diff --git a/dist/whispergate/index.ts b/dist/whispergate/index.ts new file mode 100644 index 0000000..61c61ad --- /dev/null +++ b/dist/whispergate/index.ts @@ -0,0 +1,1105 @@ +import fs from "node:fs"; +import path from "node:path"; +import type { OpenClawPluginApi } from "openclaw/plugin-sdk"; +import { evaluateDecision, resolvePolicy, type ChannelPolicy, type Decision, type WhisperGateConfig } from "./rules.js"; +import { checkTurn, advanceTurn, resetTurn, onNewMessage, onSpeakerDone, initTurnOrder, getTurnDebugInfo } from "./turn-manager.js"; +import { startModeratorPresence, stopModeratorPresence } from "./moderator-presence.js"; + +type DiscordControlAction = "channel-private-create" | "channel-private-update" | "member-list"; + +type DecisionRecord = { + decision: Decision; + createdAt: number; + needsRestore?: boolean; +}; + +type PolicyState = { + filePath: string; + channelPolicies: Record; +}; + +type DebugConfig = { + enableDebugLogs?: boolean; + debugLogChannelIds?: string[]; +}; + +const sessionDecision = new Map(); +const sessionAllowed = new Map(); // Track if session was allowed to speak (true) or forced no-reply (false) +const sessionInjected = new Set(); // Track which sessions have already injected the end marker +const sessionChannelId = new Map(); // Track sessionKey -> channelId mapping +const sessionAccountId = new Map(); // Track sessionKey -> accountId mapping +const MAX_SESSION_DECISIONS = 2000; +const DECISION_TTL_MS = 5 * 60 * 1000; +function buildEndMarkerInstruction(endSymbols: string[], isGroupChat: boolean): string { + const symbols = endSymbols.length > 0 ? endSymbols.join("") : "🔚"; + let instruction = `你的这次发言必须以${symbols}作为结尾。除非你的回复是 gateway 关键词(如 NO_REPLY、HEARTBEAT_OK),这些关键词不要加${symbols}。`; + if (isGroupChat) { + instruction += `\n\n群聊发言规则:如果这条消息与你无关、不需要你回应、或你没有有价值的补充,请主动回复 NO_REPLY。不要为了说话而说话。`; + } + return instruction; +} + +const policyState: PolicyState = { + filePath: "", + channelPolicies: {}, +}; + +function normalizeChannel(ctx: Record): string { + const candidates = [ctx.commandSource, ctx.messageProvider, ctx.channelId, ctx.channel]; + for (const c of candidates) { + if (typeof c === "string" && c.trim()) return c.trim().toLowerCase(); + } + return ""; +} + +/** + * Extract the actual Discord channel ID from a conversationId or "to" field. + * OpenClaw uses format "channel:" for Discord conversations. + * Also tries event.to and event.metadata.to as fallbacks. + */ +function extractDiscordChannelId(ctx: Record, event?: Record): string | undefined { + const candidates: unknown[] = [ + ctx.conversationId, + event?.to, + (event?.metadata as Record)?.to, + ]; + for (const c of candidates) { + if (typeof c === "string" && c.trim()) { + const s = c.trim(); + // Handle "channel:123456" format + if (s.startsWith("channel:")) { + const id = s.slice("channel:".length); + if (/^\d+$/.test(id)) return id; + } + // Handle "discord:channel:123456" format + if (s.startsWith("discord:channel:")) { + const id = s.slice("discord:channel:".length); + if (/^\d+$/.test(id)) return id; + } + // If it's a raw snowflake (all digits), use directly + if (/^\d{15,}$/.test(s)) return s; + } + } + return undefined; +} + +function normalizeSender(event: Record, ctx: Record): string | undefined { + const direct = [ctx.senderId, ctx.from, event.from]; + for (const v of direct) { + if (typeof v === "string" && v.trim()) return v.trim(); + } + + const meta = (event.metadata || ctx.metadata) as Record | undefined; + if (!meta) return undefined; + const metaCandidates = [meta.senderId, meta.sender_id, meta.userId, meta.user_id]; + for (const v of metaCandidates) { + if (typeof v === "string" && v.trim()) return v.trim(); + } + + return undefined; +} + +function extractUntrustedConversationInfo(text: string): Record | undefined { + const marker = "Conversation info (untrusted metadata):"; + const idx = text.indexOf(marker); + if (idx < 0) return undefined; + const tail = text.slice(idx + marker.length); + const m = tail.match(/```json\s*([\s\S]*?)\s*```/i); + if (!m) return undefined; + try { + const parsed = JSON.parse(m[1]); + return parsed && typeof parsed === "object" ? (parsed as Record) : undefined; + } catch { + return undefined; + } +} + +function deriveDecisionInputFromPrompt( + prompt: string, + messageProvider?: string, + channelIdFromCtx?: string, +): { + channel: string; + channelId?: string; + senderId?: string; + content: string; + conv: Record; +} { + const conv = extractUntrustedConversationInfo(prompt) || {}; + const channel = (messageProvider || "").toLowerCase(); + + // Priority: ctx.channelId > conv.chat_id > conversation_label > conv.channel_id + let channelId = channelIdFromCtx; + if (!channelId) { + // Try chat_id field (format "channel:123456") + if (typeof conv.chat_id === "string" && conv.chat_id.startsWith("channel:")) { + channelId = conv.chat_id.slice("channel:".length); + } + // Try conversation_label (format "Guild #name channel id:123456") + if (!channelId && typeof conv.conversation_label === "string") { + const labelMatch = conv.conversation_label.match(/channel id:(\d+)/); + if (labelMatch) channelId = labelMatch[1]; + } + // Try channel_id field directly + if (!channelId && typeof conv.channel_id === "string" && conv.channel_id) { + channelId = conv.channel_id; + } + } + + const senderId = + (typeof conv.sender_id === "string" && conv.sender_id) || + (typeof conv.sender === "string" && conv.sender) || + undefined; + + return { channel, channelId, senderId, content: prompt, conv }; +} + +function pruneDecisionMap(now = Date.now()) { + for (const [k, v] of sessionDecision.entries()) { + if (now - v.createdAt > DECISION_TTL_MS) sessionDecision.delete(k); + } + + if (sessionDecision.size <= MAX_SESSION_DECISIONS) return; + const keys = sessionDecision.keys(); + while (sessionDecision.size > MAX_SESSION_DECISIONS) { + const k = keys.next(); + if (k.done) break; + sessionDecision.delete(k.value); + } +} + + +function getLivePluginConfig(api: OpenClawPluginApi, fallback: WhisperGateConfig): WhisperGateConfig { + const root = (api.config as Record) || {}; + const plugins = (root.plugins as Record) || {}; + const entries = (plugins.entries as Record) || {}; + const entry = (entries.whispergate as Record) || {}; + const cfg = (entry.config as Record) || {}; + if (Object.keys(cfg).length > 0) { + // Merge with defaults to ensure optional fields have values + return { + enableDiscordControlTool: true, + enableWhispergatePolicyTool: true, + discordControlApiBaseUrl: "http://127.0.0.1:8790", + enableDebugLogs: false, + debugLogChannelIds: [], + ...cfg, + } as WhisperGateConfig; + } + return fallback; +} + +function resolvePoliciesPath(api: OpenClawPluginApi, config: WhisperGateConfig): string { + return api.resolvePath(config.channelPoliciesFile || "~/.openclaw/whispergate-channel-policies.json"); +} + +function ensurePolicyStateLoaded(api: OpenClawPluginApi, config: WhisperGateConfig) { + if (policyState.filePath) return; + const filePath = resolvePoliciesPath(api, config); + policyState.filePath = filePath; + + try { + if (!fs.existsSync(filePath)) { + fs.mkdirSync(path.dirname(filePath), { recursive: true }); + fs.writeFileSync(filePath, "{}\n", "utf8"); + policyState.channelPolicies = {}; + return; + } + + const raw = fs.readFileSync(filePath, "utf8"); + const parsed = JSON.parse(raw) as Record; + policyState.channelPolicies = parsed && typeof parsed === "object" ? parsed : {}; + } catch (err) { + api.logger.warn(`whispergate: failed init policy file ${filePath}: ${String(err)}`); + policyState.channelPolicies = {}; + } +} + +/** Resolve agentId → Discord accountId from config bindings */ +function resolveAccountId(api: OpenClawPluginApi, agentId: string): string | undefined { + const root = (api.config as Record) || {}; + const bindings = root.bindings as Array> | undefined; + if (!Array.isArray(bindings)) return undefined; + for (const b of bindings) { + if (b.agentId === agentId) { + const match = b.match as Record | undefined; + if (match?.channel === "discord" && typeof match.accountId === "string") { + return match.accountId; + } + } + } + return undefined; +} + +/** + * Get all Discord bot accountIds from config bindings. + */ +function getAllBotAccountIds(api: OpenClawPluginApi): string[] { + const root = (api.config as Record) || {}; + const bindings = root.bindings as Array> | undefined; + if (!Array.isArray(bindings)) return []; + const ids: string[] = []; + for (const b of bindings) { + const match = b.match as Record | undefined; + if (match?.channel === "discord" && typeof match.accountId === "string") { + ids.push(match.accountId); + } + } + return ids; +} + +/** + * Track which bot accountIds have been seen in each channel via message_received. + * Key: channelId, Value: Set of accountIds seen. + */ +const channelSeenAccounts = new Map>(); + +/** + * Record a bot accountId seen in a channel. + * Returns true if this is a new account for this channel (turn order should be updated). + */ +function recordChannelAccount(channelId: string, accountId: string): boolean { + let seen = channelSeenAccounts.get(channelId); + if (!seen) { + seen = new Set(); + channelSeenAccounts.set(channelId, seen); + } + if (seen.has(accountId)) return false; + seen.add(accountId); + return true; +} + +/** + * Get the list of bot accountIds seen in a channel. + * Only returns accounts that are also in the global bindings (actual bots). + */ +function getChannelBotAccountIds(api: OpenClawPluginApi, channelId: string): string[] { + const allBots = new Set(getAllBotAccountIds(api)); + const seen = channelSeenAccounts.get(channelId); + if (!seen) return []; + return [...seen].filter(id => allBots.has(id)); +} + +/** + * Ensure turn order is initialized for a channel. + * Uses only bot accounts that have been seen in this channel. + */ +function ensureTurnOrder(api: OpenClawPluginApi, channelId: string): void { + const botAccounts = getChannelBotAccountIds(api, channelId); + if (botAccounts.length > 0) { + initTurnOrder(channelId, botAccounts); + } +} + +/** + * Build agent identity string for injection into group chat prompts. + */ +function buildAgentIdentity(api: OpenClawPluginApi, agentId: string): string | undefined { + const root = (api.config as Record) || {}; + const bindings = root.bindings as Array> | undefined; + const agents = ((root.agents as Record)?.list as Array>) || []; + if (!Array.isArray(bindings)) return undefined; + + // Find accountId for this agent + let accountId: string | undefined; + for (const b of bindings) { + if (b.agentId === agentId) { + const match = b.match as Record | undefined; + if (match?.channel === "discord" && typeof match.accountId === "string") { + accountId = match.accountId; + break; + } + } + } + if (!accountId) return undefined; + + // Find agent name + const agent = agents.find((a: Record) => a.id === agentId); + const name = (agent?.name as string) || agentId; + + // Find Discord bot user ID from account token (not available directly) + // We'll use accountId as the identifier + return `你是 ${name}(Discord 账号: ${accountId})。`; +} + +// --- Moderator bot helpers --- + +/** Extract Discord user ID from a bot token (base64-encoded in first segment) */ +function userIdFromToken(token: string): string | undefined { + try { + const segment = token.split(".")