{ "name": "Exemplar Moderation", "nodes": [ { "parameters": { "httpMethod": "POST", "path": "moderation", "responseMode": "responseNode", "authentication": "headerAuth", "options": {} }, "id": "a1b2c3d4-0001-4000-8000-000000000001", "name": "Moderation Webhook", "type": "n8n-nodes-base.webhook", "typeVersion": 1, "position": [ -1040, 0 ], "notes": "Receives POST from the bot's AI-moderation orchestrator. Header Auth: X-API-Key must match N8N_API_KEY.", "credentials": { "httpHeaderAuth": { "name": "" } } }, { "parameters": { "jsCode": "// Flatten the bot payload. n8n's Webhook node nests the POST body under\n// `body`; fall back to the root for manual test executions.\nconst b = $input.first().json.body ?? $input.first().json;\n\nreturn [\n {\n json: {\n userId: b.userId ?? '',\n userName: b.userName ?? '',\n message: b.message ?? '',\n channelId: b.channelId ?? '',\n channelName: b.channelName ?? '',\n serverId: b.serverId ?? '',\n recentWarnings: Array.isArray(b.recentWarnings) ? b.recentWarnings : [],\n serverRules: b.serverRules ?? '',\n mode: b.mode ?? 'moderate',\n timestamp: b.timestamp ?? new Date().toISOString(),\n },\n },\n];" }, "id": "a1b2c3d4-0002-4000-8000-000000000002", "name": "Extract payload", "type": "n8n-nodes-base.code", "typeVersion": 2, "position": [ -820, 0 ], "notes": "Unwrap the webhook body and normalise defaults." }, { "parameters": { "promptType": "define", "text": "=Channel: {{ $json.channelName }}\nUser: {{ $json.userName }}\nServer rules: {{ $json.serverRules }}\nRecent warnings: {{ JSON.stringify($json.recentWarnings) }}\nMessage: {{ $json.message }}", "options": { "systemMessage": "/no_think\nYou are a Discord content moderator for a single Discord server. For each incoming message decide exactly one of four actions:\n\n- \"allow\" \u2014 the message is fine.\n- \"warn\" \u2014 the message breaks community norms (light insult, mild spam, off-topic noise). The user receives a tracked warning.\n- \"timeout\" \u2014 the message is bad enough to silence the user temporarily (harassment, repeated rule-breaking, NSFW where forbidden). You MUST include a duration: \"30s\", \"5m\", \"1h\", \"1d\". Cap your suggestion at \"1d\" unless extreme.\n- \"delete\" \u2014 the message itself must be removed but the user is otherwise fine (spam links, leaked secrets, accidental personal info).\n\nContext the bot gives you in each request:\n- \"Server rules\": the operator's server-specific rules. If present, treat these as the primary authority. If empty, fall back to the generic baseline below.\n- \"Recent warnings\": the user's last 5 warnings (any age) with reason and ISO timestamp. Use them to judge \"repeated\" rule-breaking objectively: a clean user gets the benefit of the doubt; a user with several recent warns for similar behaviour should escalate toward \"timeout\".\n\nGeneric community baseline (apply when Server rules is empty):\n- No slurs, harassment, or targeted hate.\n- No spam, mass-mentions, or link-flooding.\n- No NSFW content outside channels explicitly designated for it.\n- No doxxing or leaking of personal information.\n- No illegal content.