healthcare-providers-verify

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime LLM context can ingest outsider-authored free text via nimble search + nimble extract when it falls back to web search and extracts NPI/provider pages (and potentially other web sources) into data.markdown, which is then used for verification and reported back to the model.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly calls the NPPES API at runtime via nimble extract ("https://npiregistry.cms.hhs.gov/api/?...") and falls back to npidb.org and nppes.cms.hhs.gov provider pages as required data sources, which are runtime external fetches whose returned content directly drives verification decisions and injected model context.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The prompt is mostly limited to benign user-directory file reads/writes and calling the nimble CLI, but it explicitly instructs sub-agents to spawn with "bypassPermissions", which is an instruction to bypass permission/security boundaries and therefore risks compromising the host state.