[0]; + // Add padding + const padded = segment + "=".repeat((4 - (segment.length % 4)) % 4); + return Buffer.from(padded, "base64").toString("utf8"); + } catch { + return undefined; + } +} + +/** Resolve accountId → Discord user ID by reading the account's bot token from config */ +function resolveDiscordUserId(api: OpenClawPluginApi, accountId: string): string | undefined { + const root = (api.config as Record) || {}; + const channels = (root.channels as Record) || {}; + const discord = (channels.discord as Record) || {}; + const accounts = (discord.accounts as Record>) || {}; + const acct = accounts[accountId]; + if (!acct?.token || typeof acct.token !== "string") return undefined; + return userIdFromToken(acct.token); +} + +/** Get the moderator bot's Discord user ID from its token */ +function getModeratorUserId(config: WhisperGateConfig): string | undefined { + if (!config.moderatorBotToken) return undefined; + return userIdFromToken(config.moderatorBotToken); +} + +/** Send a message as the moderator bot via Discord REST API */ +async function sendModeratorMessage(token: string, channelId: string, content: string, logger: { info: (msg: string) => void; warn: (msg: string) => void }): Promise { + try { + const r = await fetch(`https://discord.com/api/v10/channels/${channelId}/messages`, { + method: "POST", + headers: { + "Authorization": `Bot ${token}`, + "Content-Type": "application/json", + }, + body: JSON.stringify({ content }), + }); + if (!r.ok) { + const text = await r.text(); + logger.warn(`whispergate: moderator send failed (${r.status}): ${text}`); + return false; + } + logger.info(`whispergate: moderator message sent to channel=${channelId}`); + return true; + } catch (err) { + logger.warn(`whispergate: moderator send error: ${String(err)}`); + return false; + } +} + +function persistPolicies(api: OpenClawPluginApi): void { + const filePath = policyState.filePath; + if (!filePath) throw new Error("policy file path not initialized"); + const before = JSON.stringify(policyState.channelPolicies, null, 2) + "\n"; + const tmp = `${filePath}.tmp`; + fs.mkdirSync(path.dirname(filePath), { recursive: true }); + fs.writeFileSync(tmp, before, "utf8"); + fs.renameSync(tmp, filePath); + api.logger.info(`whispergate: policy file persisted: ${filePath}`); +} + +function pickDefined(input: Record) { + const out: Record = {}; + for (const [k, v] of Object.entries(input)) { + if (v !== undefined) out[k] = v; + } + return out; +} + +function shouldDebugLog(cfg: DebugConfig, channelId?: string): boolean { + if (!cfg.enableDebugLogs) return false; + const allow = Array.isArray(cfg.debugLogChannelIds) ? cfg.debugLogChannelIds : []; + if (allow.length === 0) return true; + if (!channelId) return true; // 允许打印,方便排查 channelId 为空的场景 + return allow.includes(channelId); +} + +function debugCtxSummary(ctx: Record, event: Record) { + const meta = ((ctx.metadata || event.metadata || {}) as Record) || {}; + return { + sessionKey: typeof ctx.sessionKey === "string" ? ctx.sessionKey : undefined, + commandSource: typeof ctx.commandSource === "string" ? ctx.commandSource : undefined, + messageProvider: typeof ctx.messageProvider === "string" ? ctx.messageProvider : undefined, + channel: typeof ctx.channel === "string" ? ctx.channel : undefined, + channelId: typeof ctx.channelId === "string" ? ctx.channelId : undefined, + senderId: typeof ctx.senderId === "string" ? ctx.senderId : undefined, + from: typeof ctx.from === "string" ? ctx.from : undefined, + metaSenderId: + typeof meta.senderId === "string" + ? meta.senderId + : typeof meta.sender_id === "string" + ? meta.sender_id + : undefined, + metaUserId: + typeof meta.userId === "string" + ? meta.userId + : typeof meta.user_id === "string" + ? meta.user_id + : undefined, + }; +} + +export default { + id: "whispergate", + name: "WhisperGate", + register(api: OpenClawPluginApi) { + // Merge pluginConfig with defaults (in case config is missing from openclaw.json) + const baseConfig = { + enableDiscordControlTool: true, + enableWhispergatePolicyTool: true, + discordControlApiBaseUrl: "http://127.0.0.1:8790", + ...(api.pluginConfig || {}), + } as WhisperGateConfig & { + enableDiscordControlTool: boolean; + discordControlApiBaseUrl: string; + discordControlApiToken?: string; + discordControlCallerId?: string; + enableWhispergatePolicyTool: boolean; + }; + + const liveAtRegister = getLivePluginConfig(api, baseConfig as WhisperGateConfig); + ensurePolicyStateLoaded(api, liveAtRegister); + + // Start moderator bot presence (keep it "online" on Discord) + if (liveAtRegister.moderatorBotToken) { + startModeratorPresence(liveAtRegister.moderatorBotToken, api.logger); + api.logger.info("whispergate: moderator bot presence starting"); + } + + api.registerTool( + { + name: "whispergate_tools", + description: "WhisperGate unified tool: Discord admin actions + in-memory policy management.", + parameters: { + type: "object", + additionalProperties: false, + properties: { + action: { + type: "string", + enum: ["channel-private-create", "channel-private-update", "member-list", "policy-get", "policy-set-channel", "policy-delete-channel", "turn-status", "turn-advance", "turn-reset"], + }, + guildId: { type: "string" }, + name: { type: "string" }, + type: { type: "number" }, + parentId: { type: "string" }, + topic: { type: "string" }, + position: { type: "number" }, + nsfw: { type: "boolean" }, + allowedUserIds: { type: "array", items: { type: "string" } }, + allowedRoleIds: { type: "array", items: { type: "string" } }, + allowMask: { type: "string" }, + denyEveryoneMask: { type: "string" }, + channelId: { type: "string" }, + mode: { type: "string", enum: ["merge", "replace"] }, + addUserIds: { type: "array", items: { type: "string" } }, + addRoleIds: { type: "array", items: { type: "string" } }, + removeTargetIds: { type: "array", items: { type: "string" } }, + denyMask: { type: "string" }, + limit: { type: "number" }, + after: { type: "string" }, + fields: { anyOf: [{ type: "string" }, { type: "array", items: { type: "string" } }] }, + dryRun: { type: "boolean" }, + listMode: { type: "string", enum: ["human-list", "agent-list"] }, + humanList: { type: "array", items: { type: "string" } }, + agentList: { type: "array", items: { type: "string" } }, + endSymbols: { type: "array", items: { type: "string" } }, + }, + required: ["action"], + }, + async execute(_id: string, params: Record) { + const live = getLivePluginConfig(api, baseConfig as WhisperGateConfig) as WhisperGateConfig & { + discordControlApiBaseUrl?: string; + discordControlApiToken?: string; + discordControlCallerId?: string; + enableDiscordControlTool?: boolean; + enableWhispergatePolicyTool?: boolean; + }; + ensurePolicyStateLoaded(api, live); + + const action = String(params.action || ""); + const discordActions = new Set(["channel-private-create", "channel-private-update", "member-list"]); + + if (discordActions.has(action)) { + if (live.enableDiscordControlTool === false) { + return { content: [{ type: "text", text: "discord actions disabled by config" }], isError: true }; + } + const baseUrl = (live.discordControlApiBaseUrl || "http://127.0.0.1:8790").replace(/\/$/, ""); + const body = pickDefined({ ...params, action: action as DiscordControlAction }); + const headers: Record = { "Content-Type": "application/json" }; + if (live.discordControlApiToken) headers.Authorization = `Bearer ${live.discordControlApiToken}`; + if (live.discordControlCallerId) headers["X-OpenClaw-Caller-Id"] = live.discordControlCallerId; + + const r = await fetch(`${baseUrl}/v1/discord/action`, { + method: "POST", + headers, + body: JSON.stringify(body), + }); + const text = await r.text(); + if (!r.ok) { + return { + content: [{ type: "text", text: `whispergate_tools discord failed (${r.status}): ${text}` }], + isError: true, + }; + } + return { content: [{ type: "text", text }] }; + } + + if (live.enableWhispergatePolicyTool === false) { + return { content: [{ type: "text", text: "policy actions disabled by config" }], isError: true }; + } + + if (action === "policy-get") { + return { + content: [{ type: "text", text: JSON.stringify({ file: policyState.filePath, policies: policyState.channelPolicies }, null, 2) }], + }; + } + + if (action === "policy-set-channel") { + const channelId = String(params.channelId || "").trim(); + if (!channelId) return { content: [{ type: "text", text: "channelId is required" }], isError: true }; + + const prev = JSON.parse(JSON.stringify(policyState.channelPolicies)); + try { + const next: ChannelPolicy = { + listMode: (params.listMode as "human-list" | "agent-list" | undefined) || undefined, + humanList: Array.isArray(params.humanList) ? (params.humanList as string[]) : undefined, + agentList: Array.isArray(params.agentList) ? (params.agentList as string[]) : undefined, + endSymbols: Array.isArray(params.endSymbols) ? (params.endSymbols as string[]) : undefined, + }; + policyState.channelPolicies[channelId] = pickDefined(next as unknown as Record) as ChannelPolicy; + persistPolicies(api); + return { content: [{ type: "text", text: JSON.stringify({ ok: true, channelId, policy: policyState.channelPolicies[channelId] }) }] }; + } catch (err) { + policyState.channelPolicies = prev; + return { content: [{ type: "text", text: `persist failed: ${String(err)}` }], isError: true }; + } + } + + if (action === "policy-delete-channel") { + const channelId = String(params.channelId || "").trim(); + if (!channelId) return { content: [{ type: "text", text: "channelId is required" }], isError: true }; + const prev = JSON.parse(JSON.stringify(policyState.channelPolicies)); + try { + delete policyState.channelPolicies[channelId]; + persistPolicies(api); + return { content: [{ type: "text", text: JSON.stringify({ ok: true, channelId, deleted: true }) }] }; + } catch (err) { + policyState.channelPolicies = prev; + return { content: [{ type: "text", text: `persist failed: ${String(err)}` }], isError: true }; + } + } + + if (action === "turn-status") { + const channelId = String(params.channelId || "").trim(); + if (!channelId) return { content: [{ type: "text", text: "channelId is required" }], isError: true }; + return { content: [{ type: "text", text: JSON.stringify(getTurnDebugInfo(channelId), null, 2) }] }; + } + + if (action === "turn-advance") { + const channelId = String(params.channelId || "").trim(); + if (!channelId) return { content: [{ type: "text", text: "channelId is required" }], isError: true }; + const next = advanceTurn(channelId); + return { content: [{ type: "text", text: JSON.stringify({ ok: true, channelId, nextSpeaker: next, ...getTurnDebugInfo(channelId) }) }] }; + } + + if (action === "turn-reset") { + const channelId = String(params.channelId || "").trim(); + if (!channelId) return { content: [{ type: "text", text: "channelId is required" }], isError: true }; + resetTurn(channelId); + return { content: [{ type: "text", text: JSON.stringify({ ok: true, channelId, ...getTurnDebugInfo(channelId) }) }] }; + } + + return { content: [{ type: "text", text: `unsupported action: ${action}` }], isError: true }; + }, + }, + { optional: false }, + ); + + api.on("message_received", async (event, ctx) => { + try { + const c = (ctx || {}) as Record; + const e = (event || {}) as Record; + // ctx.channelId is the platform name (e.g. "discord"), NOT the Discord channel snowflake. + // Extract the real Discord channel ID from conversationId or event.to. + const preChannelId = extractDiscordChannelId(c, e); + const livePre = getLivePluginConfig(api, baseConfig as WhisperGateConfig) as WhisperGateConfig & DebugConfig; + if (shouldDebugLog(livePre, preChannelId)) { + api.logger.info(`whispergate: debug message_received preflight ctx=${JSON.stringify(debugCtxSummary(c, e))}`); + } + + // Turn management on message received + if (preChannelId) { + ensureTurnOrder(api, preChannelId); + // event.from is often the channel target (e.g. "discord:channel:xxx"), NOT the sender. + // The actual sender ID is in event.metadata.senderId. + const metadata = (e as Record).metadata as Record | undefined; + const from = (typeof metadata?.senderId === "string" && metadata.senderId) + || (typeof (e as Record).from === "string" ? (e as Record).from as string : ""); + + // Ignore moderator bot messages — they don't affect turn state + const moderatorUserId = getModeratorUserId(livePre); + if (moderatorUserId && from === moderatorUserId) { + if (shouldDebugLog(livePre, preChannelId)) { + api.logger.info(`whispergate: ignoring moderator message in channel=${preChannelId}`); + } + // Don't call onNewMessage — moderator messages are transparent to turn logic + } else { + const humanList = livePre.humanList || livePre.bypassUserIds || []; + const isHuman = humanList.includes(from); + const senderAccountId = typeof c.accountId === "string" ? c.accountId : undefined; + + // Track which bot accounts are present in this channel + if (senderAccountId && senderAccountId !