\n\nReturn ONLY valid JSON, no markdown fences, no commentary. Shape:\n {\"action\": \"\", \"reason\": \"\", \"duration\": \"\"}\n\nBe conservative: prefer \"allow\" over false positives. The bot fails open on malformed JSON, so a partial output is worse than \"allow\"." } }, "id": "a1b2c3d4-0003-4000-8000-000000000003", "name": "Moderator", "type": "@n8n/n8n-nodes-langchain.agent", "typeVersion": 3.1, "position": [ -560, 0 ], "notes": "Classifies the message. Output lands in $json.output for the parser." }, { "parameters": { "model": "qwen3:8b", "options": { "format": "json", "temperature": 0.1, "numCtx": 4096, "keepAlive": "30m" } }, "id": "a1b2c3d4-0004-4000-8000-000000000004", "name": "Qwen3 8B (moderation)", "type": "@n8n/n8n-nodes-langchain.lmChatOllama", "typeVersion": 1, "position": [ -560, 220 ], "credentials": { "ollamaApi": { "name": "" } }, "notes": "format:json forces valid JSON. Swap the model for qwen3.5:9b or qwen2.5:7b-instruct as you prefer." }, { "parameters": { "jsCode": "// Parse + validate the LLM verdict and shape it into the exact response the\n// bot reads (result.data.verdict). Fail open to `allow` on any problem.\nconst raw = $input.first().json;\n\n// The Agent node returns its text in `output`; the other keys cover plain\n// chat-model / chain setups so this parser is drop-in either way.\nconst candidates = [\n raw.output,\n raw.text,\n raw.message?.content,\n raw.content,\n raw.output_text,\n raw.choices?.[0]?.message?.content,\n];\nlet text = candidates.find((c) => typeof c === 'string' && c.length > 0) ?? '';\n\n// Strip ```json fences and any qwen block if thinking wasn't disabled.\ntext = text.replace(/[\\s\\S]*?<\\/think>/gi, '');\ntext = text.replace(/^```(?:json)?\\s*/i, '').replace(/```\\s*$/i, '').trim();\n\nconst ALLOWED = new Set(['allow', 'warn', 'timeout', 'delete']);\nconst DURATION_RE = /^\\d+[smhd]$/;\n\nfunction failOpen(reason) {\n return [{ json: { verdict: { action: 'allow', reason: `parse-fallback: ${reason}` } } }];\n}\n\nlet parsed;\ntry {\n parsed = JSON.parse(text);\n} catch (err) {\n return failOpen('invalid JSON');\n}\n\nif (!parsed || typeof parsed !== 'object') return failOpen('not an object');\nif (!ALLOWED.has(parsed.action)) return failOpen('unknown action');\nif (parsed.action === 'timeout') {\n if (typeof parsed.duration !== 'string' || !DURATION_RE.test(parsed.duration)) {\n return failOpen('timeout missing valid duration');\n }\n}\n\nconst verdict = { action: parsed.action };\nif (typeof parsed.reason === 'string') verdict.reason = parsed.reason;\nif (typeof parsed.duration === 'string') verdict.duration = parsed.duration;\n\nreturn [{ json: { verdict } }];" }, "id": "a1b2c3d4-0005-4000-8000-000000000005", "name": "Parse verdict", "type": "n8n-nodes-base.code", "typeVersion": 2, "position": [ -300, 0 ], "notes": "Validates action + duration; fails open to allow." }, { "parameters": { "respondWith": "json", "responseBody": "={{ { \"verdict\": $json.verdict } }}", "options": { "responseCode": 200 } }, "id": "a1b2c3d4-0006-4000-8000-000000000006", "name": "Respond to bot", "type": "n8n-nodes-base.respondToWebhook", "typeVersion": 1, "position": [ -60, 0 ], "notes": "Returns { verdict: {...} } \u2014 the shape the bot reads as result.data.verdict." } ], "connections": { "Moderation Webhook": { "main": [ [ { "node": "Extract payload", "type": "main", "index": 0 } ] ] }, "Extract payload": { "main": [ [ { "node": "Moderator", "type": "main", "index": 0 } ] ] }, "Qwen3 8B (moderation)": { "ai_languageModel": [ [ { "node": "Moderator", "type": "ai_languageModel", "index": 0 } ] ] }, "Moderator": { "main": [ [ { "node": "Parse verdict", "type": "main", "index": 0 } ] ] }, "Parse verdict": { "main": [ [ { "node": "Respond to bot", "type": "main", "index": 0 } ] ] } }, "active": false, "settings": { "executionOrder": "v1" }, "tags": [] }