feat(fabric): dynamic-subscription via fabric-register openclaw tool

The plugin manifest declared `fabric-register` as a tool name but
tools.ts never registered it — recruitment fell through to the
standalone `/root/.openclaw/bin/fabric-register` binary, which writes
~/.openclaw/fabric-identity.json correctly but exits without notifying
the running plugin. That left fabric inbound's subscription as a
connect-time snapshot: every new agent required a gateway restart
between `new-agent` and the interview's sub-discussion or the message
had no socket to dispatch on.

Wire the full path:
  - `FabricInbound.addAccount(entry)` — login, upsert identity, open socket(s),
    track per-agent so removeAccount can teardown cleanly. Idempotent: a
    second call replaces the previous socket (used post-onboard when the
    agent rotates off the shared `interviewee` placeholder onto its own
    apikey).
  - `FabricInbound.removeAccount(agentId)` — disconnect sockets, clear
    timers + per-agent caches.
  - `__fabric.addAccount` / `removeAccount` — cross-plugin bridge so the
    `fabric-register` tool can reach the live FabricInbound instance from
    its tool handler context.
  - `fabric-register` openclaw tool — validates apiKey, calls
    `__fabric.addAccount`, returns `{ok, fabricUserId, displayName}`.
    Accepts `agentId` arg so recruitment can bind on behalf of a
    freshly-created agent before that agent has a session of its own.

Removes the "restart the gateway" advice from the ctxGuild
"agent not registered" error message — operators should now call the
tool path instead.

After this lands + the ClawSkills register-agent script flip (separate
commit), `recruitment.new-agent` -> interviewer sub-discussion runs
without a gateway restart in between.

README.md

@@ -68,6 +68,16 @@ Two ways, both write the same identity registry the transport reads:
 ## Config
 
 - `channels.fabric.centerApiBase` — e.g. `http://localhost:7001/api`
+- `channels.fabric.commandsSyncKey` — **required**; must equal the guild's
+  `FABRIC_BACKEND_GUILD_COMMANDS_SYNC_KEY` (Guild C-2). Read it from the
+  guild with `docker exec fabric-backend-guild node dist/cli/print-commands-sync-key.js`.
+  Sourced from config only — never from the environment.
+- `channels.fabric.coalesce` — default `true`. OpenClaw splits one agent
+  turn whose blocks are `text → thinking/tool → text` into multiple
+  `deliver()` calls; this buffers them and posts ONE Fabric message at the
+  deterministic turn boundary (right after the inbound reply dispatch
+  resolves — no hooks, no timers, no idle guessing). `false` = raw
+  per-segment posting.
 - `channels.fabric.accounts.<agentId>` = `{ fabricApiKey, enabled }`
   (**agent = account**; the account id is the OpenClaw agentId)
 - plugin `identityFilePath` — default `~/.openclaw/fabric-identity.json`
@@ -100,6 +110,20 @@ Then `openclaw gateway restart`.
   `members` (list the channel's member userIds) · `join` (this agent
   joins) · `leave` (this agent leaves).
 
+## Slash-command catalog
+
+On `gateway_start` the plugin syncs OpenClaw's native-command catalog to
+each guild (`command-sync.ts`: `listNativeCommandSpecsForConfig` +
+`findCommandByNativeName`, dynamic arg `choices` snapshotted via
+`resolveCommandArgChoices`) → `PUT /api/commands`. This is exactly the data
+Discord registers as slash commands; Fabric's frontend uses it for `/`
+autocomplete. Fabric deliberately does **not** advertise the
+`nativeCommands` channel capability — it stays a text-command surface, so a
+`/<cmd>` message is delivered normally and OpenClaw's command system
+executes it (text-command + command session + auth), reusing the standard
+inbound path. Only `/no-reply` and `/force-proceed` stay
+server-intercepted by the guild.
+
 ## Install / build
 
 ```bash

dist/fabric/index.js

@@ -5,16 +5,24 @@
 // the OpenClawPluginApi for runtime startup (transport + tools).
 import { defineChannelPluginEntry } from 'openclaw/plugin-sdk/core';
 import { fabricChannelPlugin } from './src/channel.js';
+import { flushAllFabric } from './src/coalesce.js';
+import { getChannelType, flushChannelMeta } from './src/channel-meta.js';
 import { FabricInbound } from './src/inbound.js';
 import { listEnabledFabricAccounts } from './src/accounts.js';
 import { registerFabricTools } from './src/tools.js';
 import { FabricClient } from './src/fabric-client.js';
 import { IdentityRegistry } from './src/identity.js';
 import { syncFabricCommands } from './src/command-sync.js';
+import { PresenceSync } from './src/presence-sync.js';
 import path from 'node:path';
 import os from 'node:os';
 let runtimeRef = null;
 let inbound = null;
+let presence = null;
+// Periodic re-harvest of presence accounts so newly-connected agents
+// (registered through tool-based identity flow AFTER initial start)
+// get picked up. Cleared on gateway_stop.
+let presenceRefreshTimer = null;
 export { fabricChannelPlugin } from './src/channel.js';
 export default defineChannelPluginEntry({
     id: 'fabric',
@@ -36,6 +44,29 @@ export default defineChannelPluginEntry({
         const client = new FabricClient(centerApiBase);
         const identity = new IdentityRegistry(idFile);
         registerFabricTools({ registerTool: (d) => api.registerTool(d), logger: api.logger }, client, identity);
+        // Cross-plugin API: globalThis.__fabric
+        // Consumed by ClawPrompts' fabric-chat-injector to narrow its prompt
+        // injection to DM-typed channels only. The channel-meta cache is
+        // populated lazily from inbound (message.created carries xType) and
+        // persisted to ~/.openclaw/fabric-channel-meta.json — so even the
+        // very first DM after a fresh gateway start hits cache from the
+        // previous run rather than firing the injector on the wrong type.
+        //
+        // null return = channel never seen (cache cold). Callers MUST NOT
+        // fall back to "assume DM" — fail closed on unknown.
+        {
+            const _G = globalThis;
+            _G['__fabric'] = { getChannelType };
+            // Flush channel-meta cache when the gateway shuts down so
+            // recently-recorded xType entries don't get lost.
+            api.on('gateway_stop', () => {
+                try {
+                    flushChannelMeta();
+                }
+                catch { /* ignore */ }
+            });
+            api.logger.info('fabric: __fabric cross-plugin API installed (getChannelType)');
+        }
         api.on('gateway_start', () => {
             const _G = globalThis;
             if (_G._fabricInboundStarted)
@@ -56,11 +87,40 @@ export default defineChannelPluginEntry({
                 return;
             }
             inbound = new FabricInbound(runtimeRef, api.config, client, identity, api.logger, accounts);
-            void inbound.start();
+            // start() resolves once all accounts have attempted login; per-
+            // agent failures are logged but don't reject. Once it resolves we
+            // can harvest the presence accounts (those that DID log in have
+            // their fabricUserId + first guild endpoint populated).
+            void inbound.start().then(() => {
+                if (!inbound)
+                    return;
+                presence = new PresenceSync(api.logger, client);
+                presence.setAccounts(inbound.getPresenceAccounts());
+                presence.start();
+                api.logger.info(`fabric: presence-sync started for ${inbound.getPresenceAccounts().length} account(s)`);
+                // Re-harvest every 5 min: catches agents added via tool-based
+                // identity provisioning after gateway_start (recruitment flow).
+                // setAccounts is idempotent — duplicates collapse on agentId.
+                presenceRefreshTimer = setInterval(() => {
+                    if (inbound && presence)
+                        presence.setAccounts(inbound.getPresenceAccounts());
+                }, 5 * 60_000);
+            });
             api.logger.info(`fabric: inbound started for ${accounts.length} account(s)`);
             void syncFabricCommands(client, cfg, accounts, api.logger);
         });
+        // Note: the per-turn coalesce flush happens deterministically in
+        // inbound.ts right after dispatchInboundReplyWithBase resolves (that
+        // is the real "all deliveries done" boundary; the agent_end hook fires
+        // BEFORE deliver()). gateway_stop only flushes any leftover buffer.
         api.on('gateway_stop', () => {
+            void flushAllFabric();
+            if (presenceRefreshTimer) {
+                clearInterval(presenceRefreshTimer);
+                presenceRefreshTimer = null;
+            }
+            presence?.stop();
+            presence = null;
             inbound?.stop();
             inbound = null;
         });

dist/fabric/src/accounts.js

@@ -7,6 +7,17 @@ const DEFAULT_CENTER = 'http://localhost:7001/api';
 function section(cfg) {
     return cfg.channels?.fabric ?? {};
 }
+// The commands-sync shared secret (channel-level only). Empty string when
+// unconfigured — callers decide how to handle (slash-command sync is then
+// rejected by the guild).
+export function resolveCommandsSyncKey(cfg) {
+    return (section(cfg).commandsSyncKey ?? '').trim();
+}
+// Whether to coalesce a split agent turn into one Fabric message
+// (channel-level). Default true.
+export function resolveCoalesce(cfg) {
+    return (cfg.channels?.fabric ?? {}).coalesce !== false;
+}
 export function listFabricAccountIds(cfg) {
     const accts = section(cfg).accounts ?? {};
     const ids = Object.keys(accts);

dist/fabric/src/channel-meta.js

@@ -0,0 +1,105 @@
+/**
+ * Channel-meta cache. Records (channelId → xType) for every fabric
+ * channel the gateway has seen at least one inbound message in.
+ *
+ * Populated lazily from inbound (`recordChannelType` is called for
+ * every `message.created` event with non-empty `xType`). Persisted to
+ * `~/.openclaw/fabric-channel-meta.json` so the cache survives
+ * gateway restarts (so the very first DM after restart still gets the
+ * right xType without waiting for a fresh inbound).
+ *
+ * Exposed cross-plugin via `globalThis.__fabric.getChannelType`. Used
+ * by ClawPrompts' fabric-chat-injector to narrow its prompt injection
+ * to xType==='dm' only.
+ *
+ * Failure mode: lookup misses (channel never seen / inbound dropped
+ * xType) return null. Callers MUST treat null as "unknown" — DO NOT
+ * fall back to "assume DM" or you re-introduce the false-positive on
+ * group channels.
+ */
+import { existsSync, mkdirSync, readFileSync, renameSync, writeFileSync } from 'node:fs';
+import { dirname, join } from 'node:path';
+import { homedir } from 'node:os';
+const CACHE_FILE = join(homedir(), '.openclaw', 'fabric-channel-meta.json');
+let memory = new Map();
+let loaded = false;
+let dirty = false;
+let flushTimer = null;
+function load() {
+    if (loaded)
+        return;
+    loaded = true;
+    try {
+        if (!existsSync(CACHE_FILE))
+            return;
+        const raw = readFileSync(CACHE_FILE, 'utf8');
+        const parsed = JSON.parse(raw);
+        for (const [k, v] of Object.entries(parsed.channels ?? {})) {
+            if (typeof k === 'string' && typeof v === 'string')
+                memory.set(k, v);
+        }
+    }
+    catch {
+        // ignore — start with empty cache on corruption
+    }
+}
+function scheduleFlush() {
+    if (flushTimer)
+        return;
+    // Debounce writes — many inbound messages may arrive in a burst.
+    // 250ms coalesces them; on gateway_stop the channel plugin can force
+    // a synchronous flush via flushChannelMeta().
+    flushTimer = setTimeout(() => {
+        flushTimer = null;
+        if (!dirty)
+            return;
+        dirty = false;
+        flushSync();
+    }, 250);
+}
+function flushSync() {
+    try {
+        const dir = dirname(CACHE_FILE);
+        if (!existsSync(dir))
+            mkdirSync(dir, { recursive: true });
+        const out = { channels: Object.fromEntries(memory) };
+        const tmp = CACHE_FILE + '.tmp';
+        writeFileSync(tmp, JSON.stringify(out, null, 2) + '\n', 'utf8');
+        renameSync(tmp, CACHE_FILE);
+    }
+    catch {
+        // swallow — cache is an optimization; loss-on-write is recoverable
+    }
+}
+/** Called by inbound on every message.created. xType empty → no-op. */
+export function recordChannelType(channelId, xType) {
+    if (!channelId || !xType)
+        return;
+    load();
+    const existing = memory.get(channelId);
+    if (existing === xType)
+        return;
+    memory.set(channelId, xType);
+    dirty = true;
+    scheduleFlush();
+}
+/** Cross-plugin lookup. null when channel never seen / unknown. */
+export function getChannelType(channelId) {
+    if (!channelId)
+        return null;
+    load();
+    return memory.get(channelId) ?? null;
+}
+/** Force-flush — called on plugin shutdown to make sure recently
+ *  recorded entries hit disk before the gateway dies. */
+export function flushChannelMeta() {
+    if (flushTimer) {
+        clearTimeout(flushTimer);
+        flushTimer = null;
+    }
+    if (dirty) {
+        dirty = false;
+        flushSync();
+    }
+}
+export const CHANNEL_META_PATH = CACHE_FILE;

dist/fabric/src/channel.js

@@ -11,6 +11,15 @@
 import { createChatChannelPlugin, createChannelPluginBase, buildChannelOutboundSessionRoute, } from 'openclaw/plugin-sdk/core';
 import { FabricClient } from './fabric-client.js';
 import { listFabricAccountIds, resolveFabricAccount, resolveDefaultFabricAccountId, } from './accounts.js';
+import { getChannelType } from './channel-meta.js';
+export function fabricPeerRoutingForXType(xType) {
+    if (xType === 'dm')
+        return { peerKind: 'direct', chatType: 'direct' };
+    return { peerKind: 'group', chatType: 'group' };
+}
+export function fabricPeerRoutingForChannel(channelId) {
+    return fabricPeerRoutingForXType(getChannelType(channelId));
+}
 // ---- target grammar: fabric:<channelId> ----
 export function stripFabricTargetPrefix(raw) {
     let s = (raw ?? '').trim();
@@ -38,13 +47,18 @@ export function resolveFabricOutboundSessionRoute(params) {
     const id = stripFabricTargetPrefix(params.target);
     if (!id)
         return null;
+    // Consult the channel-meta cache populated by inbound — DM channels
+    // need peer.kind='direct' so the outbound session key matches the
+    // inbound one. Cache miss falls back to 'group' (the pre-fix default,
+    // no regression on cold cache).
+    const { peerKind, chatType } = fabricPeerRoutingForChannel(id);
     return buildChannelOutboundSessionRoute({
         cfg: params.cfg,
         agentId: params.agentId,
         channel: 'fabric',
         accountId: params.accountId,
-        peer: { kind: 'group', id },
-        chatType: 'group',
+        peer: { kind: peerKind, id },
+        chatType,
         from: `fabric:channel:${id}`,
         to: `fabric:${id}`,
     });
@@ -103,6 +117,19 @@ export const fabricChannelPlugin = createChatChannelPlugin({
                 resolveAccount: (cfg, accountId) => resolveFabricAccount(cfg, accountId),
                 defaultAccountId: (cfg) => resolveDefaultFabricAccountId(cfg),
                 isConfigured: (account) => Boolean(account.fabricApiKey),
+                // openclaw's channelManager.getRuntimeSnapshot() — called every minute
+                // by the channel-health-monitor — defaults `configured: true` when the
+                // plugin doesn't expose describeAccount (see applyDescribedAccountFields
+                // in server-channels). Without this, fabric's synthetic 'default'
+                // account (returned by listFabricAccountIds when channels.fabric.accounts
+                // is empty — the prod shape) gets snapshot {enabled:true, configured:true,
+                // running:false} → isManagedAccount=true → not-running → restart loop
+                // every ~10 min, logging `[fabric:default] health-monitor: restarting`.
+                // Mirror isConfigured here so the snapshot truthfully reports false for
+                // any account without a fabricApiKey.
+                describeAccount: (account) => ({
+                    configured: Boolean(account.fabricApiKey),
+                }),
             },
             // Minimal setup adapter: Fabric is configured directly under
             // channels.fabric.* (no interactive wizard). applyAccountConfig is the
@@ -137,7 +164,10 @@ export const fabricChannelPlugin = createChatChannelPlugin({
         attachedResults: {
             channel: 'fabric',
             sendText: async (ctx) => {
-                // openclaw passes config under cfg or config depending on path
+                // openclaw passes config under cfg or config depending on path.
+                // Note: inbound agent replies go through inbound.ts `deliver`
+                // (where turn coalescing happens). This path is for any direct
+                // outbound sends and posts immediately.
                 const cfg = (ctx.cfg ?? ctx.config ?? {});
                 try {
                     const r = await sendToFabric(cfg, ctx.accountId ?? null, ctx.to, ctx.text);

dist/fabric/src/coalesce.js

@@ -0,0 +1,75 @@
+// Deterministic turn coalescer.
+//
+// OpenClaw calls the Fabric `deliver` callback once per assistant text
+// segment; a thinking/tool block between two text blocks is a delivery
+// boundary, so one agent turn of `text → thinking/tool → text` arrives as
+// multiple deliver() calls. There is no turn id on the delivery, so we
+// BUFFER segments by Fabric channelId and post the merged message when the
+// turn truly ends. The flush is driven by inbound.ts right after
+// `dispatchInboundReplyWithBase` resolves — that only happens AFTER every
+// deliver() of the turn, a deterministic boundary (NOT a timer, NOT the
+// agent_end hook, which fires before deliver()). `coalesce=false` posts
+// each segment immediately.
+const SAFETY_FLUSH_MS = 120_000; // leak-guard only; not the flush mechanism
+export function normChannelId(x) {
+    const s = String(x ?? '');
+    return s.startsWith('fabric:') ? s.slice('fabric:'.length) : s;
+}
+const pendingByChannel = new Map();
+async function flushChannel(channelId, reason) {
+    const p = pendingByChannel.get(channelId);
+    if (!p)
+        return;
+    pendingByChannel.delete(channelId);
+    clearTimeout(p.safety);
+    const text = p.parts.join('\n\n').trim();
+    if (!text)
+        return;
+    try {
+        await p.post(text);
+        p.log?.(`fabric: flushed ${p.parts.length} segment(s) channel=${channelId} (${reason})`);
+    }
+    catch (e) {
+        p.log?.(`fabric: flush FAILED channel=${channelId} (${reason}): ${String(e)}`);
+    }
+}
+// Buffer one delivered segment (or send immediately when coalesce=false).
+// `post` performs the real Fabric postMessage with the caller's already
+// resolved guild/token; on flush it is called once with the merged text.
+export async function enqueueDelivery(params) {
+    const cid = normChannelId(params.channelId);
+    const text = (params.text ?? '').trim();
+    if (!text)
+        return;
+    if (!params.coalesce) {
+        await params.post(text);
+        return;
+    }
+    const existing = pendingByChannel.get(cid);
+    if (existing) {
+        existing.parts.push(text);
+        existing.post = params.post; // freshest guild/token closure
+        existing.log = params.log;
+    }
+    else {
+        pendingByChannel.set(cid, {
+            parts: [text],
+            post: params.post,
+            log: params.log,
+            safety: setTimeout(() => void flushChannel(cid, 'safety-timeout'), SAFETY_FLUSH_MS),
+        });
+    }
+}
+// Called by the agent_end hook with the hook ctx's channelId (bare or
+// fabric:-prefixed). Deterministic per-turn boundary.
+export async function flushFabricForChannel(rawChannelId) {
+    const cid = normChannelId(rawChannelId);
+    if (cid)
+        await flushChannel(cid, 'dispatch-end');
+}
+// gateway_stop: flush anything still buffered.
+export async function flushAllFabric() {
+    for (const cid of [...pendingByChannel.keys()]) {
+        await flushChannel(cid, 'gateway_stop');
+    }
+}

dist/fabric/src/command-sync.js

@@ -6,6 +6,7 @@
 // dynamic arg `choices` to a static snapshot here (like Discord does at
 // registration time).
 import { listNativeCommandSpecsForConfig, findCommandByNativeName, resolveCommandArgChoices, } from 'openclaw/plugin-sdk/native-command-registry';
+import { resolveCommandsSyncKey } from './accounts.js';
 function normChoice(c) {
     if (typeof c === 'string')
         return { value: c, label: c };
@@ -62,6 +63,14 @@ export function buildFabricCommandSpecs(cfg) {
 // Push the catalog to every guild the known agents belong to (idempotent;
 // the catalog is OpenClaw-global, so one PUT per guild is enough).
 export async function syncFabricCommands(client, cfg, accounts, log) {
+    // Guild C-2: the sync key comes from the channel config only (schema
+    // marks it required). Without it the guild rejects the catalog write.
+    const syncKey = resolveCommandsSyncKey(cfg);
+    if (!syncKey) {
+        log.warn('fabric: channels.fabric.commandsSyncKey is not set — skipping ' +
+            'slash-command sync (set it to the guild FABRIC_BACKEND_GUILD_COMMANDS_SYNC_KEY)');
+        return;
+    }
     let specs;
     try {
         specs = buildFabricCommandSpecs(cfg);
@@ -88,7 +97,7 @@ export async function syncFabricCommands(client, cfg, accounts, log) {
             if (!tok)
                 continue;
             try {
-                await client.syncCommands(g.endpoint, tok, specs);
+                await client.syncCommands(g.endpoint, tok, specs, syncKey);
                 done.add(g.nodeId);
                 log.info(`fabric: synced ${specs.length} slash command(s) -> ${g.nodeId}`);
             }

dist/fabric/src/fabric-client.js

@@ -23,12 +23,13 @@ export class FabricClient {
     }
     // Generic JSON request (GET/PUT/PATCH/DELETE). Empty 2xx body -> null
     // (Fabric returns an empty body when a channel has no canvas).