== "default") { + const isNew = recordChannelAccount(preChannelId, senderAccountId); + if (isNew) { + // Re-initialize turn order with updated channel membership + ensureTurnOrder(api, preChannelId); + api.logger.info(`whispergate: new account ${senderAccountId} seen in channel=${preChannelId}, turn order updated`); + } + } + + onNewMessage(preChannelId, senderAccountId, isHuman); + if (shouldDebugLog(livePre, preChannelId)) { + api.logger.info(`whispergate: turn onNewMessage channel=${preChannelId} from=${from} isHuman=${isHuman} accountId=${senderAccountId ?? "unknown"}`); + } + } + } + } catch (err) { + api.logger.warn(`whispergate: message hook failed: ${String(err)}`); + } + }); + + api.on("before_model_resolve", async (event, ctx) => { + const key = ctx.sessionKey; + if (!key) return; + + const live = getLivePluginConfig(api, baseConfig as WhisperGateConfig) as WhisperGateConfig & DebugConfig; + ensurePolicyStateLoaded(api, live); + + const prompt = ((event as Record).prompt as string) || ""; + + if (live.enableDebugLogs) { + api.logger.info( + `whispergate: DEBUG_BEFORE_MODEL_RESOLVE ctx=${JSON.stringify({ sessionKey: ctx.sessionKey, messageProvider: ctx.messageProvider, agentId: ctx.agentId })} ` + + `promptPreview=${prompt.slice(0, 300)}`, + ); + } + + const derived = deriveDecisionInputFromPrompt(prompt, ctx.messageProvider, ctx.channelId); + // Fallback: extract channelId from sessionKey (format "agent::discord:channel:") + if (!derived.channelId && key) { + const skMatch = key.match(/:channel:(\d+)$/); + if (skMatch) derived.channelId = skMatch[1]; + } + // Only proceed if: discord channel AND prompt contains untrusted metadata + const hasConvMarker = prompt.includes("Conversation info (untrusted metadata):"); + if (live.discordOnly !== false && (!hasConvMarker || derived.channel !== "discord")) return; + + // Always save channelId and accountId mappings for use in later hooks + if (derived.channelId) { + sessionChannelId.set(key, derived.channelId); + } + const resolvedAccountId = resolveAccountId(api, ctx.agentId || ""); + if (resolvedAccountId) { + sessionAccountId.set(key, resolvedAccountId); + } + + let rec = sessionDecision.get(key); + if (!rec || Date.now() - rec.createdAt > DECISION_TTL_MS) { + if (rec) sessionDecision.delete(key); + const decision = evaluateDecision({ + config: live, + channel: derived.channel, + channelId: derived.channelId, + channelPolicies: policyState.channelPolicies, + senderId: derived.senderId, + content: derived.content, + }); + rec = { decision, createdAt: Date.now() }; + sessionDecision.set(key, rec); + pruneDecisionMap(); + if (shouldDebugLog(live, derived.channelId)) { + api.logger.info( + `whispergate: debug before_model_resolve recompute session=${key} ` + + `channel=${derived.channel} channelId=${derived.channelId ?? ""} senderId=${derived.senderId ?? ""} ` + + `convSenderId=${String((derived.conv as Record).sender_id ?? "")} ` + + `convSender=${String((derived.conv as Record).sender ?? "")} ` + + `convChannelId=${String((derived.conv as Record).channel_id ?? "")} ` + + `decision=${decision.reason} shouldNoReply=${decision.shouldUseNoReply} shouldInject=${decision.shouldInjectEndMarkerPrompt}`, + ); + } + } + + // Turn-based check: ALWAYS check turn order regardless of evaluateDecision result. + // This ensures only the current speaker can respond even for human messages. + if (derived.channelId) { + ensureTurnOrder(api, derived.channelId); + const accountId = resolveAccountId(api, ctx.agentId || ""); + if (accountId) { + const turnCheck = checkTurn(derived.channelId, accountId); + if (!turnCheck.allowed) { + // Forced no-reply - record this session as not allowed to speak + sessionAllowed.set(key, false); + api.logger.info( + `whispergate: turn gate blocked session=${key} accountId=${accountId} currentSpeaker=${turnCheck.currentSpeaker} reason=${turnCheck.reason}`, + ); + return { + providerOverride: live.noReplyProvider, + modelOverride: live.noReplyModel, + }; + } + // Allowed to speak - record this session as allowed + sessionAllowed.set(key, true); + } + } + + if (!rec.decision.shouldUseNoReply) { + // 如果之前有 no-reply 执行过,现在不需要了,清除 override 恢复原模型 + if (rec.needsRestore) { + sessionDecision.delete(key); + return { + providerOverride: undefined, + modelOverride: undefined, + }; + } + return; + } + + // 标记这次执行了 no-reply,下次需要恢复模型 + rec.needsRestore = true; + sessionDecision.set(key, rec); + + // 无论是否有缓存,只要 debug flag 开启就打印决策详情 + if (live.enableDebugLogs) { + const prompt = ((event as Record).prompt as string) || ""; + const hasConvMarker = prompt.includes("Conversation info (untrusted metadata):"); + api.logger.info( + `whispergate: DEBUG_NO_REPLY_TRIGGER session=${key} ` + + `channel=${derived.channel} channelId=${derived.channelId ?? ""} senderId=${derived.senderId ?? ""} ` + + `convSenderId=${String((derived.conv as Record).sender_id ?? "")} ` + + `convSender=${String((derived.conv as Record).sender ?? "")} ` + + `decision=${rec.decision.reason} ` + + `shouldNoReply=${rec.decision.shouldUseNoReply} shouldInject=${rec.decision.shouldInjectEndMarkerPrompt} ` + + `hasConvMarker=${hasConvMarker} promptLen=${prompt.length}`, + ); + } + + api.logger.info( + `whispergate: override model for session=${key}, provider=${live.noReplyProvider}, model=${live.noReplyModel}, reason=${rec.decision.reason}`, + ); + + return { + providerOverride: live.noReplyProvider, + modelOverride: live.noReplyModel, + }; + }); + + api.on("before_prompt_build", async (event, ctx) => { + const key = ctx.sessionKey; + if (!key) return; + + const live = getLivePluginConfig(api, baseConfig as WhisperGateConfig) as WhisperGateConfig & DebugConfig; + ensurePolicyStateLoaded(api, live); + + let rec = sessionDecision.get(key); + if (!rec || Date.now() - rec.createdAt > DECISION_TTL_MS) { + if (rec) sessionDecision.delete(key); + + const prompt = ((event as Record).prompt as string) || ""; + const derived = deriveDecisionInputFromPrompt(prompt, ctx.messageProvider, ctx.channelId); + + const decision = evaluateDecision({ + config: live, + channel: derived.channel, + channelId: derived.channelId, + channelPolicies: policyState.channelPolicies, + senderId: derived.senderId, + content: derived.content, + }); + rec = { decision, createdAt: Date.now() }; + if (shouldDebugLog(live, derived.channelId)) { + api.logger.info( + `whispergate: debug before_prompt_build recompute session=${key} ` + + `channel=${derived.channel} channelId=${derived.channelId ?? ""} senderId=${derived.senderId ?? ""} ` + + `convSenderId=${String((derived.conv as Record).sender_id ?? "")} ` + + `convSender=${String((derived.conv as Record).sender ?? "")} ` + + `convChannelId=${String((derived.conv as Record).channel_id ?? "")} ` + + `decision=${decision.reason} shouldNoReply=${decision.shouldUseNoReply} shouldInject=${decision.shouldInjectEndMarkerPrompt}`, + ); + } + } + + sessionDecision.delete(key); + + // Only inject once per session (one-time injection) + if (sessionInjected.has(key)) { + if (shouldDebugLog(live, undefined)) { + api.logger.info( + `whispergate: debug before_prompt_build session=${key} inject skipped (already injected)`, + ); + } + return; + } + + if (!rec.decision.shouldInjectEndMarkerPrompt) { + if (shouldDebugLog(live, undefined)) { + api.logger.info( + `whispergate: debug before_prompt_build session=${key} inject=false reason=${rec.decision.reason}`, + ); + } + return; + } + + // Resolve end symbols from config/policy for dynamic instruction + const prompt = ((event as Record).prompt as string) || ""; + const derived = deriveDecisionInputFromPrompt(prompt, ctx.messageProvider, ctx.channelId); + const policy = resolvePolicy(live, derived.channelId, policyState.channelPolicies); + const isGroupChat = derived.conv.is_group_chat === true || derived.conv.is_group_chat === "true"; + const instruction = buildEndMarkerInstruction(policy.endSymbols, isGroupChat); + + // Inject agent identity for group chats + let identity = ""; + if (isGroupChat && ctx.agentId) { + const idStr = buildAgentIdentity(api, ctx.agentId); + if (idStr) identity = idStr + "\n\n"; + } + + // Mark session as injected (one-time injection) + sessionInjected.add(key); + + api.logger.info(`whispergate: prepend end marker instruction for session=${key}, reason=${rec.decision.reason} isGroupChat=${isGroupChat}`); + return { prependContext: identity + instruction }; + }); + + // Register slash commands for Discord + api.registerCommand({ + name: "whispergate", + description: "WhisperGate 频道策略管理", + acceptsArgs: true, + handler: async (cmdCtx) => { + const args = cmdCtx.args || ""; + const parts = args.trim().split(/\s+/); + const subCmd = parts[0] || "help"; + + if (subCmd === "help") { + return { text: `WhisperGate 命令:\n` + + `/whispergate status - 显示当前频道状态\n` + + `/whispergate turn-status - 显示轮流发言状态\n` + + `/whispergate turn-advance - 手动推进轮流\n` + + `/whispergate turn-reset - 重置轮流顺序` }; + } + + if (subCmd === "status") { + return { text: JSON.stringify({ policies: policyState.channelPolicies }, null, 2) }; + } + + if (subCmd === "turn-status") { + const channelId = cmdCtx.channelId; + if (!channelId) return { text: "无法获取频道ID", isError: true }; + return { text: JSON.stringify(getTurnDebugInfo(channelId), null, 2) }; + } + + if (subCmd === "turn-advance") { + const channelId = cmdCtx.channelId; + if (!channelId) return { text: "无法获取频道ID", isError: true }; + const next = advanceTurn(channelId); + return { text: JSON.stringify({ ok: true, nextSpeaker: next }) }; + } + + if (subCmd === "turn-reset") { + const channelId = cmdCtx.channelId; + if (!channelId) return { text: "无法获取频道ID", isError: true }; + resetTurn(channelId); + return { text: JSON.stringify({ ok: true }) }; + } + + return { text: `未知子命令: ${subCmd}`, isError: true }; + }, + }); + + // Handle NO_REPLY detection before message write + // This is where we detect if agent output is NO_REPLY and handle turn advancement + // NOTE: This hook is synchronous, do not use async/await + api.on("before_message_write", (event, ctx) => { + try { + // Debug: print all available keys in event and ctx + api.logger.info( + `whispergate: DEBUG before_message_write eventKeys=${JSON.stringify(Object.keys(event ?? {}))} ctxKeys=${JSON.stringify(Object.keys(ctx ?? {}))}`, + ); + + // before_message_write ctx only has { agentId, sessionKey }. + // Use session mappings populated during before_model_resolve for channelId/accountId. + // Content comes from event.message (AgentMessage). + let