The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external web APIs (api.coingecko.com, httpbin.org), which introduces a surface for indirect prompt injection. Malicious data returned by these external services could potentially influence the agent's behavior if not correctly delimited from instructions.
Ingestion points: The response.body is read and processed in the Motoko getIcpPriceUsd and Rust fetch_price examples (SKILL.md).
Boundary markers: The code examples do not explicitly implement boundary markers or instructions to ignore embedded prompts within the retrieved HTTP body.
Capability inventory: The skill utilizes Call.httpRequest (Motoko) and ic_cdk::management_canister::http_request (Rust) to perform network operations and can influence canister state based on the results.
Sanitization: While the examples show parsing JSON data, there is no explicit sanitization of the content to prevent instructions embedded in the external data from being interpreted by the agent.
[EXTERNAL_DOWNLOADS]: The skill documents and provides code for fetching data from external domains (api.coingecko.com and httpbin.org). These are well-known services commonly used for financial price data and HTTP request testing respectively, and the network operations trace back to the vendor's documented platform functionality.

https-outcalls