The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/fhir_package_manager.py utility fetches FHIR terminology and profile packages from the official industry registry at https://packages.fhir.org. It handles metadata retrieval via HTTPS and extracts package archives (tar/zip) into a local cache directory (~/.fhir/packages/).
[COMMAND_EXECUTION]: The skill documentation and assets reference the use of various command-line tools for FHIR Implementation Guide (IG) development, including fsh-sushi and gofsh. It also mentions the execution of shell scripts (_genonce.sh, _updatePublisher.sh) which are standard parts of the HL7 FHIR toolchain for rendering documentation.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes external FHIR resources that could contain malicious natural language instructions.
Ingestion points: FHIR resources are ingested via scripts/fhir_package_manager.py (downloaded packages) and assets/fhir_server.py (HTTP API inputs).
Boundary markers: Standard JSON parsing is used; however, no explicit instructions are provided to the model to ignore embedded natural language within the resource fields.
Capability inventory: The skill has capabilities for local file system access (to manage the package cache) and outbound network requests.
Sanitization: The skill relies on structural validation using Pydantic models (fhir.resources) and TypeScript interfaces. This validates the data schema but does not sanitize potential natural language instructions embedded in string fields.

fhir-software