key = ctx.sessionKey; + let channelId: string | undefined; + let accountId: string | undefined; + + // Get from session mapping (set in before_model_resolve) + if (key) { + channelId = sessionChannelId.get(key); + accountId = sessionAccountId.get(key); + } + + // Extract content from event.message (AgentMessage) + let content = ""; + const msg = (event as Record).message as Record | undefined; + if (msg) { + // AgentMessage may have content as string or nested + if (typeof msg.content === "string") { + content = msg.content; + } else if (Array.isArray(msg.content)) { + // content might be an array of parts (Anthropic format) + for (const part of msg.content) { + if (typeof part === "string") content += part; + else if (part && typeof part === "object" && typeof (part as Record).text === "string") { + content += (part as Record).text; + } + } + } + } + // Fallback to event.content + if (!content) { + content = ((event as Record).content as string) || ""; + } + + // Always log for debugging - show all available info + api.logger.info( + `whispergate: DEBUG before_message_write session=${key ?? "undefined"} channel=${channelId ?? "undefined"} accountId=${accountId ?? "undefined"} contentType=${typeof content} content=${String(content).slice(0, 200)}`, + ); + + if (!key || !channelId || !accountId) return; + + const live = getLivePluginConfig(api, baseConfig as WhisperGateConfig) as WhisperGateConfig & DebugConfig; + + const trimmed = content.trim(); + const isNoReply = /^NO_REPLY$/i.test(trimmed); + + // Log turn state for debugging + const turnDebug = getTurnDebugInfo(channelId); + api.logger.info( + `whispergate: DEBUG turn state channel=${channelId} turnOrder=${JSON.stringify(turnDebug.turnOrder)} currentSpeaker=${turnDebug.currentSpeaker} noRepliedThisCycle=${JSON.stringify([...turnDebug.noRepliedThisCycle])}`, + ); + + if (!isNoReply) { + api.logger.info( + `whispergate: before_message_write content is not NO_REPLY, skipping channel=${channelId}`, + ); + return; + } + + // Check if this session was forced no-reply or allowed to speak + const wasAllowed = sessionAllowed.get(key); + api.logger.info( + `whispergate: DEBUG NO_REPLY detected session=${key} wasAllowed=${wasAllowed}`, + ); + + if (wasAllowed === undefined) return; // No record, skip + + if (wasAllowed === false) { + // Forced no-reply - do not advance turn + sessionAllowed.delete(key); + api.logger.info( + `whispergate: before_message_write forced no-reply session=${key} channel=${channelId} - not advancing turn`, + ); + return; + } + + // Allowed to speak (current speaker) but chose NO_REPLY - advance turn + ensureTurnOrder(api, channelId); + const nextSpeaker = onSpeakerDone(channelId, accountId, true); + + sessionAllowed.delete(key); + + api.logger.info( + `whispergate: before_message_write real no-reply session=${key} channel=${channelId} nextSpeaker=${nextSpeaker ?? "dormant"}`, + ); + + // If all agents NO_REPLY'd (dormant), don't trigger handoff + if (!nextSpeaker) { + if (shouldDebugLog(live, channelId)) { + api.logger.info( + `whispergate: before_message_write all agents no-reply, going dormant - no handoff`, + ); + } + return; + } + + // Trigger moderator handoff message (fire-and-forget, don't await) + if (live.moderatorBotToken) { + const nextUserId = resolveDiscordUserId(api, nextSpeaker); + if (nextUserId) { + const handoffMsg = `轮到(<@${nextUserId}>)了,如果没有想说的请直接回复NO_REPLY`; + void sendModeratorMessage(live.moderatorBotToken, channelId, handoffMsg, api.logger).catch((err) => { + api.logger.warn(`whispergate: before_message_write handoff failed: ${String(err)}`); + }); + } else { + api.logger.warn(`whispergate: cannot resolve Discord userId for next speaker accountId=${nextSpeaker}`); + } + } + } catch (err) { + api.logger.warn(`whispergate: before_message_write hook failed: ${String(err)}`); + } + }); + + // Turn advance: when an agent sends a message, check if it signals end of turn + api.on("message_sent", async (event, ctx) => { + try { + const key = ctx.sessionKey; + const c = (ctx || {}) as Record; + const e = (event || {}) as Record; + + // Always log raw context first for debugging + api.logger.info( + `whispergate: DEBUG message_sent RAW ctxKeys=${JSON.stringify(Object.keys(c))} eventKeys=${JSON.stringify(Object.keys(e))} ` + + `ctx.channelId=${String(c.channelId ?? "undefined")} ctx.conversationId=${String(c.conversationId ?? "undefined")} ` + + `ctx.accountId=${String(c.accountId ?? "undefined")} event.to=${String(e.to ?? "undefined")} ` + + `session=${key ?? "undefined"}`, + ); + + // ctx.channelId is the platform name (e.g. "discord"), NOT the Discord channel snowflake. + // Extract real Discord channel ID from conversationId or event.to. + let channelId = extractDiscordChannelId(c, e); + // Fallback: sessionKey mapping + if (!channelId && key) { + channelId = sessionChannelId.get(key); + } + // Fallback: parse from sessionKey + if (!channelId && key) { + const skMatch = key.match(/:channel:(\d+)$/); + if (skMatch) channelId = skMatch[1]; + } + const accountId = (ctx.accountId as string | undefined) || (key ? sessionAccountId.get(key) : undefined); + const content = (event.content as string) || ""; + + // Debug log + api.logger.info( + `whispergate: DEBUG message_sent RESOLVED session=${key ?? "undefined"} channelId=${channelId ?? "undefined"} accountId=${accountId ?? "undefined"} content=${content.slice(0, 100)}`, + ); + + if (!channelId || !accountId) return; + + const live = getLivePluginConfig(api, baseConfig as WhisperGateConfig) as WhisperGateConfig & DebugConfig; + ensurePolicyStateLoaded(api, live); + const policy = resolvePolicy(live, channelId, policyState.channelPolicies); + + const trimmed = content.trim(); + const isEmpty = trimmed.length === 0; + const isNoReply = /^NO_REPLY$/i.test(trimmed) || /^HEARTBEAT_OK$/i.test(trimmed); + const lastChar = trimmed.length > 0 ? Array.from(trimmed).pop() || "" : ""; + const hasEndSymbol = !!lastChar && policy.endSymbols.includes(lastChar); + const wasNoReply = isEmpty || isNoReply; + + if (wasNoReply || hasEndSymbol) { + const nextSpeaker = onSpeakerDone(channelId, accountId, wasNoReply); + const trigger = wasNoReply ? (isEmpty ? "empty" : "no_reply_keyword") : "end_symbol"; + api.logger.info( + `whispergate: turn onSpeakerDone channel=${channelId} from=${accountId} next=${nextSpeaker ?? "dormant"} trigger=${trigger}`, + ); + // Moderator handoff: when current speaker NO_REPLY'd and there's a next speaker, + // send a handoff message via the moderator bot to trigger the next agent + if (wasNoReply && nextSpeaker && live.moderatorBotToken) { + const nextUserId = resolveDiscordUserId(api, nextSpeaker); + if (nextUserId) { + const handoffMsg = `轮到(<@${nextUserId}>)了,如果没有想说的请直接回复NO_REPLY`; + sendModeratorMessage(live.moderatorBotToken, channelId, handoffMsg, api.logger); + } else { + api.logger.warn(`whispergate: cannot resolve Discord userId for next speaker accountId=${nextSpeaker}`); + } + } + } + } catch (err) { + api.logger.warn(`whispergate: message_sent hook failed: ${String(err)}`); + } + }); + }, +}; diff --git a/docs/TURN-WAKEUP-PROBLEM.md b/docs/TURN-WAKEUP-PROBLEM.md new file mode 100644 index 0000000..f56b59f --- /dev/null +++ b/docs/TURN-WAKEUP-PROBLEM.md @@ -0,0 +1,108 @@ +# Turn-Based Speaking: Wakeup Problem + +## Context + +WhisperGate implements turn-based speaking for Discord group channels where multiple AI agents coexist. Only one agent (the "current speaker") is allowed to respond at a time. Others are silenced via a no-reply model override. + +## The Problem + +When the current speaker responds with **NO_REPLY** (decides the message is not relevant to them), the turn advances to the next agent. However, **the next agent has no trigger to start speaking**. + +### Why This Happens + +1. A message arrives in the Discord channel +2. OpenClaw routes it to **all** agent sessions in that channel simultaneously +3. The WhisperGate plugin intercepts at `before_model_resolve`: + - Current speaker → allowed to process + - Everyone else → forced to no-reply model (message is "consumed" silently) +4. Current speaker processes the message and returns NO_REPLY +5. `message_sent` hook detects NO_REPLY → turn advances to next agent +6. **But the next agent already "consumed" the message in step 3** — their session processed it (as no-reply) and moved on +7. No new message exists to trigger the next agent + +### The Result + +After a NO_REPLY, the next speaker sits idle until a **new** message arrives in the channel (from a human or another source). The original message that should have been passed to the next speaker is lost. + +## When This Matters + +- **Single-round conversation**: Human asks a question → Agent A says NO_REPLY → Agent B should answer but can't +- **Chain conversations**: Agent A defers → Agent B defers → Agent C should speak but never gets triggered + +## When This Doesn't Matter + +- **End-symbol responses**: When an agent actually speaks (ends with 🔚), the turn advances and the next agent will respond to the **next** message. This is fine. +- **Human-driven channels**: If humans keep sending messages, the dormant state resolves quickly. + +## Possible Solutions + +### 1. Synthetic Trigger Message (Plugin-Side) + +After detecting NO_REPLY and advancing the turn, the plugin sends a **synthetic message** to the channel that triggers the next agent. + +**Challenges:** +- The plugin SDK (`message_sent` hook) doesn't have an API to inject messages into agent sessions +- Sending a real Discord message (even invisible like zero-width space) creates noise and may confuse other agents +- The synthetic message wouldn't contain the original user's context + +### 2. Deferred Evaluation (Don't Block in before_model_resolve) + +Instead of blocking non-speakers at `before_model_resolve`, let all agents receive the message but inject a "you are not the current speaker, reply NO_REPLY" instruction. The current speaker gets a normal prompt. + +After the current speaker responds with NO_REPLY, the plugin would need to **re-trigger** the next agent's session with the same message. + +**Challenges:** +- All agents still consume tokens for the NO_REPLY evaluation +- Re-triggering a session with an already-processed message requires OpenClaw internal APIs + +### 3. Queue + Replay (Plugin-Side State) + +The plugin stores the original message when it arrives. After NO_REPLY, it replays the message by injecting it into the next speaker's session. + +**Challenges:** +- Requires access to session injection API (not available in current plugin SDK) +- Managing the message queue adds complexity + +### 4. Gateway-Level Support (OpenClaw Core Change) + +Add a plugin hook return value like `{ defer: true }` in `before_model_resolve` that tells OpenClaw: "don't process this message yet, but keep it pending." When the turn advances, the plugin could call `api.retrigger(sessionKey)` to replay the pending message. + +**Challenges:** +- Requires changes to OpenClaw core, not just the plugin +- Needs design discussion with the OpenClaw team + +### 5. Bot-to-Bot Handoff via Discord Message + +When current speaker NO_REPLYs, have **that bot** send a brief handoff message in the channel: e.g., "(轮到下一位)" or a reaction. This real Discord message triggers all agents, and the turn manager ensures only the next speaker responds. + +**Challenges:** +- Adds visible noise to the channel (could use a convention like a specific emoji reaction) +- The no-reply'd bot can't send messages (it was silenced) +- Could use the discord-control-api to send as a different