AI models output standard markdown, but messaging platforms use their own formatting syntax. The IMessageFormatter interface provides a per-channel normalization pipeline that converts markdown to platform-native formatting before messages are sent.

WhatsApp uses custom delimiters instead of standard markdown:

IMessageFormatter (Lis.Core)
    ↑
    └── WhatsAppFormatter (Lis.Channels/WhatsApp)
        └── Injected into WhatsAppClient via DI

AI Response → ResponseDirectives.Parse() → IMessageFormatter.Format() → GOWA → WhatsApp
                                                   ↑
                                          WhatsAppFormatter

The formatter runs inside WhatsAppClient.SendMessageAsync(), so:

• The DB stores the original markdown (better for AI context window)
• WhatsApp receives the platform-formatted version

Code blocks (fenced and inline) are extracted into placeholders before any transformations run, then restored at the end. This ensures formatting inside code examples is never modified.

1. Create a class implementing IMessageFormatter in the channel's namespace
2. Register it in the channel's Add*() setup method
3. Inject into the channel's IChannelClient implementation

Example for a hypothetical Telegram channel:

public sealed class TelegramFormatter : IMessageFormatter
{
    public string Format(string content) { /* ... */ }
}