hermoso 0.1.498 → 0.1.500

package/README.md

@@ -5,7 +5,7 @@ scripts. Research the ads already winning in a market, generate image & video ad
 composited in, copy + CTA included), publish them to your own social channels, and build & manage the ad
 campaigns behind them, all over [MCP](https://modelcontextprotocol.io) tools, a CLI, or installable Claude skills.
 
-**1110 tools.** `tools/list` is always the authoritative set; `hermoso_capabilities` (free) returns the live model
+**1182 tools.** `tools/list` is always the authoritative set; `hermoso_capabilities` (free) returns the live model
 catalog with exact per-render credit costs plus the full capability map.
 
 **No markup on models** on Starter, Pro and Studio. Stop overpaying markups on AI models. Plans: [hermoso.ai/pricing](https://hermoso.ai/pricing).
@@ -23,9 +23,9 @@ charges per API request (`hermoso_credits` prints the exact credits per X post).
 
 **What it connects to.** Ad platforms: Meta, Google Ads, TikTok Ads, LinkedIn Ads, Reddit Ads, X Ads,
 Pinterest Ads, Snapchat Ads, Microsoft Advertising, Apple Search Ads, ChatGPT Ads and AppLovin Ads, plus
-product feeds in Google Merchant Center. Publishing and scheduling — **eleven** channels: Facebook, Instagram,
-Threads, TikTok, YouTube, X, LinkedIn, Pinterest, Bluesky, Telegram and Google Business Profile. Messaging: WhatsApp (you message a
-person, so it is not a twelfth publishing channel) and DM automations on Instagram, Facebook Pages and
+product feeds in Google Merchant Center. Publishing and scheduling — **fifteen** channels: Facebook, Instagram,
+Threads, TikTok, YouTube, X, LinkedIn, Pinterest, Bluesky, Telegram, WordPress, Discord, Slack, Whop and Google Business Profile. Messaging: WhatsApp (you message a
+person, so it is not a sixteenth publishing channel) and DM automations on Instagram, Facebook Pages and
 Messenger, and X (a keyword comment, a DM, a story reply or an ig.me / m.me link gets an automatic reply). Ad research: the Meta, Google and LinkedIn ad libraries plus organic TikTok,
 Instagram, YouTube, Threads and Reddit. Analytics: Google Analytics 4, Google Search Console and every
 connected platform's own post and campaign insights. Files: Google Drive, Sheets, Docs and OneDrive.
@@ -242,7 +242,7 @@ block entirely if you signed in above; it is there for CI, where the process can
 
 Then ask your agent: *“Generate an image ad with Hermoso.”*
 
-### What the 1110 tools cover
+### What the 1182 tools cover
 
 **Ad spy / research** — `find_competitors`, `competitor_teardown`, `pull_competitor_ads`, `research_ads`; the
 Meta / Google / LinkedIn ad libraries (`search_meta_ads`, `search_google_ads`, `search_linkedin_ads`); organic
@@ -278,13 +278,16 @@ Real faces from your own photos need a paid plan; AI people work on every plan.
 per-render credit cost) with no ad framing: `generate_image` / `generate_video` with `useBrand:false`,
 `generate_voice`, `generate_text`. The image default is Nano Banana 2.1.
 