bot + +### 6. Timer-Based Retry (Pragmatic) + +After advancing the turn, set a short timer (e.g., 2-3 seconds). If no new message has arrived, send a minimal trigger. This could be an internal "nudge" if the SDK supports it. + +**Challenges:** +- Timing is fragile +- Still needs a mechanism to trigger the next agent + +## Recommendation + +**Solution 5 (Bot-to-Bot Handoff)** is the most pragmatic with current constraints. The implementation would be: + +1. In the `message_sent` hook, after detecting NO_REPLY and advancing the turn: +2. Use the discord-control-api to send a short message (e.g., `[轮转]` or a specific emoji) from the **next speaker's bot account** in the channel +3. This real Discord message triggers OpenClaw to route it to all agents +4. The turn manager allows only the (now-current) next speaker to respond +5. The next speaker sees the original conversation context in their session history and responds appropriately + +**Downside:** Adds a visible "[轮转]" message. Could be mitigated by immediately deleting it after delivery, or using a reaction instead of a message. + +## Open Questions + +1. Does the OpenClaw plugin SDK support injecting messages into sessions? +2. Can plugins access the Discord client to send messages directly? +3. Would an OpenClaw core `defer`/`retrigger` mechanism be feasible? +4. Is visible channel noise acceptable for the handoff message? diff --git a/plugin/index.ts b/plugin/index.ts index 20ac75e..61c61ad 100644 --- a/plugin/index.ts +++ b/plugin/index.ts @@ -2,6 +2,8 @@ import fs from "node:fs"; import path from "node:path"; import type { OpenClawPluginApi } from "openclaw/plugin-sdk"; import { evaluateDecision, resolvePolicy, type ChannelPolicy, type Decision, type WhisperGateConfig } from "./rules.js"; +import { checkTurn, advanceTurn, resetTurn, onNewMessage, onSpeakerDone, initTurnOrder, getTurnDebugInfo } from "./turn-manager.js"; +import { startModeratorPresence, stopModeratorPresence } from "./moderator-presence.js"; type DiscordControlAction = "channel-private-create" | "channel-private-update" | "member-list"; @@ -22,11 +24,19 @@ type DebugConfig = { }; const sessionDecision = new Map(); +const sessionAllowed = new Map(); // Track if session was allowed to speak (true) or forced no-reply (false) +const sessionInjected = new Set(); // Track which sessions have already injected the end marker +const sessionChannelId = new Map(); // Track sessionKey -> channelId mapping +const sessionAccountId = new Map(); // Track sessionKey -> accountId mapping const MAX_SESSION_DECISIONS = 2000; const DECISION_TTL_MS = 5 * 60 * 1000; -function buildEndMarkerInstruction(endSymbols: string[]): string { +function buildEndMarkerInstruction(endSymbols: string[], isGroupChat: boolean): string { const symbols = endSymbols.length > 0 ? endSymbols.join("") : "🔚"; - return `你的这次发言必须以${symbols}作为结尾。除非你的回复是 gateway 关键词(如 NO_REPLY、HEARTBEAT_OK),这些关键词不要加${symbols}。`; + let instruction = `你的这次发言必须以${symbols}作为结尾。除非你的回复是 gateway 关键词(如 NO_REPLY、HEARTBEAT_OK),这些关键词不要加${symbols}。`; + if (isGroupChat) { + instruction += `\n\n群聊发言规则:如果这条消息与你无关、不需要你回应、或你没有有价值的补充,请主动回复 NO_REPLY。不要为了说话而说话。`; + } + return instruction; } const policyState: PolicyState = { @@ -42,6 +52,37 @@ function normalizeChannel(ctx: Record): string { return ""; } +/** + * Extract the actual Discord channel ID from a conversationId or "to" field. + * OpenClaw uses format "channel:" for Discord conversations. + * Also tries event.to and event.metadata.to as fallbacks. + */ +function extractDiscordChannelId(ctx: Record, event?: Record): string | undefined { + const candidates: unknown[] = [ + ctx.conversationId, + event?.to, + (event?.metadata as Record)?.to, + ]; + for (const c of candidates) { + if (typeof c === "string" && c.trim()) { + const s = c.trim(); + // Handle "channel:123456" format + if (s.startsWith("channel:")) { + const id = s.slice("channel:".length); + if (/^\d+$/.test(id)) return id; + } + // Handle "discord:channel:123456" format + if (s.startsWith("discord:channel:")) { + const id = s.slice("discord:channel:".length); + if (/^\d+$/.test(id)) return id; + } + // If it's a raw snowflake (all digits), use directly + if (/^\d{15,}$/.test(s)) return s; + } + } + return undefined; +} + function normalizeSender(event: Record, ctx: Record): string | undefined { const direct = [ctx.senderId, ctx.from, event.from]; for (const v of direct) { @@ -76,6 +117,7 @@ function extractUntrustedConversationInfo(text: string): Record function deriveDecisionInputFromPrompt( prompt: string, messageProvider?: string, + channelIdFromCtx?: string, ): { channel: string; channelId?: string; @@ -85,11 +127,25 @@ function deriveDecisionInputFromPrompt( } { const conv = extractUntrustedConversationInfo(prompt) || {}; const channel = (messageProvider || "").toLowerCase(); - const channelId = - (typeof conv.channel_id === "string" && conv.channel_id) || - (typeof conv.chat_id === "string" && conv.chat_id.startsWith("channel:") - ? conv.chat_id.slice("channel:".length) - : undefined); + + // Priority: ctx.channelId > conv.chat_id > conversation_label > conv.channel_id + let channelId = channelIdFromCtx; + if (!channelId) { + // Try chat_id field (format "channel:123456") + if (typeof conv.chat_id === "string" && conv.chat_id.startsWith("channel:")) { + channelId = conv.chat_id.slice("channel:".length); + } + // Try conversation_label (format "Guild #name channel id:123456") + if (!channelId && typeof conv.conversation_label === "string") { + const labelMatch = conv.conversation_label.match(/channel id:(\d+)/); + if (labelMatch) channelId = labelMatch[1]; + } + // Try channel_id field directly + if (!channelId && typeof conv.channel_id === "string" && conv.channel_id) { + channelId = conv.channel_id; + } + } + const senderId = (typeof conv.sender_id === "string" && conv.sender_id) || (typeof conv.sender === "string" && conv.sender) || @@ -159,6 +215,168 @@ function ensurePolicyStateLoaded(api: OpenClawPluginApi, config: WhisperGateConf } } +/** Resolve agentId → Discord accountId from config bindings */ +function resolveAccountId(api: OpenClawPluginApi, agentId: string): string | undefined { + const root = (api.config as Record) || {}; + const bindings = root.bindings as Array> | undefined; + if (!Array.isArray(bindings)) return undefined; + for (const b of bindings) { + if (b.agentId === agentId) { + const match = b.match as Record | undefined; + if (match?.channel === "discord" && typeof match.accountId === "string") { + return match.accountId; + } + } + } + return undefined; +} + +/** + * Get all Discord bot accountIds from config bindings. + */ +function getAllBotAccountIds(api: OpenClawPluginApi): string[] { + const root = (api.config as Record) || {}; + const bindings = root.bindings as Array> | undefined; + if (!Array.isArray(bindings)) return []; + const ids: string[] = []; + for (const b of bindings) { + const match = b.match as Record | undefined; + if (match?.channel === "discord" && typeof match.accountId === "string") { + ids.push(match.accountId); + } + } + return ids; +} + +/** + * Track which bot accountIds have been seen in each channel via message_received. + * Key: channelId, Value: Set of accountIds seen. + */ +const channelSeenAccounts = new Map>(); + +/** + * Record a bot accountId seen in a channel. + * Returns true if this is a new account for this channel (turn order should be updated). + */ +function recordChannelAccount(channelId: string, accountId: string): boolean { + let seen = channelSeenAccounts.get(channelId); + if (!seen) { + seen = new Set(); + channelSeenAccounts.set(channelId, seen); + } + if (seen.has(accountId)) return false; + seen.add(accountId); + return true; +} + +/** + * Get the list of bot accountIds seen in a channel. + * Only returns accounts that are also in the global bindings (actual bots). + */ +function getChannelBotAccountIds(api: OpenClawPluginApi, channelId: string): string[] { + const allBots = new Set(getAllBotAccountIds(api)); + const seen = channelSeenAccounts.get(channelId); + if (!seen) return []; + return [...seen].filter(id => allBots.has(id)); +} + +/** + * Ensure turn order is initialized for a channel. + * Uses only bot accounts that have been seen in this channel. + */ +function ensureTurnOrder(api: OpenClawPluginApi, channelId: string): void { + const botAccounts = getChannelBotAccountIds(api, channelId); + if (botAccounts.length > 0) { + initTurnOrder(channelId, botAccounts); + } +} + +/** + * Build agent identity string for injection into group chat prompts. + */ +function buildAgentIdentity(api: OpenClawPluginApi, agentId: string): string | undefined { + const root = (api.config as Record) || {}; + const bindings = root.bindings as Array> | undefined; + const agents = ((root.agents as Record)?.list as Array>) || []; + if (!Array.isArray(bindings)) return undefined; + + // Find accountId for this agent + let accountId: string | undefined; + for (const b of bindings) { + if (b.agentId === agentId) { + const match = b.match as Record | undefined; + if (match?.channel === "discord" && typeof match.accountId === "string") { + accountId = match.accountId; + break; + } + } + } + if (!accountId) return undefined; + + // Find agent name + const agent = agents.find((a: Record) => a.id === agentId); + const name = (agent?.name as string) || agentId; + + // Find Discord bot user ID from account token (not available directly) + // We'll use accountId as the identifier + return `你是 ${name}(Discord 账号: ${accountId})。`; +} + +// --- Moderator bot helpers --- + +/** Extract Discord user ID from a bot token (base64-encoded in first segment) */ +function userIdFromToken(token: string): string | undefined { + try { + const segment = token.split(".")[0]; + // Add padding + const padded = segment + "=".repeat((4 - (segment.length % 4)) % 4); + return Buffer.from(padded, "base64").toString("utf8"); + } catch { + return undefined; + } +} + +/** Resolve accountId → Discord user ID by reading the account's bot token from config */ +function resolveDiscordUserId(api: OpenClawPluginApi, accountId: string): string | undefined { + const root = (api.config as Record) || {}; + const channels = (root.channels as Record) || {}; + const discord = (channels.discord as Record) || {}; + const accounts = (discord.accounts as Record>) || {}; + const acct = accounts[accountId]; + if (!acct?.token || typeof acct.token !