-    async req(method, url, auth, body) {
+    async req(method, url, auth, body, extraHeaders) {
         const res = await fetch(url, {
             method,
             headers: {
                 ...(body !== undefined ? { 'content-type': 'application/json' } : {}),
                 ...(auth ? { authorization: `Bearer ${auth}` } : {}),
+                ...(extraHeaders ?? {}),
             },
             body: body !== undefined ? JSON.stringify(body) : undefined,
         });
@@ -62,6 +63,11 @@ export class FabricClient {
     createChannel(guildEndpoint, guildToken, body) {
         return this.post(`${guildEndpoint}/api/channels`, body, guildToken);
     }
+    // PATCH /api/channels/:id — backend currently only patches `purpose`.
+    // Caller must be a member of the channel (or any user if public).
+    setChannelPurpose(guildEndpoint, guildToken, channelId, purpose) {
+        return this.req('PATCH', `${guildEndpoint}/api/channels/${channelId}`, guildToken, { purpose });
+    }
     closeChannel(guildEndpoint, guildToken, channelId) {
         return this.post(`${guildEndpoint}/api/channels/${channelId}/close`, {}, guildToken);
     }
@@ -74,8 +80,11 @@ export class FabricClient {
     // Register the OpenClaw slash-command catalog with this guild (idempotent
     // full replace). The frontend GETs it for `/` autocomplete; execution
     // still flows as a normal /<cmd> message into OpenClaw's command system.
-    syncCommands(guildEndpoint, guildToken, commands) {
-        return this.req('PUT', `${guildEndpoint}/api/commands`, guildToken, { commands });
+    syncCommands(guildEndpoint, guildToken, commands, syncKey) {
+        // Guild C-2: the shared key is sourced from the channel config
+        // (channels.fabric.commandsSyncKey) and must equal the guild's
+        // FABRIC_BACKEND_GUILD_COMMANDS_SYNC_KEY for the catalog write.
+        return this.req('PUT', `${guildEndpoint}/api/commands`, guildToken, { commands }, syncKey ? { 'x-commands-sync-key': syncKey } : undefined);
     }
     // [{ userId, bypass }] — bypass is true only for discuss/work bypass-list
     channelMembers(guildEndpoint, guildToken, channelId) {
@@ -101,4 +110,31 @@ export class FabricClient {
     removeCanvas(endpoint, token, channelId) {
         return this.req('DELETE', this.canvasUrl(endpoint, channelId), token);
     }
+    // ---- channel discovery + message read (used by the agent-facing
+    // fabric-channel-list / fabric-message-history tools) ----
+    /**
+     * List channels in a guild visible to the calling user. Backend
+     * filters to public + channels the user is a member of.
+     */
+    listChannels(guildEndpoint, guildToken, guildNodeId) {
+        return this.req('GET', `${guildEndpoint}/api/channels?guildId=${encodeURIComponent(guildNodeId)}`, guildToken);
+    }
+    /**
+     * Page through a channel's message history by `seq`.
+     *
+     * Backend defaults: 50 / call, max 200. The `seq` field starts at 1
+     * per channel; pass `seqFrom=channel.lastSeq - N + 1` to get the
+     * tail. Page metadata in the response describes what to ask next.
+     */
+    listMessages(guildEndpoint, guildToken, channelId, opts = {}) {
+        const qs = new URLSearchParams();
+        if (opts.seqFrom !== undefined)
+            qs.set('seq_from', String(opts.seqFrom));
+        if (opts.seqTo !== undefined)
+            qs.set('seq_to', String(opts.seqTo));
+        if (opts.limit !== undefined)
+            qs.set('limit', String(opts.limit));
+        const url = `${guildEndpoint}/api/channels/${channelId}/messages` + (qs.toString() ? `?${qs}` : '');
+        return this.req('GET', url, guildToken);
+    }
 }

dist/fabric/src/inbound.js

@@ -3,6 +3,10 @@ import { tmpdir } from 'node:os';
 import { join } from 'node:path';
 import { io } from 'socket.io-client';
 import { dispatchInboundReplyWithBase } from 'openclaw/plugin-sdk/inbound-reply-dispatch';
+import { resolveCoalesce } from './accounts.js';
+import { fabricPeerRoutingForXType } from './channel.js';
+import { recordChannelType } from './channel-meta.js';
+import { enqueueDelivery, flushFabricForChannel } from './coalesce.js';
 export class FabricInbound {
     core;
     cfg;
@@ -12,12 +16,139 @@ export class FabricInbound {
     accounts;
     sockets = [];
     seen = new Set();
+    // Timers that periodically re-sync channel membership per (agent, guild).
+    // Without this, the agent's socket.io subscriptions are a snapshot taken
+    // at connect time — any channel the agent joins later (e.g. a fresh DM
+    // created by another user) is unreachable until the gateway restarts.
+    channelSyncTimers = [];
+    // Resync cadence. Backend doesn't push a `channel.joined` event, so we
+    // poll. 60s keeps the lag bounded without hammering the backend.
+    static CHANNEL_SYNC_INTERVAL_MS = 60_000;
     // Guild access tokens are short-lived (~15 min). The socket survives via
     // socket.io reconnect, but the token captured at connect time goes stale,
     // so HTTP calls (attachment download, posting the reply) start 401ing.
     // Re-login per agent on a short TTL to keep a fresh token.
     tokenCache = new Map();
     static TOKEN_TTL_MS = 8 * 60 * 1000;
+    // Per-channel serial work queue. Every inbound socket message for a
+    // channel awaits the previous task for that same channel, so model
+    // turns never interleave. Map key = channelId; value is the tail of
+    // the chain (an in-flight promise the next task awaits).
+    //
+    // Why per-channel and not per-agent: a single agent may sit in
+    // several triage / general channels; we want each channel to flow at
+    // its own speed but the SAME channel's traffic to be strictly serial.
+    // For dm and discuss the queue also serialises but those traditionally
+    // had at-most-one-in-flight anyway via the turn engine.
+    channelChains = new Map();
+    // Agent.status snapshot cache (5s TTL) — keeps the HF /calendar/
+    // agent/status round-trip off the hot path for back-to-back triage
+    // messages. Short TTL because status flips are rare-but-meaningful.
+    agentStatusCache = new Map();
+    static AGENT_STATUS_TTL_MS = 5_000;
+    // Triage messages that arrived while the on-duty agent wasn't on_call
+    // — sit here until either (a) the agent becomes on_call and the next
+    // triage arrival drains them, or (b) the gateway restarts (lost; ok
+    // because the underlying Fabric messages are persisted and re-fetched
+    // on agent reconnect's history sweep).
+    pendingTriageGated = [];
+    // Schedule `task` to run after every previous task on the same
+    // channel has completed. Returns the promise so callers can await
+    // their own result if they need to; the chain itself is fire-and-
+    // forget from the socket.on handler.
+    enqueueChannelTask(channelId, task) {
+        const prev = this.channelChains.get(channelId) ?? Promise.resolve();
+        const next = prev.then(task).catch((err) => {
+            this.log.warn(`fabric: per-channel task failed channel=${channelId}: ${String(err)}`);
+        });
+        this.channelChains.set(channelId, next);
+        // Best-effort cleanup so the Map doesn't grow without bound for
+        // long-running gateways: drop the entry when the chain settles, but
+        // only if it's still the latest reference (newer enqueue may have
+        // overwritten it in the meantime).
+        void next.finally(() => {
+            if (this.channelChains.get(channelId) === next) {
+                this.channelChains.delete(channelId);
+            }
+        });
+        return next;
+    }
+    // Hit HF backend to check whether `agentId` is currently on_call.
+    // Cached for 5s. Failures (network, 404, etc.) are treated as "not
+    // on_call" — triage stays gated rather than risking a confused wake.
+    async checkAgentOnCall(agentId) {
+        const cached = this.agentStatusCache.get(agentId);
+        if (cached && Date.now() - cached.at < FabricInbound.AGENT_STATUS_TTL_MS) {
+            return cached.onCall;
+        }
+        const base = (process.env.HF_API_BASE_URL ?? '').trim() || 'https://monitor.hangman-lab.top';
+        // CLAW_IDENTIFIER resolution priority:
+        //   1. HF_CLAW_IDENTIFIER env (operator override)
+        //   2. openclaw config `plugins.harbor-forge.identifier` (what the HF
+        //      plugin itself uses — keeps the two in sync without an extra
+        //      env per service unit)
+        //   3. os.hostname() last-resort fallback (often wrong: e.g. sim
+        //      container hostname is `server.t2` but HF agent row has
+        //      `claw_identifier=sim-t2`; matching is mandatory for the HF
+        //      backend's _require_agent() check)
+        let claw = (process.env.HF_CLAW_IDENTIFIER ?? '').trim();
+        if (!claw) {
+            try {
+                // openclaw config shape (verified in sim):
+                //   { plugins: { entries: { 'harbor-forge': { config: { identifier } } } } }
+                const cfg = this.cfg;
+                const fromCfg = cfg?.plugins?.entries?.['harbor-forge']?.config?.identifier;
+                if (fromCfg && typeof fromCfg === 'string' && fromCfg.trim()) {
+                    claw = fromCfg.trim();
+                }
+            }
+            catch {
+                /* fall through to hostname */
+            }
+        }
+        if (!claw) {
+            claw = (await import('os')).hostname();
+        }
+        let onCall = false;
+        try {
+            const url = `${base.replace(/\/$/, '')}/calendar/agent/status?agent_id=${encodeURIComponent(agentId)}`;
+            const res = await fetch(url, {
+                headers: { 'X-Agent-ID': agentId, 'X-Claw-Identifier': claw },
+            });
+            if (res.ok) {
+                const data = (await res.json());
+                onCall = (data.status ?? '').toLowerCase() === 'on_call';
+            }
+        }
+        catch (err) {
+            this.log.warn(`fabric: HF status check failed agent=${agentId}: ${String(err)}`);
+        }
+        this.agentStatusCache.set(agentId, { onCall, at: Date.now() });
+        return onCall;
+    }
+    // FIFO drain of all triage-gated messages for `agentId` (called when
+    // we just learned they're on_call). Each drained message is dispatched
+    // through its own channel chain so per-channel serial order is kept.
+    async drainGatedFor(agentId) {
+        const keep = [];
+        const drain = [];
+        for (const item of this.pendingTriageGated) {
+            if (item.agentId === agentId)
+                drain.push(item);
+            else
+                keep.push(item);
+        }
+        if (drain.length === 0)
+            return;
+        this.pendingTriageGated = keep;
+        for (const item of drain) {
+            this.log.info(`fabric: triage drain agent=${item.agentId} channel=${item.channelId} msg=${item.m.messageId}`);
+            // Re-enqueue via the per-channel chain so ordering is preserved.
+            this.enqueueChannelTask(item.channelId, async () => {
+                await this.dispatch(item.agentId, item.g, item.channelId, item.m, item.session);
+            });
+        }
+    }
     // Return a fresh guild access token for the agent, re-authenticating with
     // the agent's Fabric API key when the cached session is stale. Falls back
     // to the connect-time session token if re-login fails.
@@ -73,38 +204,173 @@ export class FabricInbound {
         }
     }
     stop() {
+        for (const t of this.channelSyncTimers)
+            clearInterval(t);
+        this.channelSyncTimers = [];
         for (const s of this.sockets)
             s.disconnect();
         this.sockets = [];
     }
+    /**
+     * Per-account metadata harvested during `start()` — used by
+     * PresenceSync to know where to push each agent's HF status.
+     *
+     * `fabricUserId` is filled from `session.user.id` after agent-login.
+     * `guildBaseUrl` is the FIRST guild the agent is connected to (multi-
+     * guild presence push is a future concern; for sim/prod-v1 each agent
+     * is in one guild).
+     *
+     * Returns ONLY agents that successfully connected — failed-login
+     * agents have no fabricUserId yet and are excluded.
+     */
+    getPresenceAccounts() {
+        const out = [];
+        for (const entry of this.identity.list()) {
+            if (!entry.fabricUserId)
+                continue;
+            const presenceGuild = this.firstGuildByAgent.get(entry.agentId);
+            if (!presenceGuild)
+                continue;
+            out.push({
+                agentId: entry.agentId,
+                fabricUserId: entry.fabricUserId,
+                guildBaseUrl: presenceGuild.endpoint,
+                guildNodeId: presenceGuild.nodeId,
+                fabricApiKey: entry.fabricApiKey,
+            });
+        }
+        return out;
+    }
+    // Filled by connectAgent for each (agent, guild). Tracks ONLY the first
+    // guild per agent (used as the presence-push target). Stores both
+    // endpoint and nodeId — presence-sync needs both: endpoint to build
+    // the URL, nodeId to pick the matching guildAccessToken from a fresh
+    // agent-login response.
+    firstGuildByAgent = new Map();
     async connectAgent(agentId, session) {
         const selfUserId = session.user.id;
+        // First-guild capture for presence-sync push target. session.guilds is
+        // already in priority order from Center; we take the first one with a
+        // valid endpoint and stop. Multi-guild presence is a future concern.
+        if (!this.firstGuildByAgent.has(agentId)) {
+            const firstGuild = session.guilds.find((g) => typeof g.endpoint === 'string' && g.endpoint.length > 0);
+            if (firstGuild)
+                this.firstGuildByAgent.set(agentId, { endpoint: firstGuild.endpoint, nodeId: firstGuild.nodeId });
+        }
         for (const g of session.guilds) {
             const tok = session.guildAccessTokens.find((t) => t.guildNodeId === g.nodeId)?.token;
             if (!tok)
                 continue;
+            // Use the *callback* form of `auth` so socket.io re-evaluates the JWT
+            // on every (re)connect. The single-shot `auth: { token: tok }` shape
+            // captured the token in closure: after socket.io's silent auto-reconnect
+            // the backend got the same JWT that expired ~15 min into the session
+            // (guildAccessToken TTL = 900s) and silently rejected the handshake at
+            // the application layer. The client's `connect` event still fired (TCP
+            // succeeded), so the plugin happily ran the channel-resync, emitted
+            // `join_channel` into the void, and logged "joined N channel(s)" while
+            // the backend was actually broadcasting message.created to a room with
+            // zero subscribers. End user symptom: DMs to agents silently dropped.
             const socket = io(`${g.endpoint}/realtime`, {
                 transports: ['websocket'],
-                auth: { token: tok },
+                auth: (cb) => {
+                    // Best-effort fresh token; on transient failure fall back to the
+                    // last known good one. tokenCache also keeps HTTP calls (attachment
+                    // download / reply post) from 401'ing in the same window.
+                    this.freshGuildToken(agentId, g.nodeId, session)
+                        .then((fresh) => cb({ token: fresh ?? tok }))
+                        .catch(() => cb({ token: tok }));
+                },
                 autoConnect: false,
             });
-            const joinAll = async () => {
+            // Tracked socket.io rooms for this (agent, guild). The initial fetch
+            // on `connect` seeds it; the periodic resync diffs against it so we
+            // only emit `join_channel` for genuinely new channels (and
+            // `leave_channel` for ones the agent is no longer in).
+            const joined = new Set();
+            const syncChannels = async (kind) => {
+                let freshTok;
                 try {
-                    const res = await fetch(`${g.endpoint}/api/channels?guildId=${encodeURIComponent(g.nodeId)}`, { headers: { authorization: `Bearer ${tok}` } });
-                    const channels = res.ok ? (await res.json()) : [];
-                    for (const c of channels)
-                        socket.emit('join_channel', { channelId: c.id });
-                    this.log.info(`fabric: agent ${agentId} joined ${channels.length} channel(s) on ${g.nodeId}`);
+                    freshTok = await this.freshGuildToken(agentId, g.nodeId, session);
                 }
                 catch {
-                    /* best effort */
+                    freshTok = tok;
+                }
+                const authTok = freshTok ?? tok;
+                try {
+                    const res = await fetch(`${g.endpoint}/api/channels?guildId=${encodeURIComponent(g.nodeId)}`, { headers: { authorization: `Bearer ${authTok}` } });
+                    if (!res.ok)
+                        return;
+                    const channels = (await res.json());
+                    const current = new Set(channels.map((c) => c.id));
+                    let added = 0;
+                    let removed = 0;
+                    for (const id of current) {
+                        if (!joined.has(id)) {
+                            socket.emit('join_channel', { channelId: id });
+                            joined.add(id);
+                            added++;
+                        }
+                    }
+                    for (const id of [...joined]) {
+                        if (!current.has(id)) {
+                            socket.emit('leave_channel', { channelId: id });
+                            joined.delete(id);
+                            removed++;
+                        }
+                    }
+                    if (kind === 'initial') {
+                        this.log.info(`fabric: agent ${agentId} joined ${current.size} channel(s) on ${g.nodeId}`);
+                    }
+                    else if (added > 0 || removed > 0) {
+                        this.log.info(`fabric: agent ${agentId} channel resync on ${g.nodeId}: +${added} -${removed} (now ${joined.size})`);
+                    }
+                }
+                catch {
+                    /* best effort — next tick will retry */
                 }
             };
-            socket.on('connect', () => void joinAll());
+            socket.on('connect', () => {
+                // On every (re)connect the server forgets prior subscriptions, so
+                // reset our local view and seed from a fresh fetch.
+                joined.clear();
+                void syncChannels('initial');
+            });
+            // Push-based membership events from the backend (companion to
+            // Fabric.Backend.Guild's RealtimeGateway.emitToUser). When the
+            // server tells us this user was added to / removed from a
+            // channel, we sub/unsub the socket.io room immediately — no
+            // 60s wait for the polling resync. Polling remains as a safety
+            // net for missed events.
+            socket.on('channel.joined', (evt) => {
+                const id = evt?.channelId;
+                if (!id || joined.has(id))
+                    return;
+                socket.emit('join_channel', { channelId: id });
+                joined.add(id);
+                this.log.info(`fabric: agent ${agentId} channel.joined push on ${g.nodeId}: ${id} (now ${joined.size})`);
+            });
+            socket.on('channel.left', (evt) => {
+                const id = evt?.channelId;
+                if (!id || !joined.has(id))
+                    return;
+                socket.emit('leave_channel', { channelId: id });
+                joined.delete(id);
+                this.log.info(`fabric: agent ${agentId} channel.left push on ${g.nodeId}: ${id} (now ${joined.size})`);
+            });
+            const syncTimer = setInterval(() => void syncChannels('resync'), FabricInbound.CHANNEL_SYNC_INTERVAL_MS);
+            this.channelSyncTimers.push(syncTimer);
             socket.on('message.created', (m) => {
                 const channelId = m.channelId ?? '';
                 if (!channelId)
                     return;
+                // Record xType into the channel-meta cache before self-author
+                // / dedup gates — channel type doesn't depend on who sent the
+                // message, and recording it on observer-only triage messages
+                // is still useful (the next consumer asking
+                // __fabric.getChannelType wants the answer regardless of
+                // whether THIS message was delivered to an agent).
+                recordChannelType(channelId, m.xType);
                 if (m.authorUserId && m.authorUserId === selfUserId)
                     return;
                 const key = `${agentId}:${m.messageId}`;
@@ -113,7 +379,31 @@ export class FabricInbound {
                 this.seen.add(key);
                 if (this.seen.size > 5000)
                     this.seen.clear();
-                void this.dispatch(agentId, g, channelId, m, session);
+                // Per-channel serial queue. Prevents concurrent model turns for
+                // the same channel — important for triage where a second wake
+                // arriving mid-reply would interleave with the in-flight one.
+                this.enqueueChannelTask(channelId, async () => {
+                    // Triage on_call gate: if the on-duty agent isn't currently
+                    // on_call per HF, don't dispatch yet — just sit on the
+                    // per-channel queue. Subsequent triage messages will recheck;
+                    // when the agent becomes on_call, the next arrival drains.
+                    //
+                    // Also handles: triage + wake=true must verify status before
+                    // committing to a model turn. Non-triage and triage observer
+                    // (wake=false) skip the gate.
+                    if (m.xType === 'triage' && m.wakeup === true) {
+                        const onCall = await this.checkAgentOnCall(agentId);
+                        if (!onCall) {
+                            this.log.info(`fabric: triage wake gated (agent=${agentId} not on_call) — re-queue msg=${m.messageId}`);
+                            this.pendingTriageGated.push({ agentId, g, channelId, m, session });
+                            return;
+                        }
+                        // Drain any previously-gated messages (FIFO) before this one,
+                        // now that we know the agent is on_call.
+                        await this.drainGatedFor(agentId);
+                    }
+                    await this.dispatch(agentId, g, channelId, m, session);
+                });
             });
             socket.connect();
             this.sockets.push(socket);
@@ -165,11 +455,19 @@ export class FabricInbound {
         const core = this.core;
         const cfg = this.cfg;
         try {
+            // Route by xType. DM channels need peer.kind='direct' so openclaw
+            // treats them as 1:1 (sessionKey 'agent:<id>:fabric:direct:<chan>'
+            // and ctx.ChatType='direct') rather than as a multi-party group.
