The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill helps developers build applications that ingest untrusted data from the UHI network, such as hospital catalogs (on_search) and service terms (on_init). This creates an attack surface for indirect prompt injection if the agent were to process these fields without sanitization.
Ingestion points: Catalogs in on_search and terms in on_init (documented in references/scaffold.md and references/design.md).
Boundary markers: The skill describes structured data fields but does not explicitly instruct the agent to use boundary markers when processing natural language content within those fields.
Capability inventory: The skill provides technical documentation and curl examples; it does not directly possess filesystem-write or arbitrary command execution capabilities.
Sanitization: The documentation does not specify sanitization or validation rules for natural language fields ingested from external providers.
[EXTERNAL_DOWNLOADS]: The skill instructs users to clone and run the 'Header Generation Utility' from the official NHA-ABDM GitHub repository (https://github.com/NHA-ABDM/UHI/tree/main/header_generator_utility). This is a resource provided by the National Health Authority of India for implementing the UHI protocol.
[CREDENTIALS_UNSAFE]: The skill provides instructions for handling Ed25519 signing keys and patient PINs. It includes security best practices, explicitly warning developers to keep private keys in server-side secret storage and to keep PINs in memory only, avoiding databases and logs.
[OBFUSCATION]: A Base64 string is included in references/scaffold.md as a placeholder for an agent's photo in a sample API request. Technical analysis of the decoded content shows it to be benign binary data consistent with an image placeholder.

uhi-consultation