nhcx-payer-coverage

The security analysis identified several functional patterns that fall within categorized threat vectors, primarily driven by the skill's core purpose of external data integration.

• [EXTERNAL_DOWNLOADS]: The skill contains instructions in SKILL.md and references/CORE.md to periodically fetch updates from the author's GitHub repository at https://raw.githubusercontent.com/nha-in/nhcx-package. While these originate from the skill author and use SHA256 checksums for validation, they involve fetching instruction files at runtime.
• [DYNAMIC_EXECUTION]: The skill uses a self-modification pattern where it replaces its own local instruction files (SKILL.md and related markdown files) when a newer version is detected on GitHub. Additionally, the skill's workflow (Step L1.2 and LOGH) requires the agent to generate and then execute a local script (e.g., report.py) to build a progress report. This involves the execution of generated code, which is noted as a medium-risk pattern but lowered here as it is a core feature of the development workflow.
• [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process FHIR bundles from external, untrusted hospital systems, creating an attack surface for indirect prompt injection.
• Ingestion points: gateway/G8-receive.md defines the HTTP endpoints (/in/, /v1/) where the system receives POST requests from NHCX containing messages from external hospitals.
• Boundary markers: The specifications do not mandate the use of strict delimiters or explicit 'ignore instructions' warnings when clinical data (such as adjudication_remarks or description) is interpolated into internal processes or previews.
• Capability inventory: The skill possesses capabilities to perform network exfiltration via gateway/G7-send.md (HTTP POST) and persistent data modification through the database scripts defined in the services/ and callbacks/ directories.
• Sanitization: There is no explicit requirement in the provided files for the sanitization of clinical text fields to prevent instructions embedded within medical data from influencing agent behavior.