+            // Without this, the agent's user-prompt metadata says
+            // 'is_group_chat: true' on a DM and downstream prompt logic
+            // (commands-handlers `isDirectMessage` checks ChatType==='direct')
+            // misclassifies the turn.
+            const { peerKind, chatType } = fabricPeerRoutingForXType(m.xType);
             const route = core.channel.routing.resolveAgentRoute({
                 cfg: this.cfg,
                 channel: 'fabric',
                 accountId: agentId,
-                peer: { kind: 'group', id: channelId },
+                peer: { kind: peerKind, id: channelId },
             });
             const storePath = core.channel.session.resolveStorePath(cfg.session?.store, {
                 agentId: route.agentId,
@@ -183,7 +481,7 @@ export class FabricInbound {
                 To: `fabric:${channelId}`,
                 SessionKey: route.sessionKey,
                 AccountId: route.accountId ?? agentId,
-                ChatType: 'group',
+                ChatType: chatType,
                 ConversationLabel: `fabric:${guild.nodeId}`,
                 SenderId: m.authorUserId ?? 'fabric',
                 Provider: 'fabric',
@@ -199,7 +497,22 @@ export class FabricInbound {
             // the woken speaker emits a normal message or /no-reply). We still
             // record the message into the agent's session so it has the full
             // channel conversation as context whenever it IS later woken.
-            if (m.wakeup !== true) {
+            //
+            // Exception: dm channels are 1:1 — there is no turn/wakeup gating;
+            // any message that isn't the agent's own (already filtered above) is
+            // always delivered to the model.
+            if (m.xType !== 'dm' && m.wakeup !== true) {
+                // Triage exception: non-wake messages (admin observer) MUST NOT
+                // enter the agent's session at all. The next time the agent
+                // wakes for a triage message, their context should contain only
+                // their own past wakeups + their own outgoing messages — never
+                // the observer-only chatter from other agents. For non-triage
+                // channels keep the legacy "record-as-history" so a later wake
+                // sees the full channel conversation.
+                if (m.xType === 'triage') {
+                    this.log.info(`fabric: triage observer skip agent=${agentId} channel=${channelId} msg=${m.messageId}`);
+                    return;
+                }
                 const ctxPayload = core.channel.reply.finalizeInboundContext(baseCtx);
                 await core.channel.session.recordInboundSession({
                     storePath,
@@ -244,8 +557,16 @@ export class FabricInbound {
                     this.log.info(`fabric: deliver agent=${agentId} channel=${channelId} len=${text.length}`);
                     if (!text || !gt)
                         return;
-                    await this.client.postMessage(guild.endpoint, gt, channelId, text, session.user.id);
-                    this.log.info(`fabric: posted reply agent=${agentId} channel=${channelId}`);
+                    // Buffer segments; the merged message is posted right after
+                    // dispatch returns (the deterministic turn boundary, see the
+                    // finally below). Disable per channel: channels.fabric.coalesce.
+                    await enqueueDelivery({
+                        channelId,
+                        text,
+                        coalesce: resolveCoalesce(this.cfg),
+                        post: (t) => this.client.postMessage(guild.endpoint, gt, channelId, t, session.user.id),
+                        log: (m) => this.log.info(m),
+                    });
                 },
                 onRecordError: (err) => this.log.warn(`fabric: session record failed agent=${agentId}: ${String(err)}`),
                 onDispatchError: (err, info) => this.log.warn(`fabric: ${info.kind} dispatch failed agent=${agentId}: ${String(err)}`),
@@ -270,5 +591,12 @@ export class FabricInbound {
         catch (err) {
             this.log.warn(`fabric: dispatch failed agent=${agentId} channel=${channelId}: ${String(err)}`);
         }
+        finally {
+            // Deterministic per-turn boundary: dispatchInboundReplyWithBase only
+            // resolves AFTER every deliver() call of this turn has run, so the
+            // buffer now holds all segments — flush them as ONE Fabric message.
+            // No hooks, no timers, no idle guessing.
+            await flushFabricForChannel(channelId);
+        }
     }
 }

dist/fabric/src/presence-sync.js

@@ -0,0 +1,139 @@
+// Guild access JWTs expire every 900s. Refresh ~2 min early to stay
+// safely inside the window even if a tick runs late.
+const TOKEN_TTL_MS = (15 - 2) * 60 * 1000;
+export class PresenceSync {
+    logger;
+    client;
+    timer = null;
+    lastStatus = new Map(); // by agentId
+    accounts = new Map();
+    tokenCache = new Map(); // by agentId
+    // Mutex flag: a tick iterates accounts serially with `await` on each
+    // agent-login + PUT round-trip, so a single tick can easily run 20+s
+    // when there are many accounts. setInterval(intervalMs) does NOT wait
+    // for the previous tick to finish — without this guard the next tick
+    // fires on top of a still-running one and two parallel iterations
+    // PUT the same agentId within milliseconds. That tipped the backend's
+    // first-time-insert race (separate fix in Fabric.Backend.Guild) into
+    // 500s on prod. Guarded ticks just skip a beat instead.
+    inflight = false;
+    constructor(logger, client) {
+        this.logger = logger;
+        this.client = client;
+    }
+    setAccounts(accounts) {
+        this.accounts.clear();
+        for (const a of accounts)
+            this.accounts.set(a.agentId, a);
+    }
+    start(intervalMs = 30_000) {
+        if (this.timer)
+            return;
+        this.timer = setInterval(() => {
+            this.tick().catch((err) => this.logger.warn(`fabric: presence-sync error: ${String(err)}`));
+        }, intervalMs);
+        // run once immediately so initial state lands fast
+        void this.tick();
+    }
+    stop() {
+        if (this.timer) {
+            clearInterval(this.timer);
+            this.timer = null;
+        }
+    }
+    /**
+     * Fetch a fresh guildAccessToken for `acct`, caching it under the
+     * agentId until just before its JWT expiry. Returns null on login
+     * failure or if the session has no matching guild — caller logs +
+     * skips the PUT.
+     */
+    async ensureGuildToken(acct) {
+        const now = Date.now();
+        const cached = this.tokenCache.get(acct.agentId);
+        if (cached && cached.expiresAt > now)
+            return cached.token;
+        let session;
+        try {
+            session = await this.client.agentLogin(acct.fabricApiKey);
+        }
+        catch (err) {
+            this.logger.warn(`fabric: presence-sync agent-login failed for ${acct.agentId}: ${String(err)}`);
+            return null;
+        }
+        const entry = session.guildAccessTokens.find((g) => g.guildNodeId === acct.guildNodeId);
+        if (!entry?.token) {
+            this.logger.warn(`fabric: presence-sync no guild token for ${acct.agentId} guild=${acct.guildNodeId}`);
+            return null;
+        }
+        this.tokenCache.set(acct.agentId, { token: entry.token, expiresAt: now + TOKEN_TTL_MS });
+        return entry.token;
+    }
+    async tick() {
+        // Mutex: see the `inflight` field declaration for the why. Drop
+        // overlapping ticks rather than letting them run concurrently —
+        // status is gated by `lastStatus !== bridge.get`, so skipping a
+        // beat costs nothing the next beat won't catch.
+        if (this.inflight)
+            return;
+        this.inflight = true;
+        try {
+            await this.tickInner();
+        }
+        finally {
+            this.inflight = false;
+        }
+    }
+    async tickInner() {
+        const bridge = globalThis['__hfAgentStatus'];
+        if (!bridge || typeof bridge.get !== 'function')
+            return; // HF plugin not loaded — skip
+        for (const [agentId, acct] of this.accounts) {
+            let status;
+            try {
+                status = await bridge.get(agentId);
+            }
+            catch {
+                continue;
+            }
+            if (!status)
+                continue;
+            if (this.lastStatus.get(agentId) === status)
+                continue; // no change → no PUT
+            const guildToken = await this.ensureGuildToken(acct);
+            if (!guildToken)
+                continue;
+            try {
+                // Endpoint: PUT /api/agents/:userId/presence. ApiKeyGuard (global
+                // APP_GUARD) requires `Authorization: Bearer <guildAccessToken>`
+                // — NOT the agent's raw fabricApiKey. Pre-v1: this loop sent
+                // x-api-key and got 401 "missing bearer token" forever. The /api
+                // prefix is required because the guild backend sets a global
+                // 'api' prefix in main.ts setGlobalPrefix('api').
+                const url = `${acct.guildBaseUrl.replace(/\/$/, '')}/api/agents/${encodeURIComponent(acct.fabricUserId)}/presence`;
+                const res = await fetch(url, {
+                    method: 'PUT',
+                    headers: {
+                        'content-type': 'application/json',
+                        authorization: `Bearer ${guildToken}`,
+                    },
+                    body: JSON.stringify({ status, source: 'hf-plugin' }),
+                });
+                if (res.ok) {
+                    this.lastStatus.set(agentId, status);
+                    this.logger.info(`fabric: presence-sync ${agentId} → ${status}`);
+                }
+                else {
+                    // 401 here usually means the cached token went stale unexpectedly
+                    // (server-side rotation or clock skew) — drop the cache so the
+                    // next tick re-logs-in.
+                    if (res.status === 401)
+                        this.tokenCache.delete(agentId);
+                    this.logger.warn(`fabric: presence-sync PUT ${agentId} failed: ${res.status}`);
+                }
+            }
+            catch (err) {
+                this.logger.warn(`fabric: presence-sync PUT ${agentId} threw: ${String(err)}`);
+            }
+        }
+    }
+}

dist/fabric/src/tools.js

@@ -25,7 +25,9 @@ export function registerFabricTools(api, client, identity) {
     // or via static config (channels.fabric.accounts.<agentId>).
     const makeCreate = (kind) => api.registerTool((ctx) => ({
         name: `create-${kind}-channel`,
-        description: `Create a Fabric ${kind} channel (x_type=${X_BY_KIND[kind]}).`,
+        description: `Create a Fabric ${kind} channel (x_type=${X_BY_KIND[kind]}). ` +
+            'Optionally pass `purpose` to describe what this channel is for — ' +
+            'agents browse channels by purpose via fabric-channel-list.',
         parameters: {
             type: 'object',
             additionalProperties: false,
@@ -37,9 +39,16 @@ export function registerFabricTools(api, client, identity) {
                 memberUserIds: { type: 'array', items: { type: 'string' } },
                 onDuty: { type: 'string', description: 'required for triage-like flows (unused for these kinds)' },
                 listeners: { type: 'array', items: { type: 'string' } },
+                purpose: {
+                    type: 'string',
+                    description: "Free-form description of what this channel is for. Optional but " +
+                        'strongly recommended so other agents can find this channel by ' +
+                        'intent (via fabric-channel-list). Can be edited later with ' +
+                        'fabric-channel-set-purpose.',
                 },
             },
-        execute: async (p) => {
+        },
+        execute: async (_id, p) => {
             const agentId = ctx.agentId;
             if (!agentId)
                 return { ok: false, error: 'no agent context' };
@@ -50,6 +59,7 @@ export function registerFabricTools(api, client, identity) {
                 xType: X_BY_KIND[kind],
                 isPublic: p.isPublic ?? false,
                 memberUserIds: p.memberUserIds ?? [],
+                ...(p.purpose !== undefined ? { purpose: p.purpose } : {}),
             });
             return { ok: true, channelId: ch.id };
         },
@@ -74,7 +84,7 @@ export function registerFabricTools(api, client, identity) {
                 callbackChannelId: { type: 'string', description: 'optional channel to also post the summary to' },
             },
         },
-        execute: async (p) => {
+        execute: async (_id, p) => {
             const agentId = ctx.agentId;
             if (!agentId)
                 return { ok: false, error: 'no agent context' };
@@ -118,7 +128,7 @@ export function registerFabricTools(api, client, identity) {
                 },
             },
         },
-        execute: async (p) => {
+        execute: async (_id, p) => {
             const agentId = ctx.agentId;
             if (!agentId)
                 return { ok: false, error: 'no agent context' };
@@ -178,7 +188,7 @@ export function registerFabricTools(api, client, identity) {
                 channelId: { type: 'string' },
             },
         },
-        execute: async (p) => {
+        execute: async (_id, p) => {
             const agentId = ctx.agentId;
             if (!agentId)
                 return { ok: false, error: 'no agent context' };
@@ -202,4 +212,252 @@ export function registerFabricTools(api, client, identity) {
             }
         },
     }));
+    // -----------------------------------------------------------------
+    // fabric-send-message: post a message into a specific channel.
+    //
+    // Unlike a normal channel reply (which goes back to whatever channel
+    // woke the agent), this lets the agent proactively initiate text into
+    // any channel they are a member of — e.g. ARD broadcasting daily
+    // workload to #agents-room, or triage agent following up on an
+    // already-routed task by commenting in #updates.
+    // -----------------------------------------------------------------
+    api.registerTool((ctx) => ({
+        name: 'fabric-send-message',
+        description: 'Send a text message into a specific Fabric channel. Author is the calling agent. ' +
+            'Requires guildNodeId + channelId + content. Returns {ok, messageId, seq}.',
+        parameters: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['guildNodeId', 'channelId', 'content'],
+            properties: {
+                guildNodeId: { type: 'string' },
+                channelId: { type: 'string' },
+                content: { type: 'string', description: 'Message body (markdown supported by the renderer).' },
+            },
+        },
+        execute: async (_id, p) => {
+            const agentId = ctx.agentId;
+            if (!agentId)
+                return { ok: false, error: 'no agent context' };
+            const { session, guild, token } = await ctxGuild(agentId, p.guildNodeId);
+            const res = (await client.postMessage(guild.endpoint, token, p.channelId, p.content, session.user.id));
+            return { ok: true, messageId: res.messageId, seq: res.seq };
+        },
+    }));
+    // -----------------------------------------------------------------
+    // fabric-channel-list: enumerate channels the calling agent can see
+    // in a given guild. Backend filters to public channels + channels the
+    // agent is a member of. Returns id / name / xType per channel so the
+    // agent can pick a channelId for fabric-send-message etc.
+    // -----------------------------------------------------------------
+    api.registerTool((ctx) => ({
+        name: 'fabric-channel-list',
+        description: 'List channels visible to the calling agent in a guild. Optional ' +
+            'nameFilter does a case-insensitive substring match client-side. ' +
+            'Use this to find a channelId before fabric-send-message / fabric-message-history.',
+        parameters: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['guildNodeId'],
+            properties: {
+                guildNodeId: { type: 'string' },
+                nameFilter: { type: 'string', description: 'optional substring match on channel name (case-insensitive)' },
+                xType: {
+                    type: 'string',
+                    enum: ['general', 'work', 'report', 'discuss', 'triage', 'custom', 'dm'],
+                    description: 'optional filter by x_type',
+                },
+                includeClosed: { type: 'boolean', description: 'default false — closed channels filtered out' },
+            },
+        },
+        execute: async (_id, p) => {
+            const agentId = ctx.agentId;
+            if (!agentId)
+                return { ok: false, error: 'no agent context' };
+            const { guild, token } = await ctxGuild(agentId, p.guildNodeId);
+            const all = await client.listChannels(guild.endpoint, token, p.guildNodeId);
+            const needle = (p.nameFilter ?? '').toLowerCase();
+            const filtered = all.filter((c) => {
+                if (!p.includeClosed && c.closed)
+                    return false;
+                if (p.xType && c.xType !== p.xType)
+                    return false;
+                if (needle && !c.name.toLowerCase().includes(needle))
+                    return false;
+                return true;
+            });
+            return {
+                ok: true,
+                count: filtered.length,
+                channels: filtered.map((c) => ({
+                    id: c.id,
+                    name: c.name,
+                    xType: c.xType,
+                    isPublic: c.isPublic,
+                    closed: c.closed,
+                    lastSeq: c.lastSeq,
+                    purpose: c.purpose ?? null,
+                })),
+            };
+        },
+    }));
+    // -----------------------------------------------------------------
+    // fabric-guild-list: enumerate guilds the calling agent belongs to.
+    // Each row carries `purpose` — free-form description of what the
+    // guild is for (admin-set). Use this as the first step when a
+    // workflow says "find the right guild for X" — pick by purpose,
+    // then fabric-channel-list to find the right channel inside it.
+    // -----------------------------------------------------------------
+    api.registerTool((ctx) => ({
+        name: 'fabric-guild-list',
+        description: 'List guilds the calling agent is a member of. Returns ' +
+            '{nodeId, name, purpose, status} per row. ' +
+            "`purpose` is a free-form description of what each guild is for — " +
+            'pick the guild whose purpose matches your intent. Use this tool ' +
+            'BEFORE fabric-channel-list when a workflow asks you to pick the ' +
+            'right guild by intent (e.g. "find a guild whose purpose mentions ' +
+            'debate broadcasts" → then list its announce-type channels).',
+        parameters: {
+            type: 'object',
+            additionalProperties: false,
+            properties: {
+                nameFilter: {
+                    type: 'string',
+                    description: 'optional case-insensitive substring match on guild name',
+                },
+                purposeFilter: {
+                    type: 'string',
+                    description: 'optional case-insensitive substring match on guild purpose ' +
+                        '(e.g. "debate", "announcements")',
+                },
+            },
+        },
+        execute: async (_id, p) => {
+            const agentId = ctx.agentId;
+            if (!agentId)
+                return { ok: false, error: 'no agent context' };
+            const entry = identity.findByAgentId(agentId);
+            if (!entry)
+                return { ok: false, error: `agent ${agentId} not registered` };
+            const session = await client.agentLogin(entry.fabricApiKey);
+            const nameNeedle = (p.nameFilter ?? '').toLowerCase();
+            const purposeNeedle = (p.purposeFilter ?? '').toLowerCase();
+            const guilds = session.guilds.filter((g) => {
+                if (nameNeedle && !g.name.toLowerCase().includes(nameNeedle))
+                    return false;
+                if (purposeNeedle) {
+                    const purp = (g.purpose ?? '').toLowerCase();
+                    if (!purp.includes(purposeNeedle))
+                        return false;
+                }
+                return true;
+            });
+            return {
+                ok: true,
+                count: guilds.length,
+                guilds: guilds.map((g) => ({
+                    nodeId: g.nodeId,
+                    name: g.name,
+                    status: g.status,
+                    purpose: g.purpose ?? null,
+                })),
+            };
+        },
+    }));
+    // -----------------------------------------------------------------
+    // fabric-channel-set-purpose: set/update a channel's free-form
+    // purpose description. Caller must be a channel member (or the
+    // channel must be public). Use this to backfill purpose on existing
+    // channels, or to refine it after a channel's role evolves.
+    // -----------------------------------------------------------------
+    api.registerTool((ctx) => ({
+        name: 'fabric-channel-set-purpose',
+        description: "Set or update a channel's free-form purpose description. " +
+            'Channel membership required (or the channel must be public). ' +
+            'Pass empty string to clear. Use this to make a channel ' +
+            'discoverable to other agents via fabric-channel-list.',
+        parameters: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['guildNodeId', 'channelId', 'purpose'],
+            properties: {
+                guildNodeId: { type: 'string' },
+                channelId: { type: 'string' },
+                purpose: {
+                    type: 'string',
+                    description: "What this channel is for. Pass '' (empty string) to clear.",
+                },
+            },
+        },
+        execute: async (_id, p) => {
+            const agentId = ctx.agentId;
+            if (!agentId)
+                return { ok: false, error: 'no agent context' };
+            const { guild, token } = await ctxGuild(agentId, p.guildNodeId);
+            const res = await client.setChannelPurpose(guild.endpoint, token, p.channelId, p.purpose);
+            return { ok: true, channel: res };
+        },
+    }));
+    // -----------------------------------------------------------------
+    // fabric-message-history: read a channel's recent message history by
+    // `seq`. Tail-by-default: when `seqFrom`/`seqTo` are omitted, returns
+    // the last `limit` messages (limit defaults to 20, max 200).
+    //
+    // Use cases: catch-up on a channel that was muted while the agent was
+    // gated; verify a previous message went through; lookup recent
+    // duplicates before opening a new task in triage.
+    // -----------------------------------------------------------------
+    api.registerTool((ctx) => ({
+        name: 'fabric-message-history',
+        description: "Read a channel's recent message history. Omit seqFrom/seqTo to " +
+            'tail (last `limit` messages, default 20, max 200). Backend ' +
+            'requires the calling agent to be a channel participant.',
+        parameters: {
+            type: 'object',
+            additionalProperties: false,
+            required: ['guildNodeId', 'channelId'],
+            properties: {
+                guildNodeId: { type: 'string' },
+                channelId: { type: 'string' },
+                seqFrom: { type: 'integer', minimum: 1, description: 'inclusive lower bound; default = tail' },
+                seqTo: { type: 'integer', minimum: 1, description: 'inclusive upper bound; default = channel head' },
+                limit: { type: 'integer', minimum: 1, maximum: 200, description: 'default 20' },
+            },
+        },
+        execute: async (_id, p) => {
+            const agentId = ctx.agentId;
+            if (!agentId)
+                return { ok: false, error: 'no agent context' };
+            const { guild, token } = await ctxGuild(agentId, p.guildNodeId);
+            const limit = p.limit ?? 20;
+            // Tail mode: discover channel head via channel listing, then ask
+            // for [head-limit+1, head]. Avoids needing the agent to know seq.
