ContentGuardSource

Namespace Tracon · Assembly Tracon.Abstractions.dll

Where the text handed to an IContentGuard came from.

public enum ContentGuardSource

Fields

Document = 3

Text extracted from a document (for example, retrieved for RAG).

ModelOutput = 4

Text the model itself generated.

SkillResource = 5

Text loaded from a skill’s own resource file.

ToolResult = 2

The result of a tool call. See ContentGuardContext.ToolName.

Unknown = 0

The source could not be determined. This is never a reason to relax a security decision — treat it at least as strictly as the most sensitive known source.

UserMessage = 1

Text a user typed.

Remarks

The trust level behind these values is not uniform: a user message can only poison the sender’s own session, while a tool result can carry text another tenant’s data wrote into a shared system (prompt injection). A guard that wants to apply different rules per source reads this value; the built-in PatternContentGuard deliberately does not — it applies the same patterns regardless of source.