feat(ai-moderation): use per-server nickname (displayName) in analysis payload

- resolveDisplayName(): member.displayName from captured metadata,
  falls back to global username
- Applied to context lines, target message blocks, and media message
  blocks — LLM sees the name the channel actually sees (nickname can
  carry moderation signal itself)
This commit is contained in:
asepharyana
2026-08-10 11:36:37 +07:00
parent 4049ab4201
commit ecbb538c9f
5 changed files with 52 additions and 3 deletions
@@ -6,6 +6,7 @@ import {
} from "../message-capture/messageMetadata.js";
import type { MessageRecord } from "../message-capture/types.js";
import { sanitizeDiscordTokens } from "./discordTokens.js";
import { resolveDisplayName } from "./moderationBuilders.js";
const logger = createChildLogger("conversationContext");
@@ -130,7 +131,7 @@ export function formatMessageForPrompt(
const mediaEvidence = formatMediaEvidenceForPrompt(msg.metadata);
const mediaSuffix = mediaEvidence ? ` ${mediaEvidence}` : "";
const refInfo = formatReferenceInfo(msg);
return `[${label}] id=${msg.id} time=${timestamp} user=${msg.username}: ${content}${mediaSuffix}${refInfo}`;
return `[${label}] id=${msg.id} time=${timestamp} user=${resolveDisplayName(msg)}: ${content}${mediaSuffix}${refInfo}`;
}
/**
@@ -36,6 +36,27 @@ export function getAnalysisContent(message: MessageRecord): string {
).trim();
}
/**
* Server nickname (member.displayName) when captured, else the author
* username. Discord shows the server nickname to other members, so the LLM
* should see the same name the channel sees — and a nickname can carry
* moderation signal itself (offensive nick + clean message → low warn).
*/
export function resolveDisplayName(msg: MessageRecord): string {
if (msg.metadata) {
try {
const meta = JSON.parse(msg.metadata) as {
member?: { displayName?: string | null } | null;
};
const dn = meta?.member?.displayName;
if (dn && dn.trim().length > 0) return dn;
} catch {
// malformed metadata — fall back to username
}
}
return msg.username;
}
/**
* Builds a <reference> XML element for reply/forward/crosspost context.
*/
@@ -21,6 +21,7 @@ import {
buildReferenceXml,
escapeXml,
getAnalysisContent,
resolveDisplayName,
} from "./moderationBuilders.js";
import {
buildSystemPrompt as buildSystemPromptModular,
@@ -308,7 +309,7 @@ export async function runTextOnlyBatch(
const userCtx = userContexts.get(msg.user_id) ?? "";
const userProfileCtx = userProfiles.get(msg.user_id) ?? "";
const refXml = await buildReferenceXml(msg);
return `<message id="${msg.id}" user="${msg.username}">\n ${userCtx}${userProfileCtx ? `\n ${userProfileCtx}` : ""}${refXml ? `\n ${refXml}` : ""}\n <content>${escapeXml(content)}</content>${webContext}${mediaEvidenceCtx}\n</message>`;
return `<message id="${msg.id}" user="${resolveDisplayName(msg)}">\n ${userCtx}${userProfileCtx ? `\n ${userProfileCtx}` : ""}${refXml ? `\n ${refXml}` : ""}\n <content>${escapeXml(content)}</content>${webContext}${mediaEvidenceCtx}\n</message>`;
}),
)
).join("\n");
@@ -39,6 +39,7 @@ import {
buildReferenceXml,
escapeXml,
getAnalysisContent,
resolveDisplayName,
} from "./moderationBuilders.js";
import {
buildCustomEmojiVisionPrompt,
@@ -367,6 +368,6 @@ export async function prepareMediaMessage(
const profile = await getUserProfile(target.user_id);
const refXml = await buildReferenceXml(target);
const messageBlock = `<message id="${escapeXml(target.id)}" user="${escapeXml(target.username)}">\n <user_reputation trust_score="${rep.trust_score}" />${profile ? `\n <user_profile>${sanitizeAiContent(profile.profile_summary)}</user_profile>` : ""}${refXml ? `\n ${refXml}` : ""}\n <content>${escapeXml(content)}</content>${mediaContext ? ` ${escapeXml(mediaContext)}` : ""}${webContext}${mediaAnalysisContext}${searxngXml}\n</message>`;
const messageBlock = `<message id="${escapeXml(target.id)}" user="${escapeXml(resolveDisplayName(target))}">\n <user_reputation trust_score="${rep.trust_score}" />${profile ? `\n <user_profile>${sanitizeAiContent(profile.profile_summary)}</user_profile>` : ""}${refXml ? `\n ${refXml}` : ""}\n <content>${escapeXml(content)}</content>${mediaContext ? ` ${escapeXml(mediaContext)}` : ""}${webContext}${mediaAnalysisContext}${searxngXml}\n</message>`;
return { targetId, messageBlock };
}
@@ -5,6 +5,7 @@ import { describe, expect, it } from "vitest";
import {
buildConversationContext,
buildLocationContext,
formatMessageForPrompt,
} from "../src/modules/ai-moderation/conversationContext.js";
import { extractOgMeta } from "../src/modules/ai-moderation/urlFetcher.js";
import type { MessageRecord } from "../src/modules/message-capture/types.js";
@@ -139,6 +140,30 @@ describe("buildConversationContext — recency gating", () => {
});
});
describe("formatMessageForPrompt — server nickname (displayName)", () => {
it("renders member.displayName when captured (per-server nickname)", () => {
const m = msg("n1", NOW - MIN);
m.metadata = JSON.stringify({
member: {
displayName: "Si Goblok Server",
roles: [],
joinedTimestamp: null,
},
});
const line = formatMessageForPrompt(m, "context");
expect(line).toContain("user=Si Goblok Server");
expect(line).not.toContain("user_user_n1");
});
it("falls back to global username when displayName missing", () => {
const line = formatMessageForPrompt(
msg("n2", NOW - MIN, "halo"),
"context",
);
expect(line).toContain("user=user_n2");
});
});
describe("buildLocationContext — channel/thread/nsfw enrichment", () => {
it("renders channel name + thread name from captured metadata", () => {
const t = target();