+            let seqFrom = p.seqFrom;
+            let seqTo = p.seqTo;
+            if (seqFrom === undefined && seqTo === undefined) {
+                const channels = await client.listChannels(guild.endpoint, token, p.guildNodeId);
+                const ch = channels.find((c) => c.id === p.channelId);
+                const head = ch?.lastSeq ?? 0;
+                seqFrom = Math.max(1, head - limit + 1);
+                seqTo = head;
+            }
+            const res = await client.listMessages(guild.endpoint, token, p.channelId, {
+                seqFrom,
+                seqTo,
+                limit,
+            });
+            return {
+                ok: true,
+                page: res.page,
+                messages: res.items.map((m) => ({
+                    messageId: m.messageId,
+                    seq: m.seq,
+                    authorUserId: m.authorUserId,
+                    content: m.content,
+                    createdAt: m.createdAt,
+                    isDeleted: m.isDeleted,
+                })),
+            };
+        },
+    }));
 }

index.ts

@@ -6,17 +6,27 @@
 import { defineChannelPluginEntry } from 'openclaw/plugin-sdk/core';
 import type { OpenClawPluginApi } from 'openclaw/plugin-sdk/core';
 import { fabricChannelPlugin } from './src/channel.js';
+import { flushAllFabric } from './src/coalesce.js';
+import { getChannelType, flushChannelMeta } from './src/channel-meta.js';
 import { FabricInbound } from './src/inbound.js';
 import { listEnabledFabricAccounts } from './src/accounts.js';
 import { registerFabricTools } from './src/tools.js';
 import { FabricClient } from './src/fabric-client.js';
 import { IdentityRegistry } from './src/identity.js';
 import { syncFabricCommands } from './src/command-sync.js';
+import { PresenceSync } from './src/presence-sync.js';
+import { SubDiscussionStore } from './src/sub-discussion-store.js';
+import { registerSubDiscussionHook } from './src/sub-discussion-hook.js';
 import path from 'node:path';
 import os from 'node:os';
 
 let runtimeRef: unknown = null;
 let inbound: FabricInbound | null = null;
+let presence: PresenceSync | null = null;
+// Periodic re-harvest of presence accounts so newly-connected agents
+// (registered through tool-based identity flow AFTER initial start)
+// get picked up. Cleared on gateway_stop.
+let presenceRefreshTimer: ReturnType<typeof setInterval> | null = null;
 
 export { fabricChannelPlugin } from './src/channel.js';
 
@@ -37,23 +47,79 @@ export default defineChannelPluginEntry({
       config?: unknown;
       pluginConfig?: { identityFilePath?: string };
       logger: { info: (m: string) => void; warn: (m: string) => void };
-      on: (ev: string, fn: () => void) => void;
+      on: (ev: string, fn: (...args: unknown[]) => unknown) => void;
       registerTool: (d: unknown) => void;
     };
-    const cfg = (api.config ?? {}) as { channels?: { fabric?: { centerApiBase?: string } } };
+    const cfg = (api.config ?? {}) as {
+      channels?: { fabric?: { centerApiBase?: string; commandsSyncKey?: string } };
+    };
     const centerApiBase = cfg.channels?.fabric?.centerApiBase ?? 'http://localhost:7001/api';
     const idFile =
       api.pluginConfig?.identityFilePath ??
       path.join(os.homedir(), '.openclaw', 'fabric-identity.json');
+    const subDiscussionFile = path.join(
+      os.homedir(),
+      '.openclaw',
+      'fabric-sub-discussion.json',
+    );
 
     // tools operate against a default Center; per-account keys come from config
     const client = new FabricClient(centerApiBase);
     const identity = new IdentityRegistry(idFile);
+    const subDiscussion = new SubDiscussionStore(subDiscussionFile);
     registerFabricTools(
       { registerTool: (d) => api.registerTool(d), logger: api.logger },
       client,
       identity,
+      subDiscussion,
+      cfg,
     );
+    // Per-(agent, channel) prompt injection for sub-discussion channels.
+    // Runs as a sibling to PrismFacet's before_prompt_build hook (and
+    // ClawPrompts' fabric-chat-injector); openclaw composes
+    // appendSystemContext from all registered handlers.
+    registerSubDiscussionHook(
+      { on: api.on, logger: api.logger },
+      subDiscussion,
+      identity,
+    );
+
+    // Cross-plugin API: globalThis.__fabric
+    // Consumed by ClawPrompts' fabric-chat-injector to narrow its prompt
+    // injection to DM-typed channels only. The channel-meta cache is
+    // populated lazily from inbound (message.created carries xType) and
+    // persisted to ~/.openclaw/fabric-channel-meta.json — so even the
+    // very first DM after a fresh gateway start hits cache from the
+    // previous run rather than firing the injector on the wrong type.
+    //
+    // null return = channel never seen (cache cold). Callers MUST NOT
+    // fall back to "assume DM" — fail closed on unknown.
+    {
+      const _G = globalThis as Record<string, unknown>;
+      _G['__fabric'] = {
+        getChannelType,
+        // Dynamic-subscription bridges: tools (notably `fabric-register`)
+        // call these to add/remove an account's inbound socket without
+        // a gateway restart. Both delegate to the live FabricInbound
+        // instance via the module-level `inbound` closure variable; the
+        // closures stay valid across gateway_start / gateway_stop
+        // because we re-assign the variable, not the property.
+        addAccount: async (entry: { agentId: string; fabricApiKey: string }) => {
+          if (!inbound) throw new Error('fabric inbound not ready yet (gateway not started?)');
+          await inbound.addAccount(entry);
+        },
+        removeAccount: (agentId: string) => {
+          if (!inbound) return;
+          inbound.removeAccount(agentId);
+        },
+      };
+      // Flush channel-meta cache when the gateway shuts down so
+      // recently-recorded xType entries don't get lost.
+      api.on('gateway_stop', () => {
+        try { flushChannelMeta(); } catch { /* ignore */ }
+      });
+      api.logger.info('fabric: __fabric cross-plugin API installed (getChannelType + addAccount + removeAccount)');
+    }
 
     api.on('gateway_start', () => {
       const _G = globalThis as Record<string, unknown>;
@@ -81,12 +147,37 @@ export default defineChannelPluginEntry({
         api.logger,
         accounts,
       );
-      void inbound.start();
+      // start() resolves once all accounts have attempted login; per-
+      // agent failures are logged but don't reject. Once it resolves we
+      // can harvest the presence accounts (those that DID log in have
+      // their fabricUserId + first guild endpoint populated).
+      void inbound.start().then(() => {
+        if (!inbound) return;
+        presence = new PresenceSync(api.logger, client);
+        presence.setAccounts(inbound.getPresenceAccounts());
+        presence.start();
+        api.logger.info(`fabric: presence-sync started for ${inbound.getPresenceAccounts().length} account(s)`);
+
+        // Re-harvest every 5 min: catches agents added via tool-based
+        // identity provisioning after gateway_start (recruitment flow).
+        // setAccounts is idempotent — duplicates collapse on agentId.
+        presenceRefreshTimer = setInterval(() => {
+          if (inbound && presence) presence.setAccounts(inbound.getPresenceAccounts());
+        }, 5 * 60_000);
+      });
       api.logger.info(`fabric: inbound started for ${accounts.length} account(s)`);
       void syncFabricCommands(client, cfg, accounts, api.logger);
     });
 
+    // Note: the per-turn coalesce flush happens deterministically in
+    // inbound.ts right after dispatchInboundReplyWithBase resolves (that
+    // is the real "all deliveries done" boundary; the agent_end hook fires
+    // BEFORE deliver()). gateway_stop only flushes any leftover buffer.
     api.on('gateway_stop', () => {
+      void flushAllFabric();
+      if (presenceRefreshTimer) { clearInterval(presenceRefreshTimer); presenceRefreshTimer = null; }
+      presence?.stop();
+      presence = null;
       inbound?.stop();
       inbound = null;
     });

openclaw.plugin.json

@@ -14,7 +14,17 @@
       "create-work-channel",
       "create-report-channel",
       "create-discussion-channel",
-      "discussion-complete"
+      "create-sub-discussion",
+      "discussion-complete",
+      "close-sub-discussion",
+      "fabric-canvas",
+      "fabric-channel",
+      "fabric-send-message",
+      "fabric-send-sys-msg",
+      "fabric-channel-list",
+      "fabric-message-history",
+      "fabric-guild-list",
+      "fabric-channel-set-purpose"
     ]
   },
   "configSchema": {
@@ -39,6 +49,15 @@
             "type": "string",
             "description": "Fabric Center API base, e.g. http://localhost:7001/api"
           },
+          "commandsSyncKey": {
+            "type": "string",
+            "minLength": 1,
+            "description": "Shared secret that must equal the guild's FABRIC_BACKEND_GUILD_COMMANDS_SYNC_KEY. Required to register the slash-command catalog (Guild C-2). Read it from the guild via: docker exec fabric-backend-guild node dist/cli/print-commands-sync-key.js"
+          },
+          "coalesce": {
+            "type": "boolean",
+            "description": "Merge a split agent turn (text → thinking/tool → text) into ONE Fabric message. Flushed deterministically on the agent_end hook. Default true; false = raw per-segment posting."
+          },
           "dmSecurity": { "type": "string" },
           "dmPolicy": { "type": "string" },
           "enabled": { "type": "boolean" },
@@ -59,10 +78,12 @@
               }
             }
           }
-        }
+        },
+        "required": ["commandsSyncKey"]
       },
       "uiHints": {
-        "centerApiBase": { "label": "Center API base" }
+        "centerApiBase": { "label": "Center API base" },
+        "commandsSyncKey": { "label": "Commands sync key" }
       }
     }
   }

src/accounts.ts

@@ -14,6 +14,15 @@ export type FabricAccountConfig = {
 
 export type FabricChannelConfig = {
   centerApiBase?: string;
+  // Shared secret matching the guild's FABRIC_BACKEND_GUILD_COMMANDS_SYNC_KEY
+  // (Guild C-2). Required by the channel config schema; sourced from config
+  // only — never from the environment.
+  commandsSyncKey?: string;
+  // Coalesce an agent turn that OpenClaw split into multiple deliveries
+  // (text → thinking/tool → text => N sendText calls) into ONE Fabric
+  // message. The flush boundary is the deterministic `agent_end` hook (not
+  // a timer). Default true; set false for raw per-segment posting.
+  coalesce?: boolean;
   accounts?: Record<string, FabricAccountConfig>;
   defaultAccount?: string;
 } & FabricAccountConfig;
@@ -35,6 +44,19 @@ function section(cfg: Cfg): FabricChannelConfig {
   return cfg.channels?.fabric ?? {};
 }
 
+// The commands-sync shared secret (channel-level only). Empty string when
+// unconfigured — callers decide how to handle (slash-command sync is then
+// rejected by the guild).
+export function resolveCommandsSyncKey(cfg: Cfg): string {
+  return (section(cfg).commandsSyncKey ?? '').trim();
+}
+
+// Whether to coalesce a split agent turn into one Fabric message
+// (channel-level). Default true.
+export function resolveCoalesce(cfg: Cfg): boolean {
+  return (cfg.channels?.fabric ?? {}).coalesce !== false;
+}
+
 export function listFabricAccountIds(cfg: Cfg): string[] {
   const accts = section(cfg).accounts ?? {};
   const ids = Object.keys(accts);

src/channel-meta.ts

@@ -0,0 +1,108 @@
+/**
+ * Channel-meta cache. Records (channelId → xType) for every fabric
+ * channel the gateway has seen at least one inbound message in.
+ *
+ * Populated lazily from inbound (`recordChannelType` is called for
+ * every `message.created` event with non-empty `xType`). Persisted to
+ * `~/.openclaw/fabric-channel-meta.json` so the cache survives
+ * gateway restarts (so the very first DM after restart still gets the
+ * right xType without waiting for a fresh inbound).
+ *
+ * Exposed cross-plugin via `globalThis.__fabric.getChannelType`. Used
+ * by ClawPrompts' fabric-chat-injector to narrow its prompt injection
+ * to xType==='dm' only.
+ *
+ * Failure mode: lookup misses (channel never seen / inbound dropped
+ * xType) return null. Callers MUST treat null as "unknown" — DO NOT
+ * fall back to "assume DM" or you re-introduce the false-positive on
+ * group channels.
+ */
+import { existsSync, mkdirSync, readFileSync, renameSync, writeFileSync } from 'node:fs';
+import { dirname, join } from 'node:path';
+import { homedir } from 'node:os';
+
+const CACHE_FILE = join(homedir(), '.openclaw', 'fabric-channel-meta.json');
+
+interface ChannelMetaFile {
+  // channelId → xType ('dm' | 'triage' | 'group' | etc.)
+  channels: Record<string, string>;
+}
+
+let memory = new Map<string, string>();
+let loaded = false;
+let dirty = false;
+let flushTimer: ReturnType<typeof setTimeout> | null = null;
+
+function load(): void {
+  if (loaded) return;
+  loaded = true;
+  try {
+    if (!existsSync(CACHE_FILE)) return;
+    const raw = readFileSync(CACHE_FILE, 'utf8');
+    const parsed = JSON.parse(raw) as ChannelMetaFile;
+    for (const [k, v] of Object.entries(parsed.channels ?? {})) {
+      if (typeof k === 'string' && typeof v === 'string') memory.set(k, v);
+    }
+  } catch {
+    // ignore — start with empty cache on corruption
+  }
+}
+
+function scheduleFlush(): void {
+  if (flushTimer) return;
+  // Debounce writes — many inbound messages may arrive in a burst.
+  // 250ms coalesces them; on gateway_stop the channel plugin can force
+  // a synchronous flush via flushChannelMeta().
+  flushTimer = setTimeout(() => {
+    flushTimer = null;
+    if (!dirty) return;
+    dirty = false;
+    flushSync();
+  }, 250);
+}
+
+function flushSync(): void {
+  try {
+    const dir = dirname(CACHE_FILE);
+    if (!existsSync(dir)) mkdirSync(dir, { recursive: true });
+    const out: ChannelMetaFile = { channels: Object.fromEntries(memory) };
+    const tmp = CACHE_FILE + '.tmp';
+    writeFileSync(tmp, JSON.stringify(out, null, 2) + '\n', 'utf8');
+    renameSync(tmp, CACHE_FILE);
+  } catch {
+    // swallow — cache is an optimization; loss-on-write is recoverable
+  }
+}
+
+/** Called by inbound on every message.created. xType empty → no-op. */
+export function recordChannelType(channelId: string, xType: string | undefined): void {
+  if (!channelId || !xType) return;
+  load();
+  const existing = memory.get(channelId);
+  if (existing === xType) return;
+  memory.set(channelId, xType);
+  dirty = true;
+  scheduleFlush();
+}
+
+/** Cross-plugin lookup. null when channel never seen / unknown. */
+export function getChannelType(channelId: string): string | null {
+  if (!channelId) return null;
+  load();
+  return memory.get(channelId) ?? null;
+}
+
+/** Force-flush — called on plugin shutdown to make sure recently
+ *  recorded entries hit disk before the gateway dies. */
+export function flushChannelMeta(): void {
+  if (flushTimer) {
+    clearTimeout(flushTimer);
+    flushTimer = null;
+  }
+  if (dirty) {
+    dirty = false;
+    flushSync();
+  }
+}
+
+export const CHANNEL_META_PATH = CACHE_FILE;

src/channel.ts

@@ -21,6 +21,39 @@ import {
   resolveDefaultFabricAccountId,
   type ResolvedFabricAccount,
 } from './accounts.js';
+import { getChannelType } from './channel-meta.js';
+
+/**
+ * Map a Fabric channel xType to an openclaw routing peer.kind / ChatType.
+ *
+ * Fabric distinguishes channels by xType ('dm' | 'triage' | 'group' |
+ * 'broadcast' | 'announce' | ...). Openclaw's session router only knows
+ * 'direct' | 'group' | 'channel'. We collapse:
+ *   - 'dm'  → 'direct' (1:1 conversation; agent always speaks)
+ *   - rest  → 'group'  (multi-party; turn-engine gates speech)
+ *
+ * Sessions are keyed by peer.kind, so inbound and outbound MUST agree —
+ * otherwise the agent's outbound message lands in a different session
+ * than the inbound that triggered it and conversation state splits.
+ *
+ * Outbound has no live xType (the agent target is just a channelId), so
+ * it consults the channel-meta cache populated by inbound. Cache miss
+ * (channel never observed) falls back to 'group' — same as the pre-fix
+ * behavior, no regression on cold cache. The proactive-DM-first-message
+ * edge case (agent DMs a channel before any inbound) still lands as
+ * 'group' on that one outbound; the next inbound + outbound pair will
+ * agree on 'direct'.
+ */
+export type FabricPeerRouting = { peerKind: 'direct' | 'group'; chatType: 'direct' | 'group' };
+
+export function fabricPeerRoutingForXType(xType: string | null | undefined): FabricPeerRouting {
+  if (xType === 'dm') return { peerKind: 'direct', chatType: 'direct' };
+  return { peerKind: 'group', chatType: 'group' };
+}
+
+export function fabricPeerRoutingForChannel(channelId: string): FabricPeerRouting {
+  return fabricPeerRoutingForXType(getChannelType(channelId));
+}
 
 type AnyCfg = { channels?: { fabric?: unknown }; [k: string]: unknown };
 
@@ -45,13 +78,18 @@ export function looksLikeFabricTargetId(raw: string): boolean {
 export function resolveFabricOutboundSessionRoute(params: ChannelOutboundSessionRouteParams) {
   const id = stripFabricTargetPrefix(params.target);
   if (!id) return null;
+  // Consult the channel-meta cache populated by inbound — DM channels
+  // need peer.kind='direct' so the outbound session key matches the
+  // inbound one. Cache miss falls back to 'group' (the pre-fix default,
+  // no regression on cold cache).
+  const { peerKind, chatType } = fabricPeerRoutingForChannel(id);
   return buildChannelOutboundSessionRoute({
     cfg: params.cfg,
     agentId: params.agentId,
     channel: 'fabric',
     accountId: params.accountId,
-    peer: { kind: 'group', id },
-    chatType: 'group',
+    peer: { kind: peerKind, id },
+    chatType,
     from: `fabric:channel:${id}`,
     to: `fabric:${id}`,
   });
@@ -115,6 +153,20 @@ export const fabricChannelPlugin = createChatChannelPlugin<ResolvedFabricAccount
       resolveAccount: (cfg, accountId) => resolveFabricAccount(cfg as never, accountId),
       defaultAccountId: (cfg) => resolveDefaultFabricAccountId(cfg as never),
       isConfigured: (account: ResolvedFabricAccount) => Boolean(account.fabricApiKey),
+      // openclaw's channelManager.getRuntimeSnapshot() — called every minute
+      // by the channel-health-monitor — defaults `configured: true` when the
+      // plugin doesn't expose describeAccount (see applyDescribedAccountFields
+      // in server-channels). Without this, fabric's synthetic 'default'
+      // account (returned by listFabricAccountIds when channels.fabric.accounts
+      // is empty — the prod shape) gets snapshot {enabled:true, configured:true,
+      // running:false} → isManagedAccount=true → not-running → restart loop
+      // every ~10 min, logging `[fabric:default] health-monitor: restarting`.
+      // Mirror isConfigured here so the snapshot truthfully reports false for
+      // any account without a fabricApiKey.
+      describeAccount: (account: ResolvedFabricAccount) => ({
+        accountId: account.accountId,
+        configured: Boolean(account.fabricApiKey),
+      }),
     },
     // Minimal setup adapter: Fabric is configured directly under
     // channels.fabric.* (no interactive wizard). applyAccountConfig is the
@@ -159,7 +211,10 @@ export const fabricChannelPlugin = createChatChannelPlugin<ResolvedFabricAccount
         cfg?: unknown;
         config?: unknown;
       }) => {
-        // openclaw passes config under cfg or config depending on path
+        // openclaw passes config under cfg or config depending on path.
+        // Note: inbound agent replies go through inbound.ts `deliver`
+        // (where turn coalescing happens). This path is for any direct
+        // outbound sends and posts immediately.
         const cfg = (ctx.cfg ?? ctx.config ?? {}) as AnyCfg;
         try {
           const r = await sendToFabric(cfg, ctx.accountId ?? null, ctx.to, ctx.text);

src/coalesce.ts

@@ -0,0 +1,93 @@
+// Deterministic turn coalescer.
+//
+// OpenClaw calls the Fabric `deliver` callback once per assistant text
+// segment; a thinking/tool block between two text blocks is a delivery
+// boundary, so one agent turn of `text → thinking/tool → text` arrives as
+// multiple deliver() calls. There is no turn id on the delivery, so we
+// BUFFER segments by Fabric channelId and post the merged message when the
+// turn truly ends. The flush is driven by inbound.ts right after
+// `dispatchInboundReplyWithBase` resolves — that only happens AFTER every
+// deliver() of the turn, a deterministic boundary (NOT a timer, NOT the
+// agent_end hook, which fires before deliver()). `coalesce=false` posts
+// each segment immediately.