== "string") return undefined; + return userIdFromToken(acct.token); +} + +/** Get the moderator bot's Discord user ID from its token */ +function getModeratorUserId(config: WhisperGateConfig): string | undefined { + if (!config.moderatorBotToken) return undefined; + return userIdFromToken(config.moderatorBotToken); +} + +/** Send a message as the moderator bot via Discord REST API */ +async function sendModeratorMessage(token: string, channelId: string, content: string, logger: { info: (msg: string) => void; warn: (msg: string) => void }): Promise { + try { + const r = await fetch(`https://discord.com/api/v10/channels/${channelId}/messages`, { + method: "POST", + headers: { + "Authorization": `Bot ${token}`, + "Content-Type": "application/json", + }, + body: JSON.stringify({ content }), + }); + if (!r.ok) { + const text = await r.text(); + logger.warn(`whispergate: moderator send failed (${r.status}): ${text}`); + return false; + } + logger.info(`whispergate: moderator message sent to channel=${channelId}`); + return true; + } catch (err) { + logger.warn(`whispergate: moderator send error: ${String(err)}`); + return false; + } +} + function persistPolicies(api: OpenClawPluginApi): void { const filePath = policyState.filePath; if (!filePath) throw new Error("policy file path not initialized"); @@ -232,9 +450,15 @@ export default { const liveAtRegister = getLivePluginConfig(api, baseConfig as WhisperGateConfig); ensurePolicyStateLoaded(api, liveAtRegister); + // Start moderator bot presence (keep it "online" on Discord) + if (liveAtRegister.moderatorBotToken) { + startModeratorPresence(liveAtRegister.moderatorBotToken, api.logger); + api.logger.info("whispergate: moderator bot presence starting"); + } + api.registerTool( { - name: "whispergateway_tools", + name: "whispergate_tools", description: "WhisperGate unified tool: Discord admin actions + in-memory policy management.", parameters: { type: "object", @@ -242,7 +466,7 @@ export default { properties: { action: { type: "string", - enum: ["channel-private-create", "channel-private-update", "member-list", "policy-get", "policy-set-channel", "policy-delete-channel"], + enum: ["channel-private-create", "channel-private-update", "member-list", "policy-get", "policy-set-channel", "policy-delete-channel", "turn-status", "turn-advance", "turn-reset"], }, guildId: { type: "string" }, name: { type: "string" }, @@ -303,7 +527,7 @@ export default { const text = await r.text(); if (!r.ok) { return { - content: [{ type: "text", text: `whispergateway_tools discord failed (${r.status}): ${text}` }], + content: [{ type: "text", text: `whispergate_tools discord failed (${r.status}): ${text}` }], isError: true, }; } @@ -355,6 +579,26 @@ export default { } } + if (action === "turn-status") { + const channelId = String(params.channelId || "").trim(); + if (!channelId) return { content: [{ type: "text", text: "channelId is required" }], isError: true }; + return { content: [{ type: "text", text: JSON.stringify(getTurnDebugInfo(channelId), null, 2) }] }; + } + + if (action === "turn-advance") { + const channelId = String(params.channelId || "").trim(); + if (!channelId) return { content: [{ type: "text", text: "channelId is required" }], isError: true }; + const next = advanceTurn(channelId); + return { content: [{ type: "text", text: JSON.stringify({ ok: true, channelId, nextSpeaker: next, ...getTurnDebugInfo(channelId) }) }] }; + } + + if (action === "turn-reset") { + const channelId = String(params.channelId || "").trim(); + if (!channelId) return { content: [{ type: "text", text: "channelId is required" }], isError: true }; + resetTurn(channelId); + return { content: [{ type: "text", text: JSON.stringify({ ok: true, channelId, ...getTurnDebugInfo(channelId) }) }] }; + } + return { content: [{ type: "text", text: `unsupported action: ${action}` }], isError: true }; }, }, @@ -365,11 +609,51 @@ export default { try { const c = (ctx || {}) as Record; const e = (event || {}) as Record; - const preChannelId = typeof c.channelId === "string" ? c.channelId : undefined; + // ctx.channelId is the platform name (e.g. "discord"), NOT the Discord channel snowflake. + // Extract the real Discord channel ID from conversationId or event.to. + const preChannelId = extractDiscordChannelId(c, e); const livePre = getLivePluginConfig(api, baseConfig as WhisperGateConfig) as WhisperGateConfig & DebugConfig; if (shouldDebugLog(livePre, preChannelId)) { api.logger.info(`whispergate: debug message_received preflight ctx=${JSON.stringify(debugCtxSummary(c, e))}`); } + + // Turn management on message received + if (preChannelId) { + ensureTurnOrder(api, preChannelId); + // event.from is often the channel target (e.g. "discord:channel:xxx"), NOT the sender. + // The actual sender ID is in event.metadata.senderId. + const metadata = (e as Record).metadata as Record | undefined; + const from = (typeof metadata?.senderId === "string" && metadata.senderId) + || (typeof (e as Record).from === "string" ? (e as Record).from as string : ""); + + // Ignore moderator bot messages — they don't affect turn state + const moderatorUserId = getModeratorUserId(livePre); + if (moderatorUserId && from === moderatorUserId) { + if (shouldDebugLog(livePre, preChannelId)) { + api.logger.info(`whispergate: ignoring moderator message in channel=${preChannelId}`); + } + // Don't call onNewMessage — moderator messages are transparent to turn logic + } else { + const humanList = livePre.humanList || livePre.bypassUserIds || []; + const isHuman = humanList.includes(from); + const senderAccountId = typeof c.accountId === "string" ? c.accountId : undefined; + + // Track which bot accounts are present in this channel + if (senderAccountId && senderAccountId !== "default") { + const isNew = recordChannelAccount(preChannelId, senderAccountId); + if (isNew) { + // Re-initialize turn order with updated channel membership + ensureTurnOrder(api, preChannelId); + api.logger.info(`whispergate: new account ${senderAccountId} seen in channel=${preChannelId}, turn order updated`); + } + } + + onNewMessage(preChannelId, senderAccountId, isHuman); + if (shouldDebugLog(livePre, preChannelId)) { + api.logger.info(`whispergate: turn onNewMessage channel=${preChannelId} from=${from} isHuman=${isHuman} accountId=${senderAccountId ?? "unknown"}`); + } + } + } } catch (err) { api.logger.warn(`whispergate: message hook failed: ${String(err)}`); } @@ -391,11 +675,25 @@ export default { ); } - const derived = deriveDecisionInputFromPrompt(prompt, ctx.messageProvider); + const derived = deriveDecisionInputFromPrompt(prompt, ctx.messageProvider, ctx.channelId); + // Fallback: extract channelId from sessionKey (format "agent::discord:channel:") + if (!derived.channelId && key) { + const skMatch = key.match(/:channel:(\d+)$/); + if (skMatch) derived.channelId = skMatch[1]; + } // Only proceed if: discord channel AND prompt contains untrusted metadata const hasConvMarker = prompt.includes("Conversation info (untrusted metadata):"); if (live.discordOnly !== false && (!hasConvMarker || derived.channel !== "discord")) return; + // Always save channelId and accountId mappings for use in later hooks + if (derived.channelId) { + sessionChannelId.set(key, derived.channelId); + } + const resolvedAccountId = resolveAccountId(api, ctx.agentId || ""); + if (resolvedAccountId) { + sessionAccountId.set(key, resolvedAccountId); + } + let rec = sessionDecision.get(key); if (!rec || Date.now() - rec.createdAt > DECISION_TTL_MS) { if (rec) sessionDecision.delete(key); @@ -422,6 +720,29 @@ export default { } } + // Turn-based check: ALWAYS check turn order regardless of evaluateDecision result. + // This ensures only the current speaker can respond even for human messages. + if (derived.channelId) { + ensureTurnOrder(api, derived.channelId); + const accountId = resolveAccountId(api, ctx.agentId || ""); + if (accountId) { + const turnCheck = checkTurn(derived.channelId, accountId); + if (!turnCheck.allowed) { + // Forced no-reply - record this session as not allowed to speak + sessionAllowed.set(key, false); + api.logger.info( + `whispergate: turn gate blocked session=${key} accountId=${accountId} currentSpeaker=${turnCheck.currentSpeaker} reason=${turnCheck.reason}`, + ); + return { + providerOverride: live.noReplyProvider, + modelOverride: live.noReplyModel, + }; + } + // Allowed to speak - record this session as allowed + sessionAllowed.set(key, true); + } + } + if (!rec.decision.shouldUseNoReply) { // 如果之前有 no-reply 执行过,现在不需要了,清除 override 恢复原模型 if (rec.needsRestore) { @@ -475,7 +796,7 @@ export default { if (rec) sessionDecision.delete(key); const prompt = ((event as Record).prompt as string) || ""; - const derived = deriveDecisionInputFromPrompt(prompt, ctx.messageProvider); + const derived = deriveDecisionInputFromPrompt(prompt, ctx.messageProvider, ctx.channelId); const decision = evaluateDecision({ config: live, @@ -499,6 +820,17 @@ export default { } sessionDecision.delete(key); + + // Only inject once per session (one-time injection) + if (sessionInjected.has(key)) { + if (shouldDebugLog(live, undefined)) { + api.logger.info( + `whispergate: debug before_prompt_build session=${key} inject skipped (already injected)`, + ); + } + return; + } + if (!rec.decision.shouldInjectEndMarkerPrompt) { if (shouldDebugLog(live, undefined)) { api.logger.info( @@ -510,12 +842,264 @@ export default { // Resolve end symbols from config/policy for dynamic instruction const prompt = ((event as Record).prompt as string) || ""; - const derived = deriveDecisionInputFromPrompt(prompt, ctx.messageProvider); + const derived = deriveDecisionInputFromPrompt(prompt, ctx.messageProvider, ctx.channelId); const policy = resolvePolicy(live, derived.channelId, policyState.channelPolicies); - const instruction = buildEndMarkerInstruction(policy.endSymbols); + const isGroupChat = derived.conv.is_group_chat === true || derived.conv.is_group_chat === "true"; + const instruction = buildEndMarkerInstruction(policy.endSymbols, isGroupChat); - api.logger.info(`whispergate: prepend end marker instruction for session=${key}, reason=${rec.decision.reason}`); - return { prependContext: instruction }; + // Inject agent identity for group chats + let identity = ""; + if (isGroupChat && ctx.agentId) { + const idStr = buildAgentIdentity(api, ctx.agentId); + if (idStr) identity = idStr + "\n\n"; + } + + // Mark session as injected (one-time injection) + sessionInjected.add(key); + + api.logger.info(`whispergate: prepend end marker instruction for session=${key}, reason=${rec.decision.reason} isGroupChat=${isGroupChat}`); + return { prependContext: identity + instruction }; + }); + + // Register slash commands for Discord + api.registerCommand({ + name: "whispergate", + description: "WhisperGate 频道策略管理", + acceptsArgs: true, + handler: async (cmdCtx) => { + const args = cmdCtx.args || ""; + const parts = args.trim().split(/\s+/); + const subCmd = parts[0] || "help"; + + if (subCmd === "help") { + return { text: `WhisperGate 命令:\n` + + `/whispergate status - 显示当前频道状态\n` + + `/whispergate turn-status - 显示轮流发言状态\n` + + `/whispergate turn-advance - 手动推进轮流\n` + + `/whispergate turn-reset - 重置轮流顺序` }; + } + + if (subCmd === "status") { + return { text: JSON.stringify({ policies: policyState.channelPolicies }, null, 2) }; + } + + if (subCmd === "turn-status") { + const channelId = cmdCtx.channelId; + if (!channelId) return { text: "无法获取频道ID", isError: true }; + return { text: JSON.stringify(getTurnDebugInfo(channelId), null, 2) }; + } + + if (subCmd === "turn-advance") { + const channelId = cmdCtx.channelId; + if (!channelId) return { text: "无法获取频道ID", isError: true }; + const next = advanceTurn(channelId); + return { text: JSON.stringify({ ok: true, nextSpeaker: next }) }; + } + + if (subCmd === "turn-reset") { + const channelId = cmdCtx.channelId; + if (!channelId) return { text: "无法获取频道ID", isError: true }; + resetTurn(channelId); + return { text: JSON.stringify({ ok: true }) }; + } + + return { text: `未知子命令: ${subCmd}`, isError: true }; + }, + }); + + // Handle NO_REPLY detection before message write + // This is where we detect if agent output is NO_REPLY and handle turn advancement + // NOTE: This hook is synchronous, do not use async/await + api.on("before_message_write", (event, ctx) => { + try { + // Debug: print all available keys in event and ctx + api.logger.info( + `whispergate: DEBUG before_message_write eventKeys=${JSON.stringify(Object.keys(event ?? {}))} ctxKeys=${JSON.stringify(Object.keys(ctx ?? {}))}`, + ); + + // before_message_write ctx only has { agentId, sessionKey }. + // Use session mappings populated during before_model_resolve for channelId/accountId. + // Content comes from event.message (AgentMessage). + let key = ctx.sessionKey; + let channelId: string | undefined; + let accountId: string | undefined; + + // Get from session mapping (set in before_model_resolve) + if (key) { + channelId = sessionChannelId.get(key); + accountId = sessionAccountId.get(key); + } + + // Extract content from event.message (AgentMessage) + let content = ""; + const msg = (event as Record).message as Record | undefined; + if (msg) { + // AgentMessage may have content as string or nested + if (typeof msg.content === "string") { + content = msg.content; + } else if (Array.isArray(msg.content)) { + // content might be an array of parts (Anthropic format) + for (const part of msg.content) { + if (typeof part === "string") content += part; + else if (part && typeof part === "object" && typeof (part as Record).text === "string") { + content += (part as Record).text; + } + } + } + } + // Fallback to event.content + if (!content) { + content = ((event as Record).content as string) || ""; + } + + // Always log for debugging - show all available info + api.logger.info( + `whispergate: DEBUG before_message_write session=${key ?? "undefined"} channel=${channelId ?? "undefined"} accountId=${accountId ?? "undefined"} contentType=${typeof content} content=${String(content).slice(0, 200)}`, + ); + + if (!key || !channelId || !accountId) return; + + const live = getLivePluginConfig(api, baseConfig as WhisperGateConfig) as WhisperGateConfig & DebugConfig; + + const trimmed = content.trim(); + const isNoReply = /^NO_REPLY$/i.test(trimmed); + + // Log turn state for debugging + const turnDebug = getTurnDebugInfo(channelId); + api.logger.info( + `whispergate: DEBUG turn state channel=${channelId} turnOrder=${JSON.stringify(turnDebug.turnOrder)} currentSpeaker=${turnDebug.currentSpeaker} noRepliedThisCycle=${JSON.stringify([...turnDebug.noRepliedThisCycle])}`, + ); + + if (!isNoReply) { + api.logger.info( + `whispergate: before_message_write content is not NO_REPLY, skipping channel=${channelId}`, + ); + return; + } + + // Check if this session was forced no-reply or allowed to speak + const wasAllowed = sessionAllowed.get(key); + api.logger.info( + `whispergate: DEBUG NO_REPLY detected session=${key} wasAllowed=${wasAllowed}`, + ); + + if (wasAllowed === undefined) return; // No record, skip + + if (wasAllowed === false) { + // Forced no-reply - do not advance turn + sessionAllowed.delete(key); + api.logger.info( + `whispergate: before_message_write forced no-reply session=${key} channel=${channelId} - not advancing turn`, + ); + return; + } + + // Allowed to speak (current speaker) but chose NO_REPLY - advance turn + ensureTurnOrder(api, channelId); + const nextSpeaker = onSpeakerDone(channelId, accountId, true); + + sessionAllowed.delete(key); + + api.logger.info( + `whispergate: before_message_write real no-reply session=${key} channel=${channelId} nextSpeaker=${nextSpeaker ?? "dormant"}`, + ); + + // If all agents NO_REPLY'd (dormant), don't trigger handoff + if (!nextSpeaker) { + if (shouldDebugLog(live, channelId)) { + api.logger.info( + `whispergate: before_message_write all agents no-reply, going dormant - no handoff`, + ); + } + return; + } + + // Trigger moderator handoff message (fire-and-forget, don't await) + if (live.moderatorBotToken) { + const nextUserId = resolveDiscordUserId(api, nextSpeaker); + if (nextUserId) { + const handoffMsg = `轮到(<@${nextUserId}>)了,如果没有想说的请直接回复NO_REPLY`; + void sendModeratorMessage(live.moderatorBotToken, channelId, handoffMsg, api.logger).catch((err) => { + api.logger.warn(`whispergate: before_message_write handoff failed: ${String(err)}`); + }); + } else { + api.logger.warn(`whispergate: cannot resolve Discord userId for next speaker accountId=${nextSpeaker}`); + } + } + } catch (err) { + api.logger.warn(`whispergate: before_message_write hook failed: ${String(err)}`); + } + }); + + // Turn advance: when an agent sends a message, check if it signals end of turn + api.on("message_sent", async (event, ctx) => { + try { + const key = ctx.sessionKey; + const c = (ctx || {}) as Record; + const e = (event || {}) as Record; + + // Always log raw context first for debugging + api.logger.info( + `whispergate: DEBUG message_sent RAW ctxKeys=${JSON.stringify(Object.keys(c))} eventKeys=${JSON.stringify(Object.keys(e))} ` + + `ctx.channelId=${String(c.channelId ?? "undefined")} ctx.conversationId=${String(c.conversationId ?? "undefined")} ` + + `ctx.accountId=${String(c.accountId ?? "undefined")} event.to=${String(e.to ?? "undefined")} ` + + `session=${key ?? "undefined"}`, + ); + + // ctx.channelId is the platform name (e.g. "discord"), NOT the Discord channel snowflake. + // Extract real Discord channel ID from conversationId or event.to. + let channelId = extractDiscordChannelId(c, e); + // Fallback: sessionKey mapping + if (!channelId && key) { + channelId = sessionChannelId.get(key); + } + // Fallback: parse from sessionKey + if (!channelId && key) { + const skMatch = key.match(/:channel:(\d+)$/); + if (skMatch) channelId = skMatch[1]; + } + const accountId = (ctx.accountId as string | undefined) || (key ? sessionAccountId.get(key) : undefined); + const content = (event.content as string) || ""; + + // Debug log + api.logger.info( + `whispergate: DEBUG message_sent RESOLVED session=${key ?? "undefined"} channelId=${channelId ?? "undefined"} accountId=${accountId ?? "undefined"} content=${content.slice(0, 100)}`, + ); + + if (!channelId || !accountId) return; + + const live = getLivePluginConfig(api, baseConfig as WhisperGateConfig) as WhisperGateConfig & DebugConfig; + ensurePolicyStateLoaded(api, live); + const policy = resolvePolicy(live, channelId, policyState.channelPolicies); + + const trimmed = content.trim(); + const isEmpty = trimmed.length === 0; + const isNoReply = /^NO_REPLY$/i.test(trimmed) || /^HEARTBEAT_OK$/i.test(trimmed); + const lastChar = trimmed.length > 0 ? Array.from(trimmed).pop() || "" : ""; + const hasEndSymbol = !!lastChar && policy.endSymbols.includes(lastChar); + const wasNoReply = isEmpty || isNoReply; + + if (wasNoReply || hasEndSymbol) { + const nextSpeaker = onSpeakerDone(channelId, accountId, wasNoReply); + const trigger = wasNoReply ? (isEmpty ? "empty" : "no_reply_keyword") : "end_symbol"; + api.logger.info( + `whispergate: turn onSpeakerDone channel=${channelId} from=${accountId} next=${nextSpeaker ?? "dormant"} trigger=${trigger}`, + ); + // Moderator handoff: when current speaker NO_REPLY'd and there's a next speaker, + // send a handoff message via the moderator bot to trigger the next agent + if (wasNoReply && nextSpeaker && live.moderatorBotToken) { + const nextUserId = resolveDiscordUserId(api, nextSpeaker); + if (nextUserId) { + const handoffMsg = `轮到(<@${nextUserId}>)了,如果没有想说的请直接回复NO_REPLY`; + sendModeratorMessage(live.moderatorBotToken, channelId, handoffMsg, api.logger); + } else { + api.logger.warn(`whispergate: cannot resolve Discord userId for next speaker accountId=${nextSpeaker}`); + } + } + } + } catch (err) { + api.logger.warn(`whispergate: message_sent hook failed: ${String(err)}`); + } }); }, }; diff --git a/plugin/moderator-presence.ts b/plugin/moderator-presence.ts new file mode 100644 index 0000000..dfe311e --- /dev/null +++ b/plugin/moderator-presence.ts @@ -0,0 +1,257 @@ +/** + * Minimal Discord Gateway connection to keep the moderator bot "online". + * Uses Node.js built-in WebSocket (Node 22+). + * + * IMPORTANT: Only ONE instance should exist per bot token. + * Uses a singleton guard to prevent multiple connections. + */ + +let ws: WebSocket | null = null; +let heartbeatInterval: ReturnType | null = null; +let heartbeatAcked = true; +let lastSequence: number | null = null; +let sessionId: string | null = null; +let resumeUrl: string | null = null; +let reconnectTimer: ReturnType | null = null; +let destroyed = false; +let started = false; // singleton guard + +type Logger = { + info: (msg: string) => void; + warn: (msg: string) => void; +}; + +const GATEWAY_URL = "wss://gateway.discord.gg/?v=10&encoding=json"; +const MAX_RECONNECT_DELAY_MS = 60_000; +let reconnectAttempts = 0; + +function sendPayload(data: Record) { + if (ws?.readyState === WebSocket.OPEN) { + ws.send(JSON.stringify(data)); + } +} + +function startHeartbeat(intervalMs: number) { + stopHeartbeat(); + heartbeatAcked = true; + + // First heartbeat after jitter + const jitter = Math.floor(Math.random() * intervalMs); + const firstTimer = setTimeout(() => { + if (destroyed) return; + if (!heartbeatAcked) { + // Missed ACK — zombie connection, close and reconnect + ws?.close(4000, "missed heartbeat ack"); + return; + } + heartbeatAcked = false; + sendPayload({ op: 1, d: lastSequence }); + + heartbeatInterval = setInterval(() => { + if (destroyed) return; + if (!heartbeatAcked) { + ws?.close(4000, "missed heartbeat ack"); + return; + } + heartbeatAcked = false; + sendPayload({ op: 1, d: lastSequence }); + }, intervalMs); + }, jitter); + + // Store the first timer so we can clear it + heartbeatInterval = firstTimer as unknown as ReturnType; +} + +function stopHeartbeat() { + if (heartbeatInterval) { + clearInterval(heartbeatInterval); + clearTimeout(heartbeatInterval as unknown as ReturnType); + heartbeatInterval = null; + } +} + +function cleanup() { + stopHeartbeat(); + if (ws) { + // Remove all handlers to avoid ghost callbacks + ws.onopen = null; + ws.onmessage = null; + ws.onclose = null; + ws.onerror = null; + try { ws.close(1000); } catch { /* ignore */ } + ws = null; + } +} + +function connect(token: string, logger: Logger, isResume = false) { + if (destroyed) return; + + // Clean up any existing connection first + cleanup(); + + const url = isResume && resumeUrl ? resumeUrl : GATEWAY_URL; + + try { + ws = new WebSocket(url); + } catch (err) { + logger.warn(`whispergate: moderator ws constructor failed: ${String(err)}`); + scheduleReconnect(token, logger, false); + return; + } + + const currentWs = ws; // capture for closure + + ws.onopen = () => { + if (currentWs !== ws || destroyed) return; // stale + + reconnectAttempts = 0; // reset on successful open + + if (isResume && sessionId) { + sendPayload({ + op: 6, + d: { token, session_id: sessionId, seq: lastSequence }, + }); + } else { + sendPayload({ + op: 2, + d: { + token, + intents: 0, + properties: { + os: "linux", + browser: "whispergate", + device: "whispergate", + }, + presence: { + status: "online", + activities: [{ name: "Moderating", type: 3 }], + }, + }, + }); + } + }; + + ws.onmessage = (evt: MessageEvent) => { + if (currentWs !== ws || destroyed) return; + + try { + const msg = JSON.parse(typeof evt.data === "string" ? evt.data : String(evt.data)); + const { op, t, s, d } = msg; + + if (s != null) lastSequence = s; + + switch (op) { + case 10: // Hello + startHeartbeat(d.heartbeat_interval); + break; + case 11: // Heartbeat ACK + heartbeatAcked = true; + break; + case 1: // Heartbeat request + sendPayload({ op: 1, d: lastSequence }); + break; + case 0: // Dispatch + if (t === "READY") { + sessionId = d.session_id; + resumeUrl = d.resume_gateway_url; + logger.info("whispergate: moderator bot connected and online"); + } + if (t === "RESUMED") { + logger.info("whispergate: moderator bot resumed"); + } + break; + case 7: // Reconnect request + logger.info("whispergate: moderator bot reconnect requested by Discord"); + cleanup(); + scheduleReconnect(token, logger, true); + break; + case 9: // Invalid Session + logger.warn(`whispergate: moderator bot invalid session, resumable=${d}`); + cleanup(); + sessionId = d ? sessionId : null; + // Wait longer before re-identifying + setTimeout(() => { + if (!destroyed) connect(token, logger, !!d && !!sessionId); + }, 3000 + Math.random() * 2000); + break; + } + } catch { + // ignore parse errors + } + }; + + ws.onclose = (evt: CloseEvent) => { + if (currentWs !