FHIR Context With Agent Skills

Agent Skills that that use FHIR APIs can indicate that they require FHIR context. This is primarily handled by metadata field of the Agent Skills specification.

Metadata Declaration

For agent skills that are configured with FHIR context the frontmatter metadata will look like:

The scopes within the po-fhir-context-scopes should indicate SMART scopes that your script requires. The user that adds the skill can then decide to authorize the agent with the requested access. You can mark a scope as required by supplying it under the po-fhir-context-required-scopes field. When marking a scope as required, a user cannot uncheck the scope and their only option is to avoid using the skill at all if they do not consent to giving the required access.

Here is a sample metadata declaration for an agent skill that has a required scope of patient/MedicationRequest.rs (which allows reading and searching on the current patient MedicationRequest resources) and an optional scope of patient/Condition.rs (which allows reading and searching for conditions for the current patient).

---
metadata:
  po-fhir-context-required-scopes:
    patient/MedicationRequest.rs
  po-fhir-context-scopes:
    patient/Patient.rs patient/Condition.rs
---

FHIR Context Environment Variables

When the FHIR context is configured Prompt opinion will pass the access token and the FHIR url as environment variables:

FHIR_BEARER_TOKEN = access token restricted to the requested scopes

FHIR_BASE_URL = the FHIR url against which the requests will be executed

Python usage example:

    token = os.environ.get("FHIR_BEARER_TOKEN")

    fhir_url = os.environ.get("FHIR_BASE_URL")