+
+const SAFETY_FLUSH_MS = 120_000; // leak-guard only; not the flush mechanism
+
+export function normChannelId(x: string | null | undefined): string {
+  const s = String(x ?? '');
+  return s.startsWith('fabric:') ? s.slice('fabric:'.length) : s;
+}
+
+type Pending = {
+  parts: string[];
+  post: (text: string) => Promise<void>;
+  log?: (m: string) => void;
+  safety: ReturnType<typeof setTimeout>;
+};
+const pendingByChannel = new Map<string, Pending>();
+
+async function flushChannel(channelId: string, reason: string): Promise<void> {
+  const p = pendingByChannel.get(channelId);
+  if (!p) return;
+  pendingByChannel.delete(channelId);
+  clearTimeout(p.safety);
+  const text = p.parts.join('\n\n').trim();
+  if (!text) return;
+  try {
+    await p.post(text);
+    p.log?.(`fabric: flushed ${p.parts.length} segment(s) channel=${channelId} (${reason})`);
+  } catch (e) {
+    p.log?.(`fabric: flush FAILED channel=${channelId} (${reason}): ${String(e)}`);
+  }
+}
+
+// Buffer one delivered segment (or send immediately when coalesce=false).
+// `post` performs the real Fabric postMessage with the caller's already
+// resolved guild/token; on flush it is called once with the merged text.
+export async function enqueueDelivery(params: {
+  channelId: string;
+  text: string;
+  coalesce: boolean;
+  post: (text: string) => Promise<void>;
+  log?: (m: string) => void;
+}): Promise<void> {
+  const cid = normChannelId(params.channelId);
+  const text = (params.text ?? '').trim();
+  if (!text) return;
+  if (!params.coalesce) {
+    await params.post(text);
+    return;
+  }
+  const existing = pendingByChannel.get(cid);
+  if (existing) {
+    existing.parts.push(text);
+    existing.post = params.post; // freshest guild/token closure
+    existing.log = params.log;
+  } else {
+    pendingByChannel.set(cid, {
+      parts: [text],
+      post: params.post,
+      log: params.log,
+      safety: setTimeout(
+        () => void flushChannel(cid, 'safety-timeout'),
+        SAFETY_FLUSH_MS,
+      ),
+    });
+  }
+}
+
+// Called by the agent_end hook with the hook ctx's channelId (bare or
+// fabric:-prefixed). Deterministic per-turn boundary.
+export async function flushFabricForChannel(
+  rawChannelId: string | null | undefined,
+): Promise<void> {
+  const cid = normChannelId(rawChannelId);
+  if (cid) await flushChannel(cid, 'dispatch-end');
+}
+
+// gateway_stop: flush anything still buffered.
+export async function flushAllFabric(): Promise<void> {
+  for (const cid of [...pendingByChannel.keys()]) {
+    await flushChannel(cid, 'gateway_stop');
+  }
+}

src/command-sync.ts

@@ -11,6 +11,7 @@ import {
   resolveCommandArgChoices,
 } from 'openclaw/plugin-sdk/native-command-registry';
 import type { FabricClient } from './fabric-client.js';
+import { resolveCommandsSyncKey } from './accounts.js';
 
 type Logger = { info: (m: string) => void; warn: (m: string) => void };
 
@@ -102,6 +103,17 @@ export async function syncFabricCommands(
   accounts: Array<{ agentId: string; fabricApiKey: string }>,
   log: Logger,
 ): Promise<void> {
+  // Guild C-2: the sync key comes from the channel config only (schema
+  // marks it required). Without it the guild rejects the catalog write.
+  const syncKey = resolveCommandsSyncKey(cfg as never);
+  if (!syncKey) {
+    log.warn(
+      'fabric: channels.fabric.commandsSyncKey is not set — skipping ' +
+        'slash-command sync (set it to the guild FABRIC_BACKEND_GUILD_COMMANDS_SYNC_KEY)',
+    );
+    return;
+  }
+
   let specs: FabricCommand[];
   try {
     specs = buildFabricCommandSpecs(cfg);
@@ -126,7 +138,7 @@ export async function syncFabricCommands(
       )?.token;
       if (!tok) continue;
       try {
-        await client.syncCommands(g.endpoint, tok, specs);
+        await client.syncCommands(g.endpoint, tok, specs, syncKey);
         done.add(g.nodeId);
         log.info(`fabric: synced ${specs.length} slash command(s) -> ${g.nodeId}`);
       } catch (err) {

src/fabric-client.ts

@@ -6,7 +6,15 @@ export type FabricSession = {
   accessToken: string;
   refreshToken: string;
   user: { id: string; email: string; name: string };
-  guilds: Array<{ nodeId: string; name: string; endpoint: string; status: string }>;
+  guilds: Array<{
+    nodeId: string;
+    name: string;
+    endpoint: string;
+    status: string;
+    // free-form description of this guild's role; admin-set on Center.
+    // null when the admin hasn't filled it in yet.
+    purpose?: string | null;
+  }>;
   guildAccessTokens: Array<{ guildNodeId: string; token: string }>;
 };
 
@@ -36,12 +44,14 @@ export class FabricClient {
     url: string,
     auth?: string,
     body?: unknown,
+    extraHeaders?: Record<string, string>,
   ): Promise<T> {
     const res = await fetch(url, {
       method,
       headers: {
         ...(body !== undefined ? { 'content-type': 'application/json' } : {}),
         ...(auth ? { authorization: `Bearer ${auth}` } : {}),
+        ...(extraHeaders ?? {}),
       },
       body: body !== undefined ? JSON.stringify(body) : undefined,
     });
@@ -98,11 +108,30 @@ export class FabricClient {
       memberUserIds?: string[];
       onDuty?: string;
       listeners?: string[];
+      // free-form purpose; optional. Existing agents can also set/update
+      // it later via setChannelPurpose().
+      purpose?: string;
     },
   ): Promise<{ id: string }> {
     return this.post(`${guildEndpoint}/api/channels`, body, guildToken);
   }
 
+  // PATCH /api/channels/:id — backend currently only patches `purpose`.
+  // Caller must be a member of the channel (or any user if public).
+  setChannelPurpose(
+    guildEndpoint: string,
+    guildToken: string,
+    channelId: string,
+    purpose: string,
+  ): Promise<{ id: string; name: string; xType: string; purpose: string | null }> {
+    return this.req(
+      'PATCH',
+      `${guildEndpoint}/api/channels/${channelId}`,
+      guildToken,
+      { purpose },
+    );
+  }
+
   closeChannel(guildEndpoint: string, guildToken: string, channelId: string): Promise<unknown> {
     return this.post(`${guildEndpoint}/api/channels/${channelId}/close`, {}, guildToken);
   }
@@ -122,8 +151,18 @@ export class FabricClient {
     guildEndpoint: string,
     guildToken: string,
     commands: unknown[],
+    syncKey: string,
   ): Promise<unknown> {
-    return this.req('PUT', `${guildEndpoint}/api/commands`, guildToken, { commands });
+    // Guild C-2: the shared key is sourced from the channel config
+    // (channels.fabric.commandsSyncKey) and must equal the guild's
+    // FABRIC_BACKEND_GUILD_COMMANDS_SYNC_KEY for the catalog write.
+    return this.req(
+      'PUT',
+      `${guildEndpoint}/api/commands`,
+      guildToken,
+      { commands },
+      syncKey ? { 'x-commands-sync-key': syncKey } : undefined,
+    );
   }
 
   // [{ userId, bypass }] — bypass is true only for discuss/work bypass-list
@@ -178,6 +217,77 @@ export class FabricClient {
   removeCanvas(endpoint: string, token: string, channelId: string): Promise<unknown> {
     return this.req('DELETE', this.canvasUrl(endpoint, channelId), token);
   }
+
+  // ---- channel discovery + message read (used by the agent-facing
+  // fabric-channel-list / fabric-message-history tools) ----
+
+  /**
+   * List channels in a guild visible to the calling user. Backend
+   * filters to public + channels the user is a member of.
+   */
+  listChannels(
+    guildEndpoint: string,
+    guildToken: string,
+    guildNodeId: string,
+  ): Promise<Array<{
+    id: string;
+    guildId: string;
+    name: string;
+    xType: string;
+    kind: string;
+    isPublic: boolean;
+    closed: boolean;
+    lastSeq: number;
+    createdAt: string;
+    purpose?: string | null;
+  }>> {
+    return this.req(
+      'GET',
+      `${guildEndpoint}/api/channels?guildId=${encodeURIComponent(guildNodeId)}`,
+      guildToken,
+    );
+  }
+
+  /**
+   * Page through a channel's message history by `seq`.
+   *
+   * Backend defaults: 50 / call, max 200. The `seq` field starts at 1
+   * per channel; pass `seqFrom=channel.lastSeq - N + 1` to get the
+   * tail. Page metadata in the response describes what to ask next.
+   */
+  listMessages(
+    guildEndpoint: string,
+    guildToken: string,
+    channelId: string,
+    opts: { seqFrom?: number; seqTo?: number; limit?: number } = {},
+  ): Promise<{
+    items: Array<{
+      messageId: string;
+      seq: number;
+      content: string;
+      authorUserId: string;
+      createdAt: string;
+      editedAt: string | null;
+      deletedAt: string | null;
+      isDeleted: boolean;
+    }>;
+    page: {
+      seqFrom: number;
+      seqTo: number;
+      limit: number;
+      returned: number;
+      hasMore: boolean;
+      nextExpectedSeq: number;
+      highestCommittedSeq: number;
+    };
+  }> {
+    const qs = new URLSearchParams();
+    if (opts.seqFrom !== undefined) qs.set('seq_from', String(opts.seqFrom));
+    if (opts.seqTo !== undefined) qs.set('seq_to', String(opts.seqTo));
+    if (opts.limit !== undefined) qs.set('limit', String(opts.limit));
+    const url = `${guildEndpoint}/api/channels/${channelId}/messages` + (qs.toString() ? `?${qs}` : '');
+    return this.req('GET', url, guildToken);
+  }
 }
 
 export type CanvasFormat = 'md' | 'html' | 'text';

src/inbound.ts

@@ -5,6 +5,10 @@ import { io, type Socket } from 'socket.io-client';
 import { dispatchInboundReplyWithBase } from 'openclaw/plugin-sdk/inbound-reply-dispatch';
 import type { FabricClient, FabricSession } from './fabric-client.js';
 import type { IdentityRegistry } from './identity.js';
+import { resolveCoalesce } from './accounts.js';
+import { fabricPeerRoutingForXType } from './channel.js';
+import { recordChannelType } from './channel-meta.js';
+import { enqueueDelivery, flushFabricForChannel } from './coalesce.js';
 
 // COMPAT NOTE (openclaw v2026.5.7): the inbound path mirrors how bundled
 // channels (nextcloud-talk) drive the kernel:
@@ -42,11 +46,52 @@ type FabricMessage = {
   channelId?: string;
   attachments?: FabricAttachment[];
   wakeup?: boolean;
+  // x-type of the channel (sent on message.created). 'dm' bypasses the
+  // wakeup gate: any message that isn't the agent's own is delivered.
+  xType?: string;
 };
 
+// Walk cfg.bindings for the entry that ties `agentId` to a fabric account.
+// Returns the binding's match.accountId (the slot label routing keys on);
+// returns undefined when the agent has no explicit fabric binding so the
+// caller can fall back to agentId without changing pre-existing semantics
+// for agents whose binding accountId == agent_id anyway.
+function findFabricBindingAccountId(cfg: unknown, agentId: string): string | undefined {
+  const bindings = (cfg as { bindings?: Array<{
+    agentId?: string;
+    match?: { channel?: string; accountId?: string };
+  }> })?.bindings;
+  if (!Array.isArray(bindings)) return undefined;
+  for (const b of bindings) {
+    if (
+      b?.agentId === agentId &&
+      b?.match?.channel === 'fabric' &&
+      typeof b?.match?.accountId === 'string' &&
+      b.match.accountId.length > 0
+    ) {
+      return b.match.accountId;
+    }
+  }
+  return undefined;
+}
+
 export class FabricInbound {
   private sockets: Socket[] = [];
+  // Per-agent socket + timer tracking. Enables `removeAccount(agentId)`
+  // to tear down ONE agent without restarting the whole inbound. New
+  // sockets get appended on `connectAgent`; both maps are emptied by
+  // `stop()`.
+  private socketsByAgent = new Map<string, Socket[]>();
+  private timersByAgent = new Map<string, NodeJS.Timeout[]>();
   private seen = new Set<string>();
+  // Timers that periodically re-sync channel membership per (agent, guild).
+  // Without this, the agent's socket.io subscriptions are a snapshot taken
+  // at connect time — any channel the agent joins later (e.g. a fresh DM
+  // created by another user) is unreachable until the gateway restarts.
+  private channelSyncTimers: NodeJS.Timeout[] = [];
+  // Resync cadence. Backend doesn't push a `channel.joined` event, so we
+  // poll. 60s keeps the lag bounded without hammering the backend.
+  private static readonly CHANNEL_SYNC_INTERVAL_MS = 60_000;
   // Guild access tokens are short-lived (~15 min). The socket survives via
   // socket.io reconnect, but the token captured at connect time goes stale,
   // so HTTP calls (attachment download, posting the reply) start 401ing.
@@ -54,6 +99,136 @@ export class FabricInbound {
   private tokenCache = new Map<string, { session: FabricSession; at: number }>();
   private static readonly TOKEN_TTL_MS = 8 * 60 * 1000;
 
+  // Per-channel serial work queue. Every inbound socket message for a
+  // channel awaits the previous task for that same channel, so model
+  // turns never interleave. Map key = channelId; value is the tail of
+  // the chain (an in-flight promise the next task awaits).
+  //
+  // Why per-channel and not per-agent: a single agent may sit in
+  // several triage / general channels; we want each channel to flow at
+  // its own speed but the SAME channel's traffic to be strictly serial.
+  // For dm and discuss the queue also serialises but those traditionally
+  // had at-most-one-in-flight anyway via the turn engine.
+  private channelChains = new Map<string, Promise<void>>();
+
+  // Agent.status snapshot cache (5s TTL) — keeps the HF /calendar/
+  // agent/status round-trip off the hot path for back-to-back triage
+  // messages. Short TTL because status flips are rare-but-meaningful.
+  private agentStatusCache = new Map<string, { onCall: boolean; at: number }>();
+  private static readonly AGENT_STATUS_TTL_MS = 5_000;
+
+  // Triage messages that arrived while the on-duty agent wasn't on_call
+  // — sit here until either (a) the agent becomes on_call and the next
+  // triage arrival drains them, or (b) the gateway restarts (lost; ok
+  // because the underlying Fabric messages are persisted and re-fetched
+  // on agent reconnect's history sweep).
+  private pendingTriageGated: Array<{
+    agentId: string;
+    g: { nodeId: string; endpoint: string };
+    channelId: string;
+    m: FabricMessage;
+    session: FabricSession;
+  }> = [];
+
+  // Schedule `task` to run after every previous task on the same
+  // channel has completed. Returns the promise so callers can await
+  // their own result if they need to; the chain itself is fire-and-
+  // forget from the socket.on handler.
+  private enqueueChannelTask(channelId: string, task: () => Promise<void>): Promise<void> {
+    const prev = this.channelChains.get(channelId) ?? Promise.resolve();
+    const next = prev.then(task).catch((err) => {
+      this.log.warn(`fabric: per-channel task failed channel=${channelId}: ${String(err)}`);
+    });
+    this.channelChains.set(channelId, next);
+    // Best-effort cleanup so the Map doesn't grow without bound for
+    // long-running gateways: drop the entry when the chain settles, but
+    // only if it's still the latest reference (newer enqueue may have
+    // overwritten it in the meantime).
+    void next.finally(() => {
+      if (this.channelChains.get(channelId) === next) {
+        this.channelChains.delete(channelId);
+      }
+    });
+    return next;
+  }
+
+  // Hit HF backend to check whether `agentId` is currently on_call.
+  // Cached for 5s. Failures (network, 404, etc.) are treated as "not
+  // on_call" — triage stays gated rather than risking a confused wake.
+  private async checkAgentOnCall(agentId: string): Promise<boolean> {
+    const cached = this.agentStatusCache.get(agentId);
+    if (cached && Date.now() - cached.at < FabricInbound.AGENT_STATUS_TTL_MS) {
+      return cached.onCall;
+    }
+    const base = (process.env.HF_API_BASE_URL ?? '').trim() || 'https://monitor.hangman-lab.top';
+    // CLAW_IDENTIFIER resolution priority:
+    //   1. HF_CLAW_IDENTIFIER env (operator override)
+    //   2. openclaw config `plugins.harbor-forge.identifier` (what the HF
+    //      plugin itself uses — keeps the two in sync without an extra
+    //      env per service unit)
+    //   3. os.hostname() last-resort fallback (often wrong: e.g. sim
+    //      container hostname is `server.t2` but HF agent row has
+    //      `claw_identifier=sim-t2`; matching is mandatory for the HF
+    //      backend's _require_agent() check)
+    let claw = (process.env.HF_CLAW_IDENTIFIER ?? '').trim();
+    if (!claw) {
+      try {
+        // openclaw config shape (verified in sim):
+        //   { plugins: { entries: { 'harbor-forge': { config: { identifier } } } } }
+        const cfg = this.cfg as {
+          plugins?: { entries?: Record<string, { config?: { identifier?: string } }> };
+        };
+        const fromCfg = cfg?.plugins?.entries?.['harbor-forge']?.config?.identifier;
+        if (fromCfg && typeof fromCfg === 'string' && fromCfg.trim()) {
+          claw = fromCfg.trim();
+        }
+      } catch {
+        /* fall through to hostname */
+      }
+    }
+    if (!claw) {
+      claw = (await import('os')).hostname();
+    }
+    let onCall = false;
+    try {
+      const url = `${base.replace(/\/$/, '')}/calendar/agent/status?agent_id=${encodeURIComponent(agentId)}`;
+      const res = await fetch(url, {
+        headers: { 'X-Agent-ID': agentId, 'X-Claw-Identifier': claw },
+      });
+      if (res.ok) {
+        const data = (await res.json()) as { status?: string };
+        onCall = (data.status ?? '').toLowerCase() === 'on_call';
+      }
+    } catch (err) {
+      this.log.warn(`fabric: HF status check failed agent=${agentId}: ${String(err)}`);
+    }
+    this.agentStatusCache.set(agentId, { onCall, at: Date.now() });
+    return onCall;
+  }
+
+  // FIFO drain of all triage-gated messages for `agentId` (called when
+  // we just learned they're on_call). Each drained message is dispatched
+  // through its own channel chain so per-channel serial order is kept.
+  private async drainGatedFor(agentId: string): Promise<void> {
+    const keep: typeof this.pendingTriageGated = [];
+    const drain: typeof this.pendingTriageGated = [];
+    for (const item of this.pendingTriageGated) {
+      if (item.agentId === agentId) drain.push(item);
+      else keep.push(item);
+    }
+    if (drain.length === 0) return;
+    this.pendingTriageGated = keep;
+    for (const item of drain) {
+      this.log.info(
+        `fabric: triage drain agent=${item.agentId} channel=${item.channelId} msg=${item.m.messageId}`,
+      );
+      // Re-enqueue via the per-channel chain so ordering is preserved.
+      this.enqueueChannelTask(item.channelId, async () => {
+        await this.dispatch(item.agentId, item.g, item.channelId, item.m, item.session);
+      });
+    }
+  }
+
   // Return a fresh guild access token for the agent, re-authenticating with
   // the agent's Fabric API key when the cached session is stale. Falls back
   // to the connect-time session token if re-login fails.
@@ -114,46 +289,293 @@ export class FabricInbound {
   }
 
   stop(): void {
+    for (const t of this.channelSyncTimers) clearInterval(t);
+    this.channelSyncTimers = [];
     for (const s of this.sockets) s.disconnect();
     this.sockets = [];
+    this.socketsByAgent.clear();
+    this.timersByAgent.clear();
   }
 
+  /**
+   * Bring up ONE new account at runtime (no gateway restart).
+   *
+   * Mirrors what `start()` does per entry: login to Center, upsert the
+   * identity registry, open the socket(s). Idempotent: re-calling with
+   * the same agentId tears down the previous socket(s) first so the
+   * fresh apikey replaces the stale one (recruitment onboard rotates
+   * the agent from the shared `interviewee` placeholder to a real
+   * per-agent apikey — the old `interviewee` socket must drop before
+   * the new one comes up or the agent ends up subscribed to both users
+   * at once).
+   *
+   * Used by the `fabric-register` openclaw tool to make recruitment
+   * end-to-end without a gateway restart between `new-agent` and the
+   * interview's sub-discussion dispatch.
+   */
+  async addAccount(entry: { agentId: string; fabricApiKey: string }): Promise<void> {
+    if (this.socketsByAgent.has(entry.agentId)) {
+      this.removeAccount(entry.agentId);
+    }
+    const session = await this.client.agentLogin(entry.fabricApiKey);
+    this.identity.upsert({
+      agentId: entry.agentId,
+      fabricApiKey: entry.fabricApiKey,
+      fabricUserId: session.user.id,
+      displayName: session.user.name,
+    });
+    await this.connectAgent(entry.agentId, session);
+    this.log.info(`fabric: agent ${entry.agentId} dynamically added as ${session.user.email}`);
+  }
+
+  /**
+   * Tear down ONE account's sockets + timers without touching others.