-**Publish to your own channels** — **eleven** of them: Facebook, Instagram and Threads (`post_to_meta`), TikTok
+**Publish to your own channels** — **fifteen** of them: Facebook, Instagram and Threads (`post_to_meta`), TikTok
 (`post_to_tiktok`), YouTube (`post_to_youtube` + `update_youtube_video`, `youtube_video_insights`, comments
 read/reply), X (`post_to_x`, `x_post_metrics`, `x_post_insights`, `x_mentions`, `list_x_dms`, `send_x_dm`),
 LinkedIn profile **and** company Pages (`post_to_linkedin`, `post_to_linkedin_page`), Pinterest
 (`post_to_pinterest` + boards), Bluesky (`post_to_bluesky`, `delete_bluesky_post`, `bluesky_post_metrics`, plus
 `list_bluesky_convos` / `read_bluesky_dm` / `send_bluesky_dm`) and Telegram (`post_to_telegram`,
-`delete_telegram_message`, `list_telegram_chats`) and Google Business Profile (below). `schedule_post` / `list_scheduled` / `cancel_scheduled` give
+`delete_telegram_message`, `list_telegram_chats`), WordPress (`post_to_wordpress`, `update_wordpress_post`,
+`delete_wordpress_post`, `list_wordpress_posts`), Discord and Slack announcements (`post_to_discord`, `edit_discord_message`,
+`delete_discord_message`, `post_to_slack`), Whop (`post_to_whop`, `edit_whop_post`, `delete_whop_post`, `get_whop_post`,
+`list_whop_destinations`) and Google Business Profile (below). `schedule_post` / `list_scheduled` / `cancel_scheduled` give
 you one content calendar over exactly that set. `upload_file` brings in any external or local media, not just
 Hermoso renders.
 *X posting bills credits per API call (X charges per request); a post containing a link costs 13× one without.*

package/bin/hermoso.mjs

@@ -153,7 +153,7 @@ async function main() {
         return out(`Balance: ${bal ?? '—'} credits`, d); }
       case 'brand': {
         // BARE `hermoso brand` SHOWS THE SAVED BRAND (2026-09-25). The docs have always listed it as "the saved brand
-        // profile" and it died with the `brand draft` usage line instead — the first command a new user copies off
+        // profile" and it printed the `brand draft` usage line instead — the first command a new user copies off
         // the CLI page. It is get_brand, run through the same handler the MCP twins use.
         if (!sub) { const reg = await import('../mcp/registry.mjs'); return await runTool(reg, 'get_brand', flags, []); }
         if (sub !== 'draft') return die('usage: hermoso brand            the saved profile (a brand, creator or personal workspace)\n       hermoso brand draft (--domain <d> | --description <t> | --social <h> --platform <p>)');

package/mcp/chatgpt-roster.mjs

@@ -0,0 +1,51 @@
+// GENERATED by tools/chatgpt-roster-rank.mjs. Do not edit by hand: re-run the tool (it ranks by the committed usage in
+// tools/lib/chatgpt-roster-usage.json). The provider families ChatGPT lists, best first; a family is all or nothing.
+// The core (every tool that needs no connection) is always listed and is not in these lists. Cap: CHATGPT_TOOL_CAP
+// (950) in mcp/tools.mjs, under ChatGPT's own 1,000-tool limit.
+export const CHATGPT_ROSTER_FAMILIES = Object.freeze([
+  'google_drive', // long tail, pinned (plugin skills)
+  'intercom', // long tail, pinned (plugin skills)
+  'meta', // publishing
+  'youtube', // publishing
+  'tiktok', // publishing
+  'x', // publishing
+  'pinterest', // publishing
+  'google_business', // publishing
+  'linkedin', // publishing
+  'google_ads', // big-platform ads
+  'google_analytics', // analytics
+  'threads', // publishing
+  'tiktok_ads', // big-platform ads
+  'google_search_console', // analytics
+  'telegram', // publishing
+  'google_tag_manager', // long tail
+  'snapchat_ads', // long tail
+  'microsoft_onedrive', // long tail
+  'bluesky', // publishing
+  'openai_ads', // big-platform ads
+  'reddit_ads', // long tail
+  'bing_webmaster', // analytics
+]);
+// Left out to stay under the cap, in rank order (the first one is the next to get in when room appears).
+export const CHATGPT_ROSTER_LEFT_OUT = Object.freeze([
+  'x_ads', // long tail
+  'mixpanel', // analytics
+  'apple_ads', // long tail
+  'pinterest_ads', // long tail
+  'discord', // publishing
+  'slack', // publishing
+  'wordpress', // publishing
+  'microsoft_ads', // big-platform ads
+  'amplitude', // analytics
+  'hubspot', // analytics
+  'klaviyo', // analytics
+  'mailchimp', // analytics
+  'posthog', // analytics
+  'stripe', // analytics
+  'applovin_ads', // long tail
+  'canva', // long tail
+  'notion', // long tail
+  'vercel', // long tail
+  'whop', // long tail
+]);
+export const CHATGPT_ROSTER_META = Object.freeze({ cap: 950, usageCollectedAt: "2026-11-03T20:25:36.816Z" });

package/mcp/client.mjs

@@ -253,10 +253,9 @@ const isTransport = (e) => !!e && e.name === 'TypeError' && TRANSPORT_RE.test(St
 
