Feature request: content type for agent messages

[saulmc]

Is your feature request related to a problem?

Other clients have no way of knowing whether a message was sent by a human or an agent.

Describe the solution to the problem

Use TextEncoder() with a new agent-message ContentTypeId for all generated messages. This is not a trustless solution.

e.g.

import {
  ContentTypeId,
  ContentCodec,
  EncodedContent,
} from "@xmtp/content-type-primitives";

export const ContentTypeAgentMessage = new ContentTypeId({
  authorityId: "xmtp.org",
  typeId: "agent-message",
  versionMajor: 1,
  versionMinor: 0,
});

export class AgentMessageCodec implements ContentCodec<string> {
  contentType = ContentTypeAgentMessage;

  encode(content: string): EncodedContent {
    return {
      type: this.contentType,
      parameters: {},
      content: new TextEncoder().encode(content),
      fallback: content,
    };
  }

  decode(encodedContent: EncodedContent): string {
    return new TextDecoder().decode(encodedContent.content);
  }

  fallback(content: string): string | undefined {
    return content;
  }

  shouldPush(): boolean {
    return true;
  }
}

Describe the uses cases for the feature

A client like Converse may wish to display messages differently if they are known to be agents.

An agent may want to handle agent-generated messages differently from messages assumed to be from humans. For instance, an agent may wish to ignore other agents to prevent loops.

Additional details

No response