directory

[METADATA_POISONING]: The skill uses deceptive metadata by claiming to be an "Official Anthropic skill" and setting the "Author" field to "Anthropic" in its technical details. These claims are inconsistent with the provided installation source and repository location (Open-Medica), which could mislead users regarding the skill's origin, ownership, and safety verification status.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted clinical data and payer policies to generate summaries and appeal letters, creating a risk for indirect prompt injection attacks.
Ingestion points: Processes medical documentation, insurance coverage policies, and patient treatment records (SKILL.md).
Boundary markers: The instructions lack explicit delimiters or "ignore embedded instructions" warnings to prevent the agent from obeying malicious commands potentially hidden within the clinical data or policy documents.
Capability inventory: The skill generates clinical narratives, appeal letters, and structured talking points for insurance reviews (SKILL.md).
Sanitization: No input sanitization or validation routines for external clinical data are described or present in the instructions.
[EXTERNAL_DOWNLOADS]: The documentation provides an installation command (npx skills add Open-Medica/open-medical-skills) to fetch the skill from the Open-Medica GitHub repository.

prior-authorization-review