 // ── THE PLATFORM'S OWN 429 NEVER REACHED US, SO IT IS RETRIED HERE (2026-10-04) ───────────────────────────────
 // Polling a job every 10 seconds from the CLI answered HTTP 429, and it was not a limit of ours: the app has no limiter
-// on GET /api/jobs/:id, and /v1 reads allow 300 a minute per key. Every one of those 429s in the Cloud Run request log
-// (36 between 18:35 and 18:39Z, on /api/credits, GET / and webhooks as much as on job polls) is the platform's "The
+// on GET /api/jobs/:id, and /v1 reads allow 300 a minute per key. A JSON-less 429 is the platform's "The
 // request was aborted because there was no available instance": a plain-text body, latency 0s, refused before the
-// container saw it, while the one instance's CPU sat pegged by local renders at 2-7 concurrent requests. The browser's
+// container sees it, and it can land on any route, not only on job polls. The browser's
 // api() has retried that answer since 2026-09-06; the CLI and the stdio MCP read it as a failed call, so one busy
 // second ended a render wait. A refusal that never reached the app ran nothing, so it is retried for ANY method; the
 // app's own refusals always answer JSON and pass through untouched (their rules are theirs). A read also rides out a
@@ -439,7 +438,9 @@ export async function connectedProviders() {
     // shopifyStore (2026-10-27): true/false when the server could tell whether this account has a Shopify store;
     // anything else is null = unknown, which holds nothing back (roster-scope.mjs toolHeldForNoShopifyStore).
     const shopifyStore = r?.shopifyStore === true ? true : r?.shopifyStore === false ? false : null;
-    return { connected: new Set(list.filter((p) => typeof p === 'string' && p)), readOk: true, offered, gated, connectLink, withheld, shopifyStore };
+    // flagsOn (2026-11-03): tools that ship dark and are switched ON on this server (roster-scope.mjs FLAGGED_OFF_TOOLS).
+    const flagsOn = new Set((Array.isArray(r?.flagsOn) ? r.flagsOn : []).filter((n) => typeof n === 'string' && n));
+    return { connected: new Set(list.filter((p) => typeof p === 'string' && p)), readOk: true, offered, gated, connectLink, withheld, flagsOn, shopifyStore };
   } catch { return { connected: new Set(), readOk: false }; }
 }
 // Upload raw file BYTES to /api/upload (150MB, persists → returns {url,kind,bytes}). Overrides the JSON content-type so

package/mcp/http.mjs

@@ -16,7 +16,7 @@ import { randomUUID, createHash } from 'node:crypto';
 import { createRequire } from 'node:module';
 import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
 import { StreamableHTTPServerTransport } from '@modelcontextprotocol/sdk/server/streamableHttp.js';
-import { registerTools, MCP_INSTRUCTIONS, instructionsFor, parseToolScope, DEFAULT_TOOL_GROUPS } from './tools.mjs';
+import { registerTools, MCP_INSTRUCTIONS, instructionsFor, parseToolScope, DEFAULT_TOOL_GROUPS, anonDiscoveryGroups } from './tools.mjs';
 import { chatgptExtAsked } from './openai-extensions.mjs';
 import { mcpCtx, connectedProviders } from './client.mjs';
 
@@ -120,24 +120,24 @@ export function mountRemoteMcp(app, { verifyBearer, publicBaseUrl, onSessionStar
 