+   * Caller is responsible for any identity-registry cleanup; this only
+   * drops the live socket subscription so the agent stops receiving
+   * Fabric pushes.
+   */
+  removeAccount(agentId: string): void {
+    const sockets = this.socketsByAgent.get(agentId);
+    if (sockets) {
+      for (const s of sockets) {
+        try { s.disconnect(); } catch { /* socket already dead */ }
+        // Also remove from the flat list so `stop()` doesn't double-close.
+        const idx = this.sockets.indexOf(s);
+        if (idx !== -1) this.sockets.splice(idx, 1);
+      }
+      this.socketsByAgent.delete(agentId);
+    }
+    const timers = this.timersByAgent.get(agentId);
+    if (timers) {
+      for (const t of timers) {
+        clearInterval(t);
+        const idx = this.channelSyncTimers.indexOf(t);
+        if (idx !== -1) this.channelSyncTimers.splice(idx, 1);
+      }
+      this.timersByAgent.delete(agentId);
+    }
+    this.firstGuildByAgent.delete(agentId);
+    this.tokenCache.delete(agentId);
+    this.agentStatusCache.delete(agentId);
+    this.log.info(`fabric: agent ${agentId} dynamically removed`);
+  }
+
+  /**
+   * Per-account metadata harvested during `start()` — used by
+   * PresenceSync to know where to push each agent's HF status.
+   *
+   * `fabricUserId` is filled from `session.user.id` after agent-login.
+   * `guildBaseUrl` is the FIRST guild the agent is connected to (multi-
+   * guild presence push is a future concern; for sim/prod-v1 each agent
+   * is in one guild).
+   *
+   * Returns ONLY agents that successfully connected — failed-login
+   * agents have no fabricUserId yet and are excluded.
+   */
+  getPresenceAccounts(): Array<{
+    agentId: string;
+    fabricUserId: string;
+    guildBaseUrl: string;
+    guildNodeId: string;
+    fabricApiKey: string;
+  }> {
+    const out: Array<{
+      agentId: string;
+      fabricUserId: string;
+      guildBaseUrl: string;
+      guildNodeId: string;
+      fabricApiKey: string;
+    }> = [];
+    for (const entry of this.identity.list()) {
+      if (!entry.fabricUserId) continue;
+      const presenceGuild = this.firstGuildByAgent.get(entry.agentId);
+      if (!presenceGuild) continue;
+      out.push({
+        agentId: entry.agentId,
+        fabricUserId: entry.fabricUserId,
+        guildBaseUrl: presenceGuild.endpoint,
+        guildNodeId: presenceGuild.nodeId,
+        fabricApiKey: entry.fabricApiKey,
+      });
+    }
+    return out;
+  }
+
+  // Filled by connectAgent for each (agent, guild). Tracks ONLY the first
+  // guild per agent (used as the presence-push target). Stores both
+  // endpoint and nodeId — presence-sync needs both: endpoint to build
+  // the URL, nodeId to pick the matching guildAccessToken from a fresh
+  // agent-login response.
+  private firstGuildByAgent = new Map<string, { endpoint: string; nodeId: string }>();
+
   private async connectAgent(agentId: string, session: FabricSession): Promise<void> {
     const selfUserId = session.user.id;
+    // First-guild capture for presence-sync push target. session.guilds is
+    // already in priority order from Center; we take the first one with a
+    // valid endpoint and stop. Multi-guild presence is a future concern.
+    if (!this.firstGuildByAgent.has(agentId)) {
+      const firstGuild = session.guilds.find((g) => typeof g.endpoint === 'string' && g.endpoint.length > 0);
+      if (firstGuild) this.firstGuildByAgent.set(agentId, { endpoint: firstGuild.endpoint, nodeId: firstGuild.nodeId });
+    }
     for (const g of session.guilds) {
       const tok = session.guildAccessTokens.find((t) => t.guildNodeId === g.nodeId)?.token;
       if (!tok) continue;
+      // Use the *callback* form of `auth` so socket.io re-evaluates the JWT
+      // on every (re)connect. The single-shot `auth: { token: tok }` shape
+      // captured the token in closure: after socket.io's silent auto-reconnect
+      // the backend got the same JWT that expired ~15 min into the session
+      // (guildAccessToken TTL = 900s) and silently rejected the handshake at
+      // the application layer. The client's `connect` event still fired (TCP
+      // succeeded), so the plugin happily ran the channel-resync, emitted
+      // `join_channel` into the void, and logged "joined N channel(s)" while
+      // the backend was actually broadcasting message.created to a room with
+      // zero subscribers. End user symptom: DMs to agents silently dropped.
       const socket = io(`${g.endpoint}/realtime`, {
         transports: ['websocket'],
-        auth: { token: tok },
+        auth: (cb) => {
+          // Best-effort fresh token; on transient failure fall back to the
+          // last known good one. tokenCache also keeps HTTP calls (attachment
+          // download / reply post) from 401'ing in the same window.
+          this.freshGuildToken(agentId, g.nodeId, session)
+            .then((fresh) => cb({ token: fresh ?? tok }))
+            .catch(() => cb({ token: tok }));
+        },
         autoConnect: false,
       });
-      const joinAll = async () => {
+      // Tracked socket.io rooms for this (agent, guild). The initial fetch
+      // on `connect` seeds it; the periodic resync diffs against it so we
+      // only emit `join_channel` for genuinely new channels (and
+      // `leave_channel` for ones the agent is no longer in).
+      const joined = new Set<string>();
+      const syncChannels = async (kind: 'initial' | 'resync') => {
+        let freshTok: string | undefined;
+        try {
+          freshTok = await this.freshGuildToken(agentId, g.nodeId, session);
+        } catch {
+          freshTok = tok;
+        }
+        const authTok = freshTok ?? tok;
         try {
           const res = await fetch(
             `${g.endpoint}/api/channels?guildId=${encodeURIComponent(g.nodeId)}`,
-            { headers: { authorization: `Bearer ${tok}` } },
+            { headers: { authorization: `Bearer ${authTok}` } },
           );
-          const channels = res.ok ? ((await res.json()) as Array<{ id: string }>) : [];
-          for (const c of channels) socket.emit('join_channel', { channelId: c.id });
-          this.log.info(`fabric: agent ${agentId} joined ${channels.length} channel(s) on ${g.nodeId}`);
+          if (!res.ok) return;
+          const channels = (await res.json()) as Array<{ id: string }>;
+          const current = new Set(channels.map((c) => c.id));
+          let added = 0;
+          let removed = 0;
+          for (const id of current) {
+            if (!joined.has(id)) {
+              socket.emit('join_channel', { channelId: id });
+              joined.add(id);
+              added++;
+            }
+          }
+          for (const id of [...joined]) {
+            if (!current.has(id)) {
+              socket.emit('leave_channel', { channelId: id });
+              joined.delete(id);
+              removed++;
+            }
+          }
+          if (kind === 'initial') {
+            this.log.info(
+              `fabric: agent ${agentId} joined ${current.size} channel(s) on ${g.nodeId}`,
+            );
+          } else if (added > 0 || removed > 0) {
+            this.log.info(
+              `fabric: agent ${agentId} channel resync on ${g.nodeId}: +${added} -${removed} (now ${joined.size})`,
+            );
+          }
         } catch {
-          /* best effort */
+          /* best effort — next tick will retry */
         }
       };
-      socket.on('connect', () => void joinAll());
+      socket.on('connect', () => {
+        // On every (re)connect the server forgets prior subscriptions, so
+        // reset our local view and seed from a fresh fetch.
+        joined.clear();
+        void syncChannels('initial');
+      });
+      // Push-based membership events from the backend (companion to
+      // Fabric.Backend.Guild's RealtimeGateway.emitToUser). When the
+      // server tells us this user was added to / removed from a
+      // channel, we sub/unsub the socket.io room immediately — no
+      // 60s wait for the polling resync. Polling remains as a safety
+      // net for missed events.
+      socket.on('channel.joined', (evt: { channelId?: string }) => {
+        const id = evt?.channelId;
+        if (!id || joined.has(id)) return;
+        socket.emit('join_channel', { channelId: id });
+        joined.add(id);
+        this.log.info(`fabric: agent ${agentId} channel.joined push on ${g.nodeId}: ${id} (now ${joined.size})`);
+      });
+      socket.on('channel.left', (evt: { channelId?: string }) => {
+        const id = evt?.channelId;
+        if (!id || !joined.has(id)) return;
+        socket.emit('leave_channel', { channelId: id });
+        joined.delete(id);
+        this.log.info(`fabric: agent ${agentId} channel.left push on ${g.nodeId}: ${id} (now ${joined.size})`);
+      });
+      const syncTimer = setInterval(
+        () => void syncChannels('resync'),
+        FabricInbound.CHANNEL_SYNC_INTERVAL_MS,
+      );
+      this.channelSyncTimers.push(syncTimer);
+      const agentTimers = this.timersByAgent.get(agentId) ?? [];
+      agentTimers.push(syncTimer);
+      this.timersByAgent.set(agentId, agentTimers);
       socket.on('message.created', (m: FabricMessage) => {
         const channelId = m.channelId ?? '';
         if (!channelId) return;
+        // Record xType into the channel-meta cache before self-author
+        // / dedup gates — channel type doesn't depend on who sent the
+        // message, and recording it on observer-only triage messages
+        // is still useful (the next consumer asking
+        // __fabric.getChannelType wants the answer regardless of
+        // whether THIS message was delivered to an agent).
+        recordChannelType(channelId, m.xType);
         if (m.authorUserId && m.authorUserId === selfUserId) return;
         const key = `${agentId}:${m.messageId}`;
         if (this.seen.has(key)) return;
         this.seen.add(key);
         if (this.seen.size > 5000) this.seen.clear();
-        void this.dispatch(agentId, g, channelId, m, session);
+        // Per-channel serial queue. Prevents concurrent model turns for
+        // the same channel — important for triage where a second wake
+        // arriving mid-reply would interleave with the in-flight one.
+        this.enqueueChannelTask(channelId, async () => {
+          // Triage on_call gate: if the on-duty agent isn't currently
+          // on_call per HF, don't dispatch yet — just sit on the
+          // per-channel queue. Subsequent triage messages will recheck;
+          // when the agent becomes on_call, the next arrival drains.
+          //
+          // Also handles: triage + wake=true must verify status before
+          // committing to a model turn. Non-triage and triage observer
+          // (wake=false) skip the gate.
+          if (m.xType === 'triage' && m.wakeup === true) {
+            const onCall = await this.checkAgentOnCall(agentId);
+            if (!onCall) {
+              this.log.info(
+                `fabric: triage wake gated (agent=${agentId} not on_call) — re-queue msg=${m.messageId}`,
+              );
+              this.pendingTriageGated.push({ agentId, g, channelId, m, session });
+              return;
+            }
+            // Drain any previously-gated messages (FIFO) before this one,
+            // now that we know the agent is on_call.
+            await this.drainGatedFor(agentId);
+          }
+          await this.dispatch(agentId, g, channelId, m, session);
+        });
       });
       socket.connect();
       this.sockets.push(socket);
+      // Track per-agent so addAccount/removeAccount can teardown
+      // independently without disturbing other agents.
+      const agentSockets = this.socketsByAgent.get(agentId) ?? [];
+      agentSockets.push(socket);
+      this.socketsByAgent.set(agentId, agentSockets);
     }
   }
 
@@ -214,11 +636,31 @@ export class FabricInbound {
     const core = this.core as Core & Record<string, unknown>;
     const cfg = this.cfg as { session?: { store?: unknown } };
     try {
+      // Route by xType. DM channels need peer.kind='direct' so openclaw
+      // treats them as 1:1 (sessionKey 'agent:<id>:fabric:direct:<chan>'
+      // and ctx.ChatType='direct') rather than as a multi-party group.
+      // Without this, the agent's user-prompt metadata says
+      // 'is_group_chat: true' on a DM and downstream prompt logic
+      // (commands-handlers `isDirectMessage` checks ChatType==='direct')
+      // misclassifies the turn.
+      const { peerKind, chatType } = fabricPeerRoutingForXType(m.xType);
+      // resolveAgentRoute needs the *binding* accountId (the channel-side
+      // slot name) — not the openclaw agentId. For most agents the binding
+      // is `{agentId: X, match: {channel: fabric, accountId: X}}` so the
+      // two coincide; but for shared-placeholder cases (e.g. the recruitment
+      // `interviewee` slot bound to multiple agents over its lifetime) the
+      // binding accountId is the slot label ("interviewee", "Neon", …) not
+      // the agent_id. Passing agentId there returned bindings=0 and silently
+      // fell back to `main`, hijacking sub-discussion turns. Look up the
+      // agent's fabric binding accountId here; fall back to agentId when no
+      // explicit binding exists (preserves prior behavior for agents with
+      // no fabric binding declared).
+      const bindingAccountId = findFabricBindingAccountId(this.cfg, agentId) ?? agentId;
       const route = core.channel.routing.resolveAgentRoute({
         cfg: this.cfg,
         channel: 'fabric',
-        accountId: agentId,
-        peer: { kind: 'group', id: channelId },
+        accountId: bindingAccountId,
+        peer: { kind: peerKind, id: channelId },
       });
       const storePath = core.channel.session.resolveStorePath(cfg.session?.store, {
         agentId: route.agentId,
@@ -233,7 +675,7 @@ export class FabricInbound {
         To: `fabric:${channelId}`,
         SessionKey: route.sessionKey,
         AccountId: route.accountId ?? agentId,
-        ChatType: 'group',
+        ChatType: chatType,
         ConversationLabel: `fabric:${guild.nodeId}`,
         SenderId: m.authorUserId ?? 'fabric',
         Provider: 'fabric',
@@ -250,7 +692,24 @@ export class FabricInbound {
       // the woken speaker emits a normal message or /no-reply). We still
       // record the message into the agent's session so it has the full
       // channel conversation as context whenever it IS later woken.
-      if (m.wakeup !== true) {
+      //
+      // Exception: dm channels are 1:1 — there is no turn/wakeup gating;
+      // any message that isn't the agent's own (already filtered above) is
+      // always delivered to the model.
+      if (m.xType !== 'dm' && m.wakeup !== true) {
+        // Triage exception: non-wake messages (admin observer) MUST NOT
+        // enter the agent's session at all. The next time the agent
+        // wakes for a triage message, their context should contain only
+        // their own past wakeups + their own outgoing messages — never
+        // the observer-only chatter from other agents. For non-triage
+        // channels keep the legacy "record-as-history" so a later wake
+        // sees the full channel conversation.
+        if (m.xType === 'triage') {
+          this.log.info(
+            `fabric: triage observer skip agent=${agentId} channel=${channelId} msg=${m.messageId}`,
+          );
+          return;
+        }
         const ctxPayload = core.channel.reply.finalizeInboundContext(baseCtx);
         await core.channel.session.recordInboundSession({
           storePath,
@@ -301,8 +760,17 @@ export class FabricInbound {
           const text = (payload?.text ?? '').trim();
           this.log.info(`fabric: deliver agent=${agentId} channel=${channelId} len=${text.length}`);
           if (!text || !gt) return;
-          await this.client.postMessage(guild.endpoint, gt, channelId, text, session.user.id);
-          this.log.info(`fabric: posted reply agent=${agentId} channel=${channelId}`);
+          // Buffer segments; the merged message is posted right after
+          // dispatch returns (the deterministic turn boundary, see the
+          // finally below). Disable per channel: channels.fabric.coalesce.
+          await enqueueDelivery({
+            channelId,
+            text,
+            coalesce: resolveCoalesce(this.cfg as never),
+            post: (t) =>
+              this.client.postMessage(guild.endpoint, gt, channelId, t, session.user.id) as Promise<void>,
+            log: (m) => this.log.info(m),
+          });
         },
         onRecordError: (err: unknown) =>
           this.log.warn(`fabric: session record failed agent=${agentId}: ${String(err)}`),
@@ -327,6 +795,12 @@ export class FabricInbound {
       this.log.info(`fabric: dispatch returned agent=${agentId} channel=${channelId}`);
     } catch (err) {
       this.log.warn(`fabric: dispatch failed agent=${agentId} channel=${channelId}: ${String(err)}`);
+    } finally {
+      // Deterministic per-turn boundary: dispatchInboundReplyWithBase only
+      // resolves AFTER every deliver() call of this turn has run, so the
+      // buffer now holds all segments — flush them as ONE Fabric message.
+      // No hooks, no timers, no idle guessing.
+      await flushFabricForChannel(channelId);
     }
   }
 }

src/presence-sync.ts

@@ -0,0 +1,177 @@
+/**
+ * presence-sync — read each connected agent's HF status (via the
+ * cross-plugin `globalThis.__hfAgentStatus.get(agentId)` exposed by
+ * HarborForge.OpenclawPlugin) and push diffs to Fabric.Backend.Guild
+ * `PUT /api/agents/:userId/presence` so the backend can apply
+ * busy-discard on `announce`-type channel deliveries.
+ *
+ * Push model: we only PUT when an agent's status actually changes
+ * (since the last push). The HF-side accessor has its own TTL cache
+ * to absorb the every-30s polling.
+ *
+ * Auth: the endpoint sits behind ApiKeyGuard (global APP_GUARD per
+ * app.module.js) which expects `Authorization: Bearer <guild-token>`
+ * — NOT the agent's fabricApiKey directly. So before each PUT we do
+ * a fresh agent-login (or reuse a cached token if still within its
+ * 15-min JWT TTL) and pull the guildAccessToken matching the target
+ * guild. Status changes are rare enough that login overhead is fine.
+ *
+ * If HF plugin isn't loaded (`__hfAgentStatus` undefined), the loop
+ * is a no-op — Fabric backend defaults presence to 'unknown' which is
+ * treated as not-busy. Announce-channel delivery still works; busy
+ * filtering simply doesn't kick in.
+ */
+import type { FabricClient } from './fabric-client.js';
+
+type HfStatus = 'idle' | 'on_call' | 'busy' | 'exhausted' | 'offline';
+type Bridge = { get(agentId: string): Promise<HfStatus | undefined> };
+type Logger = { info: (m: string) => void; warn: (m: string) => void };
+
+export interface PresenceSyncAccount {
+  agentId: string;
+  fabricUserId: string; // the agent's Fabric Center user id (UUID)
+  guildBaseUrl: string; // e.g. https://fabric.hangman-lab.top/guild/<id>
+  guildNodeId: string;  // which guildAccessTokens[].guildNodeId to pick
+  fabricApiKey: string; // existing per-account key (used for agent-login)
+}
+
+// Guild access JWTs expire every 900s. Refresh ~2 min early to stay
+// safely inside the window even if a tick runs late.
+const TOKEN_TTL_MS = (15 - 2) * 60 * 1000;
+
+interface CachedToken {
+  token: string;
+  expiresAt: number; // epoch ms
+}
+
+export class PresenceSync {
+  private timer: ReturnType<typeof setInterval> | null = null;
+  private readonly lastStatus = new Map<string, HfStatus>(); // by agentId
+  private readonly accounts = new Map<string, PresenceSyncAccount>();
+  private readonly tokenCache = new Map<string, CachedToken>(); // by agentId
+
+  // Mutex flag: a tick iterates accounts serially with `await` on each
+  // agent-login + PUT round-trip, so a single tick can easily run 20+s
+  // when there are many accounts. setInterval(intervalMs) does NOT wait
+  // for the previous tick to finish — without this guard the next tick
+  // fires on top of a still-running one and two parallel iterations
+  // PUT the same agentId within milliseconds. That tipped the backend's
+  // first-time-insert race (separate fix in Fabric.Backend.Guild) into
+  // 500s on prod. Guarded ticks just skip a beat instead.
+  private inflight = false;
+
+  constructor(private readonly logger: Logger, private readonly client: FabricClient) {}
+
+  setAccounts(accounts: PresenceSyncAccount[]): void {
+    this.accounts.clear();
+    for (const a of accounts) this.accounts.set(a.agentId, a);
+  }
+
+  start(intervalMs = 30_000): void {
+    if (this.timer) return;
+    this.timer = setInterval(() => {
+      this.tick().catch((err) => this.logger.warn(`fabric: presence-sync error: ${String(err)}`));
+    }, intervalMs);
+    // run once immediately so initial state lands fast
+    void this.tick();
+  }
+
+  stop(): void {
+    if (this.timer) {
+      clearInterval(this.timer);
+      this.timer = null;
+    }
+  }
+
+  /**
+   * Fetch a fresh guildAccessToken for `acct`, caching it under the
+   * agentId until just before its JWT expiry. Returns null on login
+   * failure or if the session has no matching guild — caller logs +
+   * skips the PUT.
