The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to configure Git-based dependencies targeting https://github.com/Dicklesworthstone/asupersync within Cargo.toml. This allows the download and integration of external code from a source outside the predefined trusted vendor list.
[INDIRECT_PROMPT_INJECTION]: The skill documentation includes patterns for building web servers that ingest and process external network data, creating a surface for indirect prompt injection.
Ingestion points: The create_item handler and to_web_request function in SKILL.md extract untrusted data from HTTP request bodies (asupersync::web::extract::Json) and headers.
Boundary markers: The instructions do not define explicit prompt delimiters or "ignore instructions" markers for the processed data, although they mention protocol-level constraints.
Capability inventory: The skill facilitates building Rust applications with capabilities for network listening (Http1Listener), network requests, and standard file/system access inherent to the Rust runtime.
Sanitization: The skill recommends best practices such as HostPolicy::allow_list and bounding request metadata, but lacks specific sanitization logic for data passed to an LLM context.

asupersync-rust