   // Per-session Streamable-HTTP transports. Each authenticated session gets its own McpServer with the same tools.
   //
-  // ── A SESSION WAS EXPENSIVE, AND THIS MAP IS WHY PROD OOM'd (2026-08-01, again 2026-08-24) ───────────────────
+  // ── A SESSION WAS EXPENSIVE, AND THIS MAP IS WHY IT IS BOUNDED (2026-08-01, again 2026-08-24) ─────────────────
   // registerTools() used to rebuild all 681 tool definitions with their zod schemas on EVERY `initialize`:
   // **114.36 MB of RETAINED heap per McpServer** (measured, node --expose-gc, 25 instances; 114.48 MB measured
   // again end-to-end through this very transport). The older note here said ~36 MB — true at the 248-tool roster
   // it was written against, and 3.2x stale by August. This map used to be unbounded and never expired — the only
-  // removal was transport.onclose, which never fires for a client that simply goes away. Prod took 238
-  // `initialize` handshakes in 49 minutes on a 4 GiB, maxScale=1 instance and died of
-  // `FATAL ERROR: Reached heap limit` six times in that hour, every crash a full outage.
+  // removal was transport.onclose, which never fires for a client that simply goes away, so a burst of
+  // `initialize` handshakes from clients that never close can grow the heap until the process reaches its
+  // memory limit.
   //
-  // THE CAP AND THE TTL WERE NEVER THE FIX, AND 2026-08-24 PROVED IT: they worked exactly as designed (peak
-  // concurrent sessions measured at exactly SESSION_MAX) and prod still crashed four times in twelve hours,
-  // because the cost is CHURN, not retention — one client opened ~175 sessions in 65 minutes. Worse, the cap was
-  // FEEDING the churn: 157 of 258 session closures in a 3-day window were 'over cap', each one answering that
+  // THE CAP AND THE TTL ARE NOT THE FIX ON THEIR OWN: they work exactly as designed (peak concurrent sessions
+  // measured at exactly SESSION_MAX) and memory still runs short under load,
+  // because the cost is CHURN, not retention — one client re-initialized over and over. Worse, the cap was
+  // FEEDING the churn: most session closures in a 3-day window were 'over cap', each one answering that
   // client's next call with a 404, which obliges it to re-initialize, which allocates another full roster and
   // evicts somebody else. A self-sustaining spiral. The durable fix is in mcp/tools.mjs: the tool definitions are
   // static per process, so they are built ONCE and replayed. A session now costs **3.06-3.19 MB** — 36x less —
   // and it is the per-`initialize` figure, not the per-live-session one, that decides whether this survives a
-  // retry storm. See tools/mcp-tool-canon-check.mjs.
+  // burst of retries. See tools/mcp-tool-canon-check.mjs.
   //
   // The 400-on-a-session-miss below was the ACCELERANT: a client that cannot re-initialize opens a NEW session
   // instead of reusing its own, so the leak fed itself. Both are fixed here — bound + expire the map, and answer
@@ -150,11 +150,10 @@ export function mountRemoteMcp(app, { verifyBearer, publicBaseUrl, onSessionStar
   const SESSION_IDLE_MS = Math.max(1000, Number(process.env.MCP_SESSION_IDLE_MS || 30 * 60e3)); // 1s floor so the expiry is TESTABLE; a short TTL is merely wasteful now that eviction is recoverable
   const sessions = new Map(); // mcp-session-id -> { transport, server, user, lastSeen }  (insertion-ordered = LRU)
   // ── A STANDALONE SSE STREAM HOLDS A CLOUD RUN REQUEST SLOT FOR AS LONG AS IT LIVES (2026-09-15) ──────────────
-  // OUTAGE, 13:00–13:22 UTC: every request to the app answered 429 "no available instance" for 22 minutes with the
-  // instance at 1% CPU. Nothing was down and nobody was busy — the instance's 200 concurrent-request slots were all
-  // held by `GET /mcp` notification streams (99 of them from ChatGPT's connector that morning, each held until the
-  // 30-minute session idle), and Cloud Run refuses a request it has no slot for BEFORE the container sees it, so
-  // the app could not even log it. maxScale=1 turns "one host holds too many streams" into a total outage.
+  // An instance's concurrent-request slots can all be held by `GET /mcp` notification streams (ChatGPT's connector
+  // opens one per session and holds it until the 30-minute session idle) while its CPU sits idle, and Cloud Run
+  // refuses a request it has no slot for BEFORE the container sees it, so the app cannot even log the refusal.
+  // Unbounded, "one host holds too many streams" saturates the instance for every other caller.
   // The GET stream is optional in the spec (a server may answer 405 or close it at any time; clients reconnect),
   // and this server sends nothing on it but a one-shot tools/list_changed nudge. So: a stream lives at most
   // MCP_GET_STREAM_MS (5 min, env-tunable) and the instance holds at most MCP_GET_STREAM_MAX of them at once —
@@ -221,13 +220,13 @@ const inflightNameOf = (body) => { const msgs = Array.isArray(body) ? body : [bo
 