+   */
+  private async ensureGuildToken(acct: PresenceSyncAccount): Promise<string | null> {
+    const now = Date.now();
+    const cached = this.tokenCache.get(acct.agentId);
+    if (cached && cached.expiresAt > now) return cached.token;
+
+    let session;
+    try {
+      session = await this.client.agentLogin(acct.fabricApiKey);
+    } catch (err) {
+      this.logger.warn(`fabric: presence-sync agent-login failed for ${acct.agentId}: ${String(err)}`);
+      return null;
+    }
+    const entry = session.guildAccessTokens.find((g) => g.guildNodeId === acct.guildNodeId);
+    if (!entry?.token) {
+      this.logger.warn(
+        `fabric: presence-sync no guild token for ${acct.agentId} guild=${acct.guildNodeId}`,
+      );
+      return null;
+    }
+    this.tokenCache.set(acct.agentId, { token: entry.token, expiresAt: now + TOKEN_TTL_MS });
+    return entry.token;
+  }
+
+  private async tick(): Promise<void> {
+    // Mutex: see the `inflight` field declaration for the why. Drop
+    // overlapping ticks rather than letting them run concurrently —
+    // status is gated by `lastStatus !== bridge.get`, so skipping a
+    // beat costs nothing the next beat won't catch.
+    if (this.inflight) return;
+    this.inflight = true;
+    try {
+      await this.tickInner();
+    } finally {
+      this.inflight = false;
+    }
+  }
+
+  private async tickInner(): Promise<void> {
+    const bridge = (globalThis as Record<string, unknown>)['__hfAgentStatus'] as Bridge | undefined;
+    if (!bridge || typeof bridge.get !== 'function') return; // HF plugin not loaded — skip
+
+    for (const [agentId, acct] of this.accounts) {
+      let status: HfStatus | undefined;
+      try {
+        status = await bridge.get(agentId);
+      } catch {
+        continue;
+      }
+      if (!status) continue;
+      if (this.lastStatus.get(agentId) === status) continue; // no change → no PUT
+
+      const guildToken = await this.ensureGuildToken(acct);
+      if (!guildToken) continue;
+
+      try {
+        // Endpoint: PUT /api/agents/:userId/presence. ApiKeyGuard (global
+        // APP_GUARD) requires `Authorization: Bearer <guildAccessToken>`
+        // — NOT the agent's raw fabricApiKey. Pre-v1: this loop sent
+        // x-api-key and got 401 "missing bearer token" forever. The /api
+        // prefix is required because the guild backend sets a global
+        // 'api' prefix in main.ts setGlobalPrefix('api').
+        const url = `${acct.guildBaseUrl.replace(/\/$/, '')}/api/agents/${encodeURIComponent(acct.fabricUserId)}/presence`;
+        const res = await fetch(url, {
+          method: 'PUT',
+          headers: {
+            'content-type': 'application/json',
+            authorization: `Bearer ${guildToken}`,
+          },
+          body: JSON.stringify({ status, source: 'hf-plugin' }),
+        });
+        if (res.ok) {
+          this.lastStatus.set(agentId, status);
+          this.logger.info(`fabric: presence-sync ${agentId} → ${status}`);
+        } else {
+          // 401 here usually means the cached token went stale unexpectedly
+          // (server-side rotation or clock skew) — drop the cache so the
+          // next tick re-logs-in.
+          if (res.status === 401) this.tokenCache.delete(agentId);
+          this.logger.warn(`fabric: presence-sync PUT ${agentId} failed: ${res.status}`);
+        }
+      } catch (err) {
+        this.logger.warn(`fabric: presence-sync PUT ${agentId} threw: ${String(err)}`);
+      }
+    }
+  }
+}

src/sub-discussion-hook.ts

@@ -0,0 +1,76 @@
+import type { IdentityRegistry } from './identity.js';
+import type { SubDiscussionStore } from './sub-discussion-store.js';
+
+// Plugin-local before_prompt_build hook that injects per-(agent, channel)
+// guides for sub-discussion channels created via the `create-sub-discussion`
+// tool. Mirrors the pattern used by ClawPrompts' fabric-chat-injector
+// (channelId-aware injection) but with content dynamically supplied at
+// channel-creation time instead of read from static files via PrismFacet's
+// router/rule registry.
+//
+// Match logic per turn:
+//   ctx.channelId  → store.find()  → sub-discussion entry
+//   ctx.agentId    → identity.findByAgentId().fabricUserId
+//                    ─ matches entry.hostUserId   → inject hostGuide
+//                    ─ matches entry.guestUserIds → inject guestGuide
+//                    ─ neither                    → no injection
+//
+// Fail-closed on unknown agentId/channelId — we never inject "the wrong"
+// guide, only the right one or nothing.
+
+const _G = globalThis as Record<string, unknown>;
+const DEDUP_KEY = '_fabricSubDiscussionHookDedup';
+
+type PromptCtx = {
+  agentId?: string;
+  channelId?: string;
+  messageProvider?: string;
+};
+
+export function registerSubDiscussionHook(
+  api: {
+    on: (hook: string, handler: (...args: unknown[]) => unknown) => void;
+    logger: { info: (m: string) => void; warn: (m: string) => void };
+  },
+  store: SubDiscussionStore,
+  identity: IdentityRegistry,
+): void {
+  if (!(_G[DEDUP_KEY] instanceof WeakSet)) _G[DEDUP_KEY] = new WeakSet<object>();
+  const dedup = _G[DEDUP_KEY] as WeakSet<object>;
+
+  api.on('before_prompt_build', async (...args: unknown[]) => {
+    const event = args[0];
+    const ctx = (args[1] ?? {}) as PromptCtx;
+    // The hook fires both for fabric-driven turns (channelId set) and
+    // for other triggers (HF wake, exec-event, etc.) — drop those.
+    if (typeof event === 'object' && event !== null) {
+      if (dedup.has(event)) return undefined;
+      dedup.add(event);
+    }
+    const agentId = (ctx.agentId ?? '').trim();
+    const channelId = (ctx.channelId ?? '').trim();
+    if (!agentId || !channelId) return undefined;
+    const provider = (ctx.messageProvider ?? '').toLowerCase();
+    if (provider && provider !== 'fabric') return undefined;
+
+    const entry = store.find(channelId);
+    if (!entry) return undefined;
+
+    const ident = identity.findByAgentId(agentId);
+    const myUserId = (ident?.fabricUserId ?? '').trim();
+    if (!myUserId) {
+      // identity registry caches fabricUserId after the first agentLogin
+      // in inbound.ts. If it's missing here, the agent likely hasn't
+      // completed login yet — skip rather than guess.
+      return undefined;
+    }
+
+    if (myUserId === entry.hostUserId) {
+      return { appendSystemContext: entry.hostGuide };
+    }
+    if (entry.guestUserIds.includes(myUserId)) {
+      return { appendSystemContext: entry.guestGuide };
+    }
+    return undefined;
+  });
+}

src/sub-discussion-store.ts

@@ -0,0 +1,76 @@
+import { readFileSync, writeFileSync, existsSync, mkdirSync } from 'node:fs';
+import { dirname } from 'node:path';
+
+// Records per-(sub-discussion-channel) state created by the
+// `create-sub-discussion` tool and consumed by:
+//   1. `before_prompt_build` hook — looks up by (agentId, channelId) to
+//      inject the host or guest guide as appendSystemContext.
+//   2. `close-sub-discussion` tool — looks up by sub channelId to find
+//      the parent channel to post the callback to and to validate that
+//      the caller is the original host.
+//
+// One sub-discussion = one channel. Lifetime: from create-sub-discussion
+// return until close-sub-discussion (or gateway-stop / disk corruption).
+// We persist to a small JSON file so a gateway restart mid-interview
+// doesn't strand both host and guest with no guide.
+
+export type SubDiscussionEntry = {
+  subChannelId: string;
+  hostAgentId: string;         // the openclaw agentId that called create-sub-discussion
+  hostUserId: string;          // the host's Fabric Center userId (session.user.id)
+  guestUserIds: string[];      // Fabric Center userIds invited as guests
+  hostGuide: string;           // appended to host's session system prompt while in this channel
+  guestGuide: string;          // appended to each guest's session system prompt while in this channel
+  callbackGuildNodeId: string; // where to post the callback when close is called
+  callbackChannelId: string;   // parent channel to post callback to (system msg)
+  createdAt: string;           // ISO timestamp
+};
+
+type StoreFile = { entries: SubDiscussionEntry[] };
+
+export class SubDiscussionStore {
+  private byChannelId = new Map<string, SubDiscussionEntry>();
+
+  constructor(private readonly filePath: string) {
+    this.load();
+  }
+
+  private load(): void {
+    if (!existsSync(this.filePath)) return;
+    try {
+      const data = JSON.parse(readFileSync(this.filePath, 'utf8')) as StoreFile;
+      for (const e of data.entries ?? []) {
+        if (e?.subChannelId) this.byChannelId.set(e.subChannelId, e);
+      }
+    } catch {
+      // Corrupt file → start empty; first mutation rewrites cleanly.
+    }
+  }
+
+  private persist(): void {
+    mkdirSync(dirname(this.filePath), { recursive: true });
+    const data: StoreFile = { entries: [...this.byChannelId.values()] };
+    writeFileSync(this.filePath, JSON.stringify(data, null, 2));
+  }
+
+  list(): SubDiscussionEntry[] {
+    return [...this.byChannelId.values()];
+  }
+
+  find(subChannelId: string): SubDiscussionEntry | undefined {
+    return this.byChannelId.get(subChannelId);
+  }
+
+  add(entry: SubDiscussionEntry): void {
+    this.byChannelId.set(entry.subChannelId, entry);
+    this.persist();
+  }
+
+  remove(subChannelId: string): SubDiscussionEntry | undefined {
+    const e = this.byChannelId.get(subChannelId);
+    if (!e) return undefined;
+    this.byChannelId.delete(subChannelId);
+    this.persist();
+    return e;
+  }
+}

src/tools.ts

@@ -1,5 +1,7 @@
 import type { FabricClient } from './fabric-client.js';
 import type { IdentityRegistry } from './identity.js';
+import type { SubDiscussionStore } from './sub-discussion-store.js';
+import { resolveCommandsSyncKey } from './accounts.js';
 
 // OpenClaw tool registration api (loose typing — concrete shape from
 // openclaw/plugin-sdk/core at host SDK version).
@@ -10,6 +12,9 @@ type ToolApi = {
 
 type Ctx = { agentId?: string };
 
+// Loose config shape — just the fields we read here.
+type ToolsCfg = { channels?: { fabric?: { commandsSyncKey?: string } }; [k: string]: unknown };
+
 const X_BY_KIND: Record<string, string> = {
   chat: 'general',
   work: 'work',
@@ -17,19 +22,34 @@ const X_BY_KIND: Record<string, string> = {
   discussion: 'discuss',
 };
 
+// Delay between create-sub-discussion's channel create and greeting post.
+// Backend pushes channel.joined to invitee sockets on create; that push
+// has to traverse socket.io rooms before the guest plugin can sub the
+// channel:<id> room. If the greeting is posted before that, the guest's
+// turn-activation wakeup misses (the socket isn't in the room yet).
+// 500ms is empirically slack enough on local sim + production t3, and
+// short enough not to feel laggy from the host's tool-result POV. Bump
+// via FABRIC_SUB_DISCUSSION_GREETING_DELAY_MS env if needed.
+const GREETING_DELAY_MS = (() => {
+  const v = Number.parseInt(process.env.FABRIC_SUB_DISCUSSION_GREETING_DELAY_MS ?? '', 10);
+  return Number.isFinite(v) && v >= 0 ? v : 500;
+})();
+
 export function registerFabricTools(
   api: ToolApi,
   client: FabricClient,
   identity: IdentityRegistry,
+  store: SubDiscussionStore,
+  cfg: ToolsCfg,
 ): void {
   // Resolve the calling agent's Fabric session + a guild's token/endpoint.
   const ctxGuild = async (agentId: string, guildNodeId: string) => {
     const entry = identity.findByAgentId(agentId);
     if (!entry)
       throw new Error(
-        `agent ${agentId} not registered — run: AGENT_ID=${agentId} ` +
-          `~/.openclaw/bin/fabric-register --api-key <fak_…> (or set ` +
-          `channels.fabric.accounts.${agentId}); then restart the gateway`,
+        `agent ${agentId} not registered — call the openclaw \`fabric-register\` ` +
+          `tool (apiKey: <fak_…>, agentId: ${agentId}); the dynamic-subscription ` +
+          `path brings the socket up immediately, no gateway restart needed`,
       );
     const session = await client.agentLogin(entry.fabricApiKey);
     const guild = session.guilds.find((g) => g.nodeId === guildNodeId);
@@ -38,15 +58,82 @@ export function registerFabricTools(
     return { session, guild, token };
   };
 
-  // NOTE: binding an agent's Fabric API key is intentionally NOT a tool.
-  // It's a one-time step done out-of-band via the installed script
-  //   ~/.openclaw/bin/fabric-register --api-key <fak_…>   (AGENT_ID or --agent-id)
-  // or via static config (channels.fabric.accounts.<agentId>).
+  // Bind an agent's Fabric API key — validates the key against Center,
+  // upserts ~/.openclaw/fabric-identity.json, AND brings up the inbound
+  // socket immediately via the live FabricInbound instance (no gateway
+  // restart). The standalone binary `~/.openclaw/bin/fabric-register`
+  // still exists for one-time bootstrap before the gateway runs, but
+  // recruitment's `register-agent` script should prefer this tool path
+  // so the new agent's socket is live before `interviewer` fires.
+  api.registerTool((ctx: Ctx) => ({
+    name: 'fabric-register',
+    description:
+      'Bind an agent to a Fabric Center API key. Validates the key, writes ' +
+      'the entry to ~/.openclaw/fabric-identity.json, and starts a live ' +
+      'inbound socket immediately so the agent receives Fabric pushes ' +
+      'without a gateway restart. Caller defaults to the current agent; ' +
+      'pass `agentId` to bind on behalf of another agent (recruitment use).',
+    parameters: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['apiKey'],
+      properties: {
+        apiKey: { type: 'string', description: 'Fabric Center API key (`fak_…`)' },
+        agentId: {
+          type: 'string',
+          description:
+            'Agent to register. Defaults to the calling agent (ctx.agentId). ' +
+            'Recruitment onboarding may override this when wiring a freshly ' +
+            'created agent before that agent has a session of its own.',
+        },
+      },
+    },
+    execute: async (_id: string, p: { apiKey: string; agentId?: string }) => {
+      const agentId = p.agentId ?? ctx.agentId;
+      if (!agentId) return { ok: false, error: 'no agent context (pass agentId)' };
+      if (!p.apiKey || typeof p.apiKey !== 'string') {
+        return { ok: false, error: 'apiKey required' };
+      }
+      // Delegate to FabricInbound.addAccount via the cross-plugin bridge.
+      // The bridge is installed in index.ts when inbound spins up; if it's
+      // not present yet, the gateway is still starting and the caller should
+      // retry (rare path — only hit during the gateway_start window).
+      const fabricApi = (globalThis as Record<string, unknown>)['__fabric'] as
+        | { addAccount?: (entry: { agentId: string; fabricApiKey: string }) => Promise<void> }
+        | undefined;
+      if (!fabricApi?.addAccount) {
+        return {
+          ok: false,
+          error:
+            'fabric inbound not ready (gateway still starting?). Fall back to ' +
+            '~/.openclaw/bin/fabric-register or retry after a few seconds.',
+        };
+      }
+      try {
+        await fabricApi.addAccount({ agentId, fabricApiKey: p.apiKey });
+      } catch (err) {
+        return {
+          ok: false,
+          error: `fabric-register failed: ${String(err)}`,
+        };
+      }
+      const entry = identity.findByAgentId(agentId);
+      return {
+        ok: true,
+        agentId,
+        fabricUserId: entry?.fabricUserId,
+        displayName: entry?.displayName,
+      };
+    },
+  }));
 
   const makeCreate = (kind: 'chat' | 'work' | 'report' | 'discussion') =>
     api.registerTool((ctx: Ctx) => ({
       name: `create-${kind}-channel`,
-      description: `Create a Fabric ${kind} channel (x_type=${X_BY_KIND[kind]}).`,
+      description:
+        `Create a Fabric ${kind} channel (x_type=${X_BY_KIND[kind]}). ` +
+        'Optionally pass `purpose` to describe what this channel is for — ' +
+        'agents browse channels by purpose via fabric-channel-list.',
       parameters: {
         type: 'object',
         additionalProperties: false,
@@ -58,13 +145,22 @@ export function registerFabricTools(
           memberUserIds: { type: 'array', items: { type: 'string' } },
           onDuty: { type: 'string', description: 'required for triage-like flows (unused for these kinds)' },
           listeners: { type: 'array', items: { type: 'string' } },
+          purpose: {
+            type: 'string',
+            description:
+              "Free-form description of what this channel is for. Optional but " +
+              'strongly recommended so other agents can find this channel by ' +
+              'intent (via fabric-channel-list). Can be edited later with ' +
+              'fabric-channel-set-purpose.',
           },
         },
-      execute: async (p: {
+      },
+      execute: async (_id: string, p: {
         guildNodeId: string;
         name: string;
         isPublic?: boolean;
         memberUserIds?: string[];
+        purpose?: string;
       }) => {
         const agentId = ctx.agentId;
         if (!agentId) return { ok: false, error: 'no agent context' };
@@ -75,6 +171,7 @@ export function registerFabricTools(
           xType: X_BY_KIND[kind],
           isPublic: p.isPublic ?? false,
           memberUserIds: p.memberUserIds ?? [],
+          ...(p.purpose !== undefined ? { purpose: p.purpose } : {}),
         });
         return { ok: true, channelId: ch.id };
       },
@@ -101,7 +198,7 @@ export function registerFabricTools(
         callbackChannelId: { type: 'string', description: 'optional channel to also post the summary to' },
       },
     },
-    execute: async (p: {
+    execute: async (_id: string, p: {
       guildNodeId: string;
       channelId: string;
       summary: string;
@@ -121,6 +218,245 @@ export function registerFabricTools(
     },
   }));
 
+  // ───────────────────────────────────────────────────────────────────
+  // create-sub-discussion: open a discuss-type sub-channel hanging off
+  // the caller's current channel. Designed for host-driven multi-turn
+  // exchanges (interview, brainstorm, narrow Q&A) where the guests are
+  // either fresh agents without workflow capability (recruitment
+  // interviewee) or peers that just need a short scoped chat without
+  // entering their own subflow.
+  //
+  // What it does on top of plain create-discussion-channel:
+  //   1. Persists a store entry indexed by the new sub channelId, carrying:
+  //        host agent + userId, guest userIds, host/guest guide texts,
+  //        callback (parent) channel info.
+  //   2. Auto-posts `greetingMsg` using the host's own Fabric account so
+  //      turn rotation's activation rule (first author → newOrder[0],
+  //      currentSpeaker → newOrder[1], wakeup → newOrder[1]) puts the
+  //      first guest on the spot immediately — no race where host posts
+  //      before guest's socket subs the channel room (we wait
+  //      GREETING_DELAY_MS for backend's channel.joined push to land).
+  //   3. The accompanying before_prompt_build hook (sub-discussion-hook
+  //      registered from index.ts) then injects `hostGuideMsg` into the
+  //      host's session prompt and `guestGuideMsg` into each guest's
+  //      session prompt whenever a turn in this channel fires — so the
+  //      two roles see different instructions, no shared guide file.
+  // ───────────────────────────────────────────────────────────────────
+  api.registerTool((ctx: Ctx) => ({
+    name: 'create-sub-discussion',
+    description:
+      'Open a host-driven sub-discussion channel (x_type=discuss) hanging off your current channel, ' +
+      'with role-specific system-prompt guides for host and guests. Use this for interviews / scoped ' +
+      'Q&A where you stay in control of when the conversation ends. Returns the sub channelId; ' +
+      'reach it via fabric-send-message in the rotating turn order. Close with close-sub-discussion ' +
+      'to write a callback back into the parent channel.',
+    parameters: {
+      type: 'object',
+      additionalProperties: false,
+      required: [
+        'guildNodeId',
+        'currentChannelId',
+        'channelName',
+        'greetingMsg',
+        'hostGuideMsg',
+        'guestGuideMsg',
+        'guests',
+      ],
+      properties: {
+        guildNodeId: { type: 'string', description: 'Fabric guild node id (same guild for parent + sub).' },
+        currentChannelId: {
+          type: 'string',
+          description: 'Channel id you are currently in (parent). Used as the callback target on close.',
+        },
+        channelName: { type: 'string', description: 'Display name for the new sub-discussion channel.' },
+        greetingMsg: {
+          type: 'string',
+          description:
+            'First message posted by YOU (the host) in the sub channel. Triggers turn rotation so ' +
+            "the first guest's session wakes immediately with both your greeting in history and the " +
+            'guest guide injected as system prompt.',
+        },
+        hostGuideMsg: {
+          type: 'string',
+          description:
+            "Appended to YOUR session's system prompt whenever a turn fires in this sub channel. " +
+            'Use it to remind yourself of the procedure (what to ask, when to call close-sub-discussion).',
+        },
+        guestGuideMsg: {
+          type: 'string',
+          description:
+            "Appended to EACH GUEST's session system prompt for turns in this sub channel. Use it to " +
+            'orient guests with no prior workflow context (e.g. a fresh interviewee). Keep it short; ' +
+            'long guides bloat every turn.',
+        },
+        guests: {
+          type: 'array',
+          items: { type: 'string' },
+          minItems: 1,
+          description:
+            'Fabric Center userIds invited as guests. Resolve via fabric-channel-list members or the ' +
+            '<name>@<role>.hangman-lab.top email convention.',
+        },
+        purpose: { type: 'string', description: 'Optional channel.purpose for discoverability.' },
+      },
+    },
+    execute: async (
+      _id: string,
+      p: {
+        guildNodeId: string;
+        currentChannelId: string;
+        channelName: string;
+        greetingMsg: string;
+        hostGuideMsg: string;
+        guestGuideMsg: string;
+        guests: string[];
+        purpose?: string;
+      },
+    ) => {
+      const agentId = ctx.agentId;
+      if (!agentId) return { ok: false, error: 'no agent context' };
+      if (!Array.isArray(p.guests) || p.guests.length === 0) {
+        return { ok: false, error: 'guests must be a non-empty array of Fabric userIds' };
+      }
+      const { session, guild, token } = await ctxGuild(agentId, p.guildNodeId);
+      const ch = await client.createChannel(guild.endpoint, token, {
+        guildId: p.guildNodeId,
+        name: p.channelName,
+        xType: 'discuss',
+        isPublic: false,
+        memberUserIds: p.guests,
+        ...(p.purpose !== undefined ? { purpose: p.purpose } : {}),
+      });
+      store.add({
+        subChannelId: ch.id,
+        hostAgentId: agentId,
+        hostUserId: session.user.id,
+        guestUserIds: [...p.guests],
+        hostGuide: p.hostGuideMsg,
+        guestGuide: p.guestGuideMsg,
+        callbackGuildNodeId: p.guildNodeId,
+        callbackChannelId: p.currentChannelId,
+        createdAt: new Date().toISOString(),
+      });
+      // Let backend's channel.joined push reach guest sockets before our
+      // greeting fires — otherwise the wakeup emitted by turn-activation
+      // races a not-yet-subscribed socket.io room.