== ws) return; // stale ws + stopHeartbeat(); + if (destroyed) return; + + const code = evt.code; + + // Non-recoverable codes — stop reconnecting + if (code === 4004) { + logger.warn("whispergate: moderator bot token invalid (4004), stopping"); + started = false; + return; + } + if (code === 4010 || code === 4011 || code === 4013 || code === 4014) { + logger.warn(`whispergate: moderator bot fatal close (${code}), re-identifying`); + sessionId = null; + scheduleReconnect(token, logger, false); + return; + } + + logger.info(`whispergate: moderator bot disconnected (code=${code}), will reconnect`); + const canResume = !!sessionId && code !== 4012; + scheduleReconnect(token, logger, canResume); + }; + + ws.onerror = () => { + // onclose will fire after this + }; +} + +function scheduleReconnect(token: string, logger: Logger, resume: boolean) { + if (destroyed) return; + if (reconnectTimer) clearTimeout(reconnectTimer); + + // Exponential backoff with cap + reconnectAttempts++; + const baseDelay = Math.min(1000 * Math.pow(2, reconnectAttempts), MAX_RECONNECT_DELAY_MS); + const jitter = Math.random() * 1000; + const delay = baseDelay + jitter; + + logger.info(`whispergate: moderator reconnect in ${Math.round(delay)}ms (attempt ${reconnectAttempts})`); + + reconnectTimer = setTimeout(() => { + reconnectTimer = null; + connect(token, logger, resume); + }, delay); +} + +/** + * Start the moderator bot's Discord Gateway connection. + * Singleton: calling multiple times with the same token is safe (no-op). + */ +export function startModeratorPresence(token: string, logger: Logger): void { + if (started) { + logger.info("whispergate: moderator presence already started, skipping"); + return; + } + started = true; + destroyed = false; + reconnectAttempts = 0; + connect(token, logger); +} + +/** + * Disconnect the moderator bot. + */ +export function stopModeratorPresence(): void { + destroyed = true; + started = false; + if (reconnectTimer) { + clearTimeout(reconnectTimer); + reconnectTimer = null; + } + cleanup(); +} diff --git a/plugin/openclaw.plugin.json b/plugin/openclaw.plugin.json index d10f3e5..1cb7fab 100644 --- a/plugin/openclaw.plugin.json +++ b/plugin/openclaw.plugin.json @@ -24,7 +24,8 @@ "discordControlApiToken": { "type": "string" }, "discordControlCallerId": { "type": "string" }, "enableDebugLogs": { "type": "boolean", "default": false }, - "debugLogChannelIds": { "type": "array", "items": { "type": "string" }, "default": [] } + "debugLogChannelIds": { "type": "array", "items": { "type": "string" }, "default": [] }, + "moderatorBotToken": { "type": "string" } }, "required": ["noReplyProvider", "noReplyModel"] } diff --git a/plugin/rules.ts b/plugin/rules.ts index ca00593..9abcda9 100644 --- a/plugin/rules.ts +++ b/plugin/rules.ts @@ -10,6 +10,8 @@ export type WhisperGateConfig = { endSymbols?: string[]; noReplyProvider: string; noReplyModel: string; + /** Discord bot token for the moderator bot (used for turn handoff messages) */ + moderatorBotToken?: string; }; export type ChannelPolicy = { diff --git a/plugin/turn-manager.ts b/plugin/turn-manager.ts new file mode 100644 index 0000000..033699e --- /dev/null +++ b/plugin/turn-manager.ts @@ -0,0 +1,233 @@ +/** + * Turn-based speaking manager for group channels. + * + * Rules: + * - Humans (humanList) are never in the turn order + * - Turn order is auto-populated from channel/server members minus humans + * - currentSpeaker can be null (dormant state) + * - When ALL agents in a cycle have NO_REPLY'd, state goes dormant (null) + * - Dormant → any new message reactivates: + * - If sender is NOT in turn order → current = first in list + * - If sender IS in turn order → current = next after sender + */ + +export type ChannelTurnState = { + /** Ordered accountIds for this channel (auto-populated, shuffled) */ + turnOrder: string[]; + /** Current speaker accountId, or null if dormant */ + currentSpeaker: string | null; + /** Set of accountIds that have NO_REPLY'd this cycle */ + noRepliedThisCycle: Set; + /** Timestamp of last state change */ + lastChangedAt: number; +}; + +const channelTurns = new Map(); + +/** Turn timeout: if the current speaker hasn't responded, auto-advance */ +const TURN_TIMEOUT_MS = 60_000; + +// --- helpers --- + +function shuffleArray(arr: T[]): T[] { + const a = [...arr]; + for (let i = a.length - 1; i > 0; i--) { + const j = Math.floor(Math.random() * (i + 1)); + [a[i], a[j]] = [a[j], a[i]]; + } + return a; +} + +// --- public API --- + +/** + * Initialize or update the turn order for a channel. + * Called with the list of bot accountIds (already filtered, humans excluded). + */ +export function initTurnOrder(channelId: string, botAccountIds: string[]): void { + const existing = channelTurns.get(channelId); + if (existing) { + // Check if membership changed + const oldSet = new Set(existing.turnOrder); + const newSet = new Set(botAccountIds); + const same = oldSet.size === newSet.size && [...oldSet].every(id => newSet.has(id)); + if (same) return; // no change + } + + channelTurns.set(channelId, { + turnOrder: shuffleArray(botAccountIds), + currentSpeaker: null, // start dormant + noRepliedThisCycle: new Set(), + lastChangedAt: Date.now(), + }); +} + +/** + * Check if the given accountId is allowed to speak. + */ +export function checkTurn(channelId: string, accountId: string): { + allowed: boolean; + currentSpeaker: string | null; + reason: string; +} { + const state = channelTurns.get(channelId); + if (!state || state.turnOrder.length === 0) { + return { allowed: true, currentSpeaker: null, reason: "no_turn_state" }; + } + + // Not in turn order (human or unknown) → always allowed + if (!state.turnOrder.includes(accountId)) { + return { allowed: true, currentSpeaker: state.currentSpeaker, reason: "not_in_turn_order" }; + } + + // Dormant → not allowed (will be activated by onNewMessage) + if (state.currentSpeaker === null) { + return { allowed: false, currentSpeaker: null, reason: "dormant" }; + } + + // Check timeout → auto-advance + if (Date.now() - state.lastChangedAt > TURN_TIMEOUT_MS) { + advanceTurn(channelId); + // Re-check after advance + const updated = channelTurns.get(channelId)!; + if (updated.currentSpeaker === accountId) { + return { allowed: true, currentSpeaker: updated.currentSpeaker, reason: "timeout_advanced_to_self" }; + } + return { allowed: false, currentSpeaker: updated.currentSpeaker, reason: "timeout_advanced_to_other" }; + } + + if (accountId === state.currentSpeaker) { + return { allowed: true, currentSpeaker: state.currentSpeaker, reason: "is_current_speaker" }; + } + + return { allowed: false, currentSpeaker: state.currentSpeaker, reason: "not_current_speaker" }; +} + +/** + * Called when a new message arrives in the channel. + * Handles reactivation from dormant state and human-triggered resets. + * + * @param senderAccountId - the accountId of the message sender (could be human/bot/unknown) + * @param isHuman - whether the sender is in the humanList + */ +export function onNewMessage(channelId: string, senderAccountId: string | undefined, isHuman: boolean): void { + const state = channelTurns.get(channelId); + if (!state || state.turnOrder.length === 0) return; + + if (isHuman) { + // Human message: activate, start from first in order + state.currentSpeaker = state.turnOrder[0]; + state.noRepliedThisCycle = new Set(); + state.lastChangedAt = Date.now(); + return; + } + + if (state.currentSpeaker !== null) { + // Already active, no change needed from incoming message + return; + } + + // Dormant state + non-human message → reactivate + if (senderAccountId && state.turnOrder.includes(senderAccountId)) { + // Sender is in turn order → next after sender + const idx = state.turnOrder.indexOf(senderAccountId); + const nextIdx = (idx + 1) % state.turnOrder.length; + state.currentSpeaker = state.turnOrder[nextIdx]; + } else { + // Sender not in turn order → start from first + state.currentSpeaker = state.turnOrder[0]; + } + state.noRepliedThisCycle = new Set(); + state.lastChangedAt = Date.now(); +} + +/** + * Called when the current speaker finishes (end symbol detected) or says NO_REPLY. + * @param wasNoReply - true if the speaker said NO_REPLY (empty/silent) + * @returns the new currentSpeaker (or null if dormant) + */ +export function onSpeakerDone(channelId: string, accountId: string, wasNoReply: boolean): string | null { + const state = channelTurns.get(channelId); + if (!state) return null; + if (state.currentSpeaker !== accountId) return state.currentSpeaker; // not current speaker, ignore + + if (wasNoReply) { + state.noRepliedThisCycle.add(accountId); + + // Check if ALL agents have NO_REPLY'd this cycle + const allNoReplied = state.turnOrder.every(id => state.noRepliedThisCycle.has(id)); + if (allNoReplied) { + // Go dormant + state.currentSpeaker = null; + state.noRepliedThisCycle = new Set(); + state.lastChangedAt = Date.now(); + return null; + } + } else { + // Successful speech resets the cycle counter + state.noRepliedThisCycle = new Set(); + } + + return advanceTurn(channelId); +} + +/** + * Advance to next speaker in order. + */ +export function advanceTurn(channelId: string): string | null { + const state = channelTurns.get(channelId); + if (!state || state.turnOrder.length === 0) return null; + + if (state.currentSpeaker === null) return null; + + const idx = state.turnOrder.indexOf(state.currentSpeaker); + const nextIdx = (idx + 1) % state.turnOrder.length; + + // Skip agents that already NO_REPLY'd this cycle + let attempts = 0; + let candidateIdx = nextIdx; + while (state.noRepliedThisCycle.has(state.turnOrder[candidateIdx]) && attempts < state.turnOrder.length) { + candidateIdx = (candidateIdx + 1) % state.turnOrder.length; + attempts++; + } + + if (attempts >= state.turnOrder.length) { + // All have NO_REPLY'd + state.currentSpeaker = null; + state.lastChangedAt = Date.now(); + return null; + } + + state.currentSpeaker = state.turnOrder[candidateIdx]; + state.lastChangedAt = Date.now(); + return state.currentSpeaker; +} + +/** + * Force reset: go dormant. + */ +export function resetTurn(channelId: string): void { + const state = channelTurns.get(channelId); + if (state) { + state.currentSpeaker = null; + state.noRepliedThisCycle = new Set(); + state.lastChangedAt = Date.now(); + } +} + +/** + * Get debug info. + */ +export function getTurnDebugInfo(channelId: string): Record { + const state = channelTurns.get(channelId); + if (!state) return { channelId, hasTurnState: false }; + return { + channelId, + hasTurnState: true, + turnOrder: state.turnOrder, + currentSpeaker: state.currentSpeaker, + noRepliedThisCycle: [...state.noRepliedThisCycle], + lastChangedAt: state.lastChangedAt, + dormant: state.currentSpeaker === null, + }; +} diff --git a/scripts/package-plugin.mjs b/scripts/package-plugin.mjs index 9e65873..29be6d3 100644 --- a/scripts/package-plugin.mjs +++ b/scripts/package-plugin.mjs @@ -8,7 +8,7 @@ const outDir = path.join(root, "dist", "whispergate"); fs.rmSync(outDir, { recursive: true, force: true }); fs.mkdirSync(outDir, { recursive: true }); -for (const f of ["index.ts", "rules.ts", "openclaw.plugin.json", "README.md", "package.json"]) { +for (const f of ["index.ts", "rules.ts", "turn-manager.ts", "moderator-presence.ts", "openclaw.plugin.json", "README.md", "package.json"]) { fs.copyFileSync(path.join(pluginDir, f), path.join(outDir, f)); }