The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle remote HTTP responses, which are untrusted external inputs. The instructions and code examples include robust security measures to mitigate risks associated with untrusted data.
Ingestion points: The HttpClient::get_text method in examples/golden-http-client/src/lib.rs and general instructions for handling API responses in SKILL.md define how external data is read into the application context.
Boundary markers: SKILL.md explicitly mandates restricting schemes, hosts, ports, and redirect targets for user-controlled destinations to prevent SSRF and other injection attacks.
Capability inventory: Capabilities are restricted to standard outbound HTTP network operations via the reqwest library.
Sanitization: The provided example in lib.rs strictly enforces body size limits using .take(self.max_body_bytes + 1) and validates UTF-8 integrity before processing data as strings.
[EXTERNAL_DOWNLOADS]: The skill manages a standard set of Rust dependencies via Cargo.toml and Cargo.lock, fetching well-known and established libraries from the official crates.io registry.
[COMMAND_EXECUTION]: The aws-lc-sys dependency, used for cryptographic operations, invokes cmake and cc during the compilation phase to build native libraries. This is standard and expected behavior for low-level cryptographic crates in the Rust ecosystem.

rust-http-client