The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements middleware that ingests untrusted data from HTTP headers (e.g., X-Request-ID, Authorization) which is then used in logging and downstream authentication logic. This creates a surface for log injection or indirect prompt injection if the logs are consumed by an LLM or other automated systems.
Ingestion points: Header extraction in RequestTrackingMiddleware (SKILL.md, SKILL_ZH.md) and auth_middleware (SKILL.md).
Boundary markers: No explicit sanitization or boundary markers are applied to header values before they are passed to log::info!.
Capability inventory: The skill utilizes logging (log::info!), request extensions, and HTTP response generation.
Sanitization: Values are converted to strings using to_str() but are not validated for malicious control characters or injection patterns.
[SAFE]: The skill demonstrates best practices for CORS in its main documentation (SKILL.md), including strict origin validation for production environments. Note that the Chinese translation (SKILL_ZH.md) contains a simplified implementation that defaults to allowing all origins; users should follow the logic provided in the primary SKILL.md file.
[SAFE]: All referenced crates (e.g., actix-web, axum, tokio, uuid, jsonwebtoken) are well-known, reputable packages in the Rust ecosystem. No suspicious remote code execution or persistence mechanisms were identified.

rust-middleware