The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data including user requirements, FHIR resource instances, and code from repository files (in 'scoped' mode).
Ingestion points: User-supplied requirements and API designs in SKILL.md; arbitrary file content in scoped review mode.
Boundary markers: The skill includes an explicit 'Prompt injection boundary' section that instructs the agent to treat all external content as data to be analyzed rather than instructions to follow, and to flag any detected directives to the user.
Capability inventory: The skill is limited to advisory output. It generates design recommendations, HTTP query examples, and JSON snippets. It has no capabilities to modify files, execute commands, or perform network operations outside of its standard agent environment.
Sanitization: No programmatic sanitization is defined, but the skill relies on instructional constraints to prevent accidental execution of embedded directives.
[PROMPT_INJECTION]: A static detector hit for instruction override patterns was analyzed and found to be a false positive. The phrase 'ignore previous instructions' appears within a defensive 'Prompt injection boundary' section as an example of what the agent should identify and report to the user, representing a security best practice rather than an attack vector.

health-fhir-api-design