+      if (GREETING_DELAY_MS > 0) {
+        await new Promise((r) => setTimeout(r, GREETING_DELAY_MS));
+      }
+      try {
+        await client.postMessage(guild.endpoint, token, ch.id, p.greetingMsg, session.user.id);
+      } catch (err) {
+        api.logger.warn(
+          `fabric: create-sub-discussion greeting post failed channel=${ch.id} err=${String(err)}`,
+        );
+      }
+      return { ok: true, subChannelId: ch.id };
+    },
+  }));
+
+  // ───────────────────────────────────────────────────────────────────
+  // close-sub-discussion: post a system-authored callback into the
+  // parent channel + close the sub-discussion channel. Only the original
+  // host can call this. Uses the Guild's x-fabric-system-key path (shared
+  // secret = commandsSyncKey) so the callback lands as a guild/system
+  // author, not the host's personal account — and can wake the host on
+  // the parent channel to continue whatever workflow opened the sub.
+  // ───────────────────────────────────────────────────────────────────
+  api.registerTool((ctx: Ctx) => ({
+    name: 'close-sub-discussion',
+    description:
+      'Close a sub-discussion channel you opened (host-only) and write a callback to the parent ' +
+      'channel as a system message. Pass `wakeupHost: false` to land the callback silently in ' +
+      'history without waking yourself.',
+    parameters: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['subChannelId', 'callbackMsg'],
+      properties: {
+        subChannelId: { type: 'string', description: 'The sub-discussion channelId returned by create-sub-discussion.' },
+        callbackMsg: {
+          type: 'string',
+          description:
+            'Content to post into the parent channel as a system-authored message. Typical content: ' +
+            'the conclusion / extracted data from the sub-discussion, so the next turn on the parent ' +
+            'channel can act on it.',
+        },
+        wakeupHost: {
+          type: 'boolean',
+          description:
+            'Whether to wake YOU (the host) on the parent channel. Default true — for recruitment ' +
+            'interview flow where the next workflow step needs to run immediately. Pass false for ' +
+            'fire-and-forget logging.',
+        },
+      },
+    },
+    execute: async (
+      _id: string,
+      p: { subChannelId: string; callbackMsg: string; wakeupHost?: boolean },
+    ) => {
+      const agentId = ctx.agentId;
+      if (!agentId) return { ok: false, error: 'no agent context' };
+      const entry = store.find(p.subChannelId);
+      if (!entry) {
+        return { ok: false, error: `sub-discussion not found: ${p.subChannelId}` };
+      }
+      if (entry.hostAgentId !== agentId) {
+        return {
+          ok: false,
+          error: `only the host (${entry.hostAgentId}) may close this sub-discussion`,
+        };
+      }
+      const systemKey = resolveCommandsSyncKey(cfg);
+      if (!systemKey) {
+        return {
+          ok: false,
+          error:
+            'channels.fabric.commandsSyncKey is not configured — close-sub-discussion needs it for ' +
+            'the x-fabric-system-key callback. Configure via openclaw config.',
+        };
+      }
+      const { guild, token } = await ctxGuild(agentId, entry.callbackGuildNodeId);
+      const wakeup = p.wakeupHost !== false;
+      // 1) Post callback into parent channel via the system-key path.
+      const url = `${guild.endpoint}/api/channels/${encodeURIComponent(entry.callbackChannelId)}/messages`;
+      const res = await fetch(url, {
+        method: 'POST',
+        headers: {
+          'content-type': 'application/json',
+          'x-fabric-system-key': systemKey,
+        },
+        body: JSON.stringify({
+          content: p.callbackMsg,
+          wakeupUserId: wakeup ? entry.hostUserId : null,
+        }),
+      }).catch((err) => {
+        throw new Error(`callback POST failed: ${String(err)}`);
+      });
+      if (!res.ok) {
+        const text = await res.text().catch(() => '');
+        return {
+          ok: false,
+          error: `callback POST ${url} -> ${res.status} ${text}`,
+        };
+      }
+      // 2) Close the sub channel using the host's own bearer (the host is
+      // a member of the channel — channel.close auth is per-member).
+      try {
+        await client.closeChannel(guild.endpoint, token, entry.subChannelId);
+      } catch (err) {
+        api.logger.warn(
+          `fabric: close-sub-discussion: sub channel close failed channel=${entry.subChannelId} err=${String(err)}`,
+        );
+      }
+      // 3) Drop the store entry so the prompt-injection hook stops firing
+      // for this channel (the sub is closed; any straggler turns would
+      // just hit the closed-channel write reject downstream).
+      store.remove(entry.subChannelId);
+      return { ok: true, closed: true };
+    },
+  }));
+
   // fabric-canvas: share / update / read / close the channel's single
   // pinned canvas document (one tool, four actions). update/close are
   // sharer-only server-side (the guild returns 403 otherwise).
@@ -151,7 +487,7 @@ export function registerFabricTools(
         },
       },
     },
-    execute: async (p: {
+    execute: async (_id: string, p: {
       action: 'read' | 'share' | 'update' | 'close';
       guildNodeId: string;
       channelId: string;
@@ -216,7 +552,7 @@ export function registerFabricTools(
         channelId: { type: 'string' },
       },
     },
-    execute: async (p: {
+    execute: async (_id: string, p: {
       action: 'members' | 'join' | 'leave';
       guildNodeId: string;
       channelId: string;
@@ -243,4 +579,359 @@ export function registerFabricTools(
       }
     },
   }));
+
+  // -----------------------------------------------------------------
+  // fabric-send-message: post a message into a specific channel.
+  //
+  // Unlike a normal channel reply (which goes back to whatever channel
+  // woke the agent), this lets the agent proactively initiate text into
+  // any channel they are a member of — e.g. ARD broadcasting daily
+  // workload to #agents-room, or triage agent following up on an
+  // already-routed task by commenting in #updates.
+  // -----------------------------------------------------------------
+  api.registerTool((ctx: Ctx) => ({
+    name: 'fabric-send-message',
+    description:
+      'Send a text message into a specific Fabric channel. Author is the calling agent. ' +
+      'Requires guildNodeId + channelId + content. Returns {ok, messageId, seq}.',
+    parameters: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['guildNodeId', 'channelId', 'content'],
+      properties: {
+        guildNodeId: { type: 'string' },
+        channelId: { type: 'string' },
+        content: { type: 'string', description: 'Message body (markdown supported by the renderer).' },
+      },
+    },
+    execute: async (_id: string, p: { guildNodeId: string; channelId: string; content: string }) => {
+      const agentId = ctx.agentId;
+      if (!agentId) return { ok: false, error: 'no agent context' };
+      const { session, guild, token } = await ctxGuild(agentId, p.guildNodeId);
+      const res = (await client.postMessage(
+        guild.endpoint,
+        token,
+        p.channelId,
+        p.content,
+        session.user.id,
+      )) as { messageId?: string; seq?: number };
+      return { ok: true, messageId: res.messageId, seq: res.seq };
+    },
+  }));
+
+  // ───────────────────────────────────────────────────────────────────
+  // fabric-send-sys-msg: post a system-authored message (author =
+  // sentinel UUID 0000…, not the calling agent) using the Guild's
+  // x-fabric-system-key path. Use for cross-agent broadcasts where you
+  // don't want the message tied to one agent's identity — Dialectic
+  // topic announcements / lifecycle events, host-system advisories,
+  // etc. Caller doesn't need to be a member of the channel (the
+  // backend isSystem branch skips assertParticipant), but must be a
+  // member of the guild (their session resolves the guild endpoint).
+  //
+  // Shared secret: reads channels.fabric.commandsSyncKey (same value
+  // as the guild's FABRIC_BACKEND_GUILD_COMMANDS_SYNC_KEY env). Empty
+  // config → tool returns ok:false with a clear error, no fall-through
+  // to regular agent posting.
+  // ───────────────────────────────────────────────────────────────────
+  api.registerTool((ctx: Ctx) => ({
+    name: 'fabric-send-sys-msg',
+    description:
+      'Send a SYSTEM-AUTHORED message into a Fabric channel (author = guild sentinel, not you). ' +
+      'Use for cross-agent broadcasts that should not be attributed to a single agent — ' +
+      'Dialectic announce-channel topic broadcasts, lifecycle events, system advisories. ' +
+      'Optionally precise-wake one recipient via wakeupUserId; otherwise the message lands ' +
+      'silently in history (no wake).',
+    parameters: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['guildNodeId', 'channelId', 'content'],
+      properties: {
+        guildNodeId: { type: 'string' },
+        channelId: { type: 'string' },
+        content: { type: 'string', description: 'Message body (markdown supported by the renderer).' },
+        wakeupUserId: {
+          type: 'string',
+          description:
+            "Optional: a single Fabric userId to wake with this message (everyone else in the " +
+            'channel sees it but with wakeup=false). Omit for fully silent broadcast.',
+        },
+      },
+    },
+    execute: async (
+      _id: string,
+      p: { guildNodeId: string; channelId: string; content: string; wakeupUserId?: string },
+    ) => {
+      const agentId = ctx.agentId;
+      if (!agentId) return { ok: false, error: 'no agent context' };
+      const systemKey = resolveCommandsSyncKey(cfg);
+      if (!systemKey) {
+        return {
+          ok: false,
+          error:
+            'channels.fabric.commandsSyncKey is not configured — fabric-send-sys-msg needs it for ' +
+            'the x-fabric-system-key header. Configure via openclaw config.',
+        };
+      }
+      const { guild } = await ctxGuild(agentId, p.guildNodeId);
+      const url = `${guild.endpoint}/api/channels/${encodeURIComponent(p.channelId)}/messages`;
+      const wakeup = typeof p.wakeupUserId === 'string' && p.wakeupUserId.trim()
+        ? p.wakeupUserId.trim()
+        : null;
+      const res = await fetch(url, {
+        method: 'POST',
+        headers: {
+          'content-type': 'application/json',
+          'x-fabric-system-key': systemKey,
+        },
+        body: JSON.stringify({ content: p.content, wakeupUserId: wakeup }),
+      });
+      if (!res.ok) {
+        const text = await res.text().catch(() => '');
+        return { ok: false, error: `POST ${url} -> ${res.status} ${text}` };
+      }
+      const json = (await res.json().catch(() => null)) as
+        | { messageId?: string; seq?: number; authorUserId?: string }
+        | null;
+      return {
+        ok: true,
+        messageId: json?.messageId,
+        seq: json?.seq,
+        authorUserId: json?.authorUserId,
+      };
+    },
+  }));
+
+  // -----------------------------------------------------------------
+  // fabric-channel-list: enumerate channels the calling agent can see
+  // in a given guild. Backend filters to public channels + channels the
+  // agent is a member of. Returns id / name / xType per channel so the
+  // agent can pick a channelId for fabric-send-message etc.
+  // -----------------------------------------------------------------
+  api.registerTool((ctx: Ctx) => ({
+    name: 'fabric-channel-list',
+    description:
+      'List channels visible to the calling agent in a guild. Optional ' +
+      'nameFilter does a case-insensitive substring match client-side. ' +
+      'Use this to find a channelId before fabric-send-message / fabric-message-history.',
+    parameters: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['guildNodeId'],
+      properties: {
+        guildNodeId: { type: 'string' },
+        nameFilter: { type: 'string', description: 'optional substring match on channel name (case-insensitive)' },
+        xType: {
+          type: 'string',
+          enum: ['general', 'work', 'report', 'discuss', 'triage', 'custom', 'dm'],
+          description: 'optional filter by x_type',
+        },
+        includeClosed: { type: 'boolean', description: 'default false — closed channels filtered out' },
+      },
+    },
+    execute: async (_id: string, p: {
+      guildNodeId: string;
+      nameFilter?: string;
+      xType?: string;
+      includeClosed?: boolean;
+    }) => {
+      const agentId = ctx.agentId;
+      if (!agentId) return { ok: false, error: 'no agent context' };
+      const { guild, token } = await ctxGuild(agentId, p.guildNodeId);
+      const all = await client.listChannels(guild.endpoint, token, p.guildNodeId);
+      const needle = (p.nameFilter ?? '').toLowerCase();
+      const filtered = all.filter((c) => {
+        if (!p.includeClosed && c.closed) return false;
+        if (p.xType && c.xType !== p.xType) return false;
+        if (needle && !c.name.toLowerCase().includes(needle)) return false;
+        return true;
+      });
+      return {
+        ok: true,
+        count: filtered.length,
+        channels: filtered.map((c) => ({
+          id: c.id,
+          name: c.name,
+          xType: c.xType,
+          isPublic: c.isPublic,
+          closed: c.closed,
+          lastSeq: c.lastSeq,
+          purpose: c.purpose ?? null,
+        })),
+      };
+    },
+  }));
+
+  // -----------------------------------------------------------------
+  // fabric-guild-list: enumerate guilds the calling agent belongs to.
+  // Each row carries `purpose` — free-form description of what the
+  // guild is for (admin-set). Use this as the first step when a
+  // workflow says "find the right guild for X" — pick by purpose,
+  // then fabric-channel-list to find the right channel inside it.
+  // -----------------------------------------------------------------
+  api.registerTool((ctx: Ctx) => ({
+    name: 'fabric-guild-list',
+    description:
+      'List guilds the calling agent is a member of. Returns ' +
+      '{nodeId, name, purpose, status} per row. ' +
+      "`purpose` is a free-form description of what each guild is for — " +
+      'pick the guild whose purpose matches your intent. Use this tool ' +
+      'BEFORE fabric-channel-list when a workflow asks you to pick the ' +
+      'right guild by intent (e.g. "find a guild whose purpose mentions ' +
+      'debate broadcasts" → then list its announce-type channels).',
+    parameters: {
+      type: 'object',
+      additionalProperties: false,
+      properties: {
+        nameFilter: {
+          type: 'string',
+          description: 'optional case-insensitive substring match on guild name',
+        },
+        purposeFilter: {
+          type: 'string',
+          description:
+            'optional case-insensitive substring match on guild purpose ' +
+            '(e.g. "debate", "announcements")',
+        },
+      },
+    },
+    execute: async (_id: string, p: { nameFilter?: string; purposeFilter?: string }) => {
+      const agentId = ctx.agentId;
+      if (!agentId) return { ok: false, error: 'no agent context' };
+      const entry = identity.findByAgentId(agentId);
+      if (!entry) return { ok: false, error: `agent ${agentId} not registered` };
+      const session = await client.agentLogin(entry.fabricApiKey);
+      const nameNeedle = (p.nameFilter ?? '').toLowerCase();
+      const purposeNeedle = (p.purposeFilter ?? '').toLowerCase();
+      const guilds = session.guilds.filter((g) => {
+        if (nameNeedle && !g.name.toLowerCase().includes(nameNeedle)) return false;
+        if (purposeNeedle) {
+          const purp = (g.purpose ?? '').toLowerCase();
+          if (!purp.includes(purposeNeedle)) return false;
+        }
+        return true;
+      });
+      return {
+        ok: true,
+        count: guilds.length,
+        guilds: guilds.map((g) => ({
+          nodeId: g.nodeId,
+          name: g.name,
+          status: g.status,
+          purpose: g.purpose ?? null,
+        })),
+      };
+    },
+  }));
+
+  // -----------------------------------------------------------------
+  // fabric-channel-set-purpose: set/update a channel's free-form
+  // purpose description. Caller must be a channel member (or the
+  // channel must be public). Use this to backfill purpose on existing
+  // channels, or to refine it after a channel's role evolves.
+  // -----------------------------------------------------------------
+  api.registerTool((ctx: Ctx) => ({
+    name: 'fabric-channel-set-purpose',
+    description:
+      "Set or update a channel's free-form purpose description. " +
+      'Channel membership required (or the channel must be public). ' +
+      'Pass empty string to clear. Use this to make a channel ' +
+      'discoverable to other agents via fabric-channel-list.',
+    parameters: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['guildNodeId', 'channelId', 'purpose'],
+      properties: {
+        guildNodeId: { type: 'string' },
+        channelId: { type: 'string' },
+        purpose: {
+          type: 'string',
+          description: "What this channel is for. Pass '' (empty string) to clear.",
+        },
+      },
+    },
+    execute: async (_id: string, p: { guildNodeId: string; channelId: string; purpose: string }) => {
+      const agentId = ctx.agentId;
+      if (!agentId) return { ok: false, error: 'no agent context' };
+      const { guild, token } = await ctxGuild(agentId, p.guildNodeId);
+      const res = await client.setChannelPurpose(
+        guild.endpoint,
+        token,
+        p.channelId,
+        p.purpose,
+      );
+      return { ok: true, channel: res };
+    },
+  }));
+
+  // -----------------------------------------------------------------
+  // fabric-message-history: read a channel's recent message history by
+  // `seq`. Tail-by-default: when `seqFrom`/`seqTo` are omitted, returns
+  // the last `limit` messages (limit defaults to 20, max 200).
+  //
+  // Use cases: catch-up on a channel that was muted while the agent was
+  // gated; verify a previous message went through; lookup recent
+  // duplicates before opening a new task in triage.
+  // -----------------------------------------------------------------
+  api.registerTool((ctx: Ctx) => ({
+    name: 'fabric-message-history',
+    description:
+      "Read a channel's recent message history. Omit seqFrom/seqTo to " +
+      'tail (last `limit` messages, default 20, max 200). Backend ' +
+      'requires the calling agent to be a channel participant.',
+    parameters: {
+      type: 'object',
+      additionalProperties: false,
+      required: ['guildNodeId', 'channelId'],
+      properties: {
+        guildNodeId: { type: 'string' },
+        channelId: { type: 'string' },
+        seqFrom: { type: 'integer', minimum: 1, description: 'inclusive lower bound; default = tail' },
+        seqTo: { type: 'integer', minimum: 1, description: 'inclusive upper bound; default = channel head' },
+        limit: { type: 'integer', minimum: 1, maximum: 200, description: 'default 20' },
+      },
+    },
+    execute: async (_id: string, p: {
+      guildNodeId: string;
+      channelId: string;
+      seqFrom?: number;
+      seqTo?: number;
+      limit?: number;
+    }) => {
+      const agentId = ctx.agentId;
+      if (!agentId) return { ok: false, error: 'no agent context' };
+      const { guild, token } = await ctxGuild(agentId, p.guildNodeId);
+      const limit = p.limit ?? 20;
+
+      // Tail mode: discover channel head via channel listing, then ask
+      // for [head-limit+1, head]. Avoids needing the agent to know seq.
+      let seqFrom = p.seqFrom;
+      let seqTo = p.seqTo;
+      if (seqFrom === undefined && seqTo === undefined) {
+        const channels = await client.listChannels(guild.endpoint, token, p.guildNodeId);
+        const ch = channels.find((c) => c.id === p.channelId);
+        const head = ch?.lastSeq ?? 0;
+        seqFrom = Math.max(1, head - limit + 1);
+        seqTo = head;
+      }
+
+      const res = await client.listMessages(guild.endpoint, token, p.channelId, {
+        seqFrom,
+        seqTo,
+        limit,
+      });
+      return {
+        ok: true,
+        page: res.page,
+        messages: res.items.map((m) => ({
+          messageId: m.messageId,
+          seq: m.seq,
+          authorUserId: m.authorUserId,
+          content: m.content,
+          createdAt: m.createdAt,
+          isDeleted: m.isDeleted,
+        })),
+      };
+    },
+  }));
 }