   // MCP spec 2025-06-18, Transports § Session Management: a server that no longer has a session MUST answer 404,
   // and on 404 the client MUST start a new session with a fresh InitializeRequest. 400 carries no such obligation,
-  // so a 400 here is permanent death for that connection — which is exactly what every deploy used to do to every
-  // hosted customer. -32001 "Session not found" is the SDK's own code for this case; we mint it ourselves because
+  // so a 400 here is permanent death for that connection, which is what a restart would otherwise do to every
+  // hosted connection. -32001 "Session not found" is the SDK's own code for this case; we mint it ourselves because
   // the SDK's branch is unreachable when there is no initialized transport left to compare the id against.
   const sessionGone = (res) => res.status(404).json({ jsonrpc: '2.0', error: { code: -32001, message: 'Session not found' }, id: null });
   // The SDK's own answer for a non-initialize request that names no session at all. We reply BEFORE constructing
-  // anything: building a 36 MB McpServer purely to have it reject the request is what turned a retry storm into
-  // an OOM. Status and message are byte-identical to the SDK's, so no client sees a behaviour change.
+  // anything: building a 36 MB McpServer purely to have it reject the request turns a burst of retries into
+  // memory pressure. Status and message are byte-identical to the SDK's, so no client sees a behaviour change.
   const needSession = (res) => res.status(400).json({ jsonrpc: '2.0', error: { code: -32000, message: 'Bad Request: Mcp-Session-Id header is required' }, id: null });
   // The one place the host name is turned into a decision, so the anon and session paths cannot diverge.
   //
@@ -407,7 +406,9 @@ const inflightNameOf = (body) => { const msgs = Array.isArray(body) ? body : [bo
     // every one of those the roster IS the product description. Serving them the core set would publish Hermoso as
     // a 22-tool server on ~430 directory pages. So an UNSTATED scope here resolves to the full pre-core-first
     // default rather than to the session default; an explicit `?tools=` still wins, exactly as it does below.
-    registerTools(server, { only: scope?.groups || [...DEFAULT_TOOL_GROUPS], directory: scope?.directory || false, widgetHost, hosted: true, ext: extAsked(req) }); // metadata only — tools/list never invokes a handler, and tools/call can't reach here
+    // A widget host in full mode gets every group here too, so the anonymous roster and the signed-in one agree (see
+    // chatgptRosterMode in tools.mjs: ChatGPT cannot reach an unlisted tool).
+    registerTools(server, { only: scope?.groups || anonDiscoveryGroups(widgetHost), directory: scope?.directory || false, widgetHost, hosted: true, ext: extAsked(req) }); // metadata only — tools/list never invokes a handler, and tools/call can't reach here
     // WHO PROBES US WITHOUT A TOKEN, BY NAME (2026-09-25). A host's add-connector dialog decides "sign-in needed or
     // not" from THIS answer: ChatGPT probes with an empty body and gets the 401; claude.ai's dialog sends a real
     // tokenless initialize (UA python-httpx) and our 200 made it pre-select "No sign-in". The UA alone cannot tell
@@ -456,7 +457,7 @@ const inflightNameOf = (body) => { const msgs = Array.isArray(body) ? body : [bo
       return challenge(res);
     }
 
-    // The stream cap, BEFORE any session work: a reconnect storm must be refused at the door, not after allocating.
+    // The stream cap, BEFORE any session work: a burst of reconnects must be refused at the door, not after allocating.
     if (req.method === 'GET') {
       if (openGetStreams >= GET_STREAM_MAX) {
         res.set('Retry-After', '30');
@@ -534,8 +535,8 @@ const inflightNameOf = (body) => { const msgs = Array.isArray(body) ? body : [bo
       // tools/call. It is advisory only: it may not change auth, scope or spend — it decides PRESENTATION.
       // A CHATGPT ACCOUNT THAT HAS MADE NOTHING YET (2026-10-19). Read once per session, ChatGPT hosts only, bounded so a slow
       // read can never hold the connect: null (no note) is the answer to every failure. The first successful tool reply of
-      // the session then offers one concrete next ask (firstRunNote in tools.mjs). 9 of 13 ChatGPT signups since the
-      // 10-15 publish spent nothing, and a live "make me a video ad" chat on a connected account never called us.
+      // the session then offers one concrete next ask (firstRunNote in tools.mjs). Most ChatGPT signups since the
+      // 10-15 publish had made nothing, and a live "make me a video ad" chat on a connected account never called us.
       let firstRun = null;
       if (widgetHost && typeof firstRunFor === 'function' && user.accountId) {
         try { firstRun = await Promise.race([Promise.resolve(firstRunFor({ accountId: user.accountId })).catch(() => null), new Promise((r) => { const t = setTimeout(() => r(null), 1500); if (t && typeof t.unref === 'function') t.unref(); })]); } catch { firstRun = null; }
@@ -578,7 +579,7 @@ const inflightNameOf = (body) => { const msgs = Array.isArray(body) ? body : [bo
       else if (m === 'tools/call') {
         entry.calls = (entry.calls || 0) + 1;
         // THE FIRST CALL IS THE MILESTONE, NOT THE LISTING (2026-09-07). The connect_mcp reward was granted on "an API
-        // key exists", and 37 sessions last week listed the roster and never called a tool while 803 grants went out.
+        // key exists", and many sessions listed the roster and never called a tool while the grant still went out.
         // A session's first tools/call is what proves the connection is USED; the hook stamps the account's touch row.
         if (entry.calls === 1 && typeof onFirstCall === 'function' && entry.user?.accountId) { try { onFirstCall({ accountId: entry.user.accountId, userId: entry.user.userId || null, client: entry.client || '', ua: entry.ua || '', src: entry.src || null }); } catch {} }
       }

package/mcp/openai-extensions.mjs

@@ -1268,7 +1268,7 @@ export const neutralNote = (t) => {
 // sweeps and drift watches queued for later (measured on a fresh local account: one `agentidle` "nudge", queued for
 // tomorrow, would have shown as "1 render in progress"). Mirrors server.js JOB_TYPES_HIDDEN_FROM_V1, and the check
 // fails if that list gains a type this one lacks. A job queued for LATER is not in progress either.
-export const PANEL_HIDDEN_JOB_TYPES = Object.freeze(['fallate', 'storysweep', 'ttpublicurl', 'whatsappcoexsync', 'purge', 'agentidle', 'welcomewait', 'agentsetupmail', 'uploadretention', 'miniappretention', 'connectordrift', 'falwatch', 'faceassetsweep', 'searchvis', 'mcpcanary', 'welcomeads', 'lifecycleedu', 'connhealth', 'chargednothing', 'emailfeedback']);
+export const PANEL_HIDDEN_JOB_TYPES = Object.freeze(['fallate', 'storysweep', 'ttpublicurl', 'whatsappcoexsync', 'purge', 'agentidle', 'welcomewait', 'agentsetupmail', 'uploadretention', 'miniappretention', 'connectordrift', 'falwatch', 'faceassetsweep', 'searchvis', 'mcpcanary', 'welcomeads', 'lifecycleedu', 'connhealth', 'chargednothing', 'badoutcome', 'emailfeedback', 'weeklyposts']);
 const ageText = (h) => (h == null || !Number.isFinite(+h) ? '' : +h < 1 ? 'just now' : +h < 48 ? `${Math.round(+h)}h ago` : `${Math.round(+h / 24)}d ago`);
 const VIDEO_RE = /\.(mp4|webm|mov|m4v)([?#]|$)/i;
 const clip = (s, n) => { const t = String(s ?? '').replace(/\s+/g, ' ').trim(); return t.length > n ? t.slice(0, n - 1) + '…' : t; };
@@ -1279,8 +1279,8 @@ const EDITABLE_POST = ['queued', 'scheduled'];
 const PROVIDER_INFO = {
   meