The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements endpoints and event handlers designed to ingest untrusted data from external sources, creating a surface for indirect prompt injection.
Ingestion points: The readBoundedJson function in references/server-boundary.md parses user-supplied HTTP request bodies, and the onEvent handler in references/realtime.md processes events emitted from the blockchain.
Boundary markers: The skill includes explicit warnings in the documentation (e.g., "Do not feed payload text to an agent as trusted instructions") and architectural advice to verify the application boundary.
Capability inventory: The skill utilizes DurablePostWriter.executeCreate for state persistence and createServerReader for authenticated network requests using a server-side access key.
Sanitization: While parsePost performs structural validation of the JSON payload, the documentation explicitly notes that text content is not sanitized and requires application-level handling.

arkiv-app-integration