//! Wrap sanitized content into an `<untrusted_content>` envelope.
//!
//! The envelope is the contract between AgentSec and the caller LLM: text
//! inside the envelope is treated as **data, not instruction**. The
//! attributes carry provenance metadata so the LLM can reason about where
//! the content came from and what was already stripped.
use Utc;
/// Wrap `body` in an `<untrusted_content>` element with provenance
/// attributes.
///
/// Schema:
///
/// ```xml
/// <untrusted_content
/// src="<url>"
/// sanitized_at="<RFC 3339 UTC>"
/// removed_patterns="<comma-separated pattern labels>">
/// <body...>
/// </untrusted_content>
/// ```
///
/// `removed` may be empty; the attribute is still emitted as an empty
/// string so downstream parsers can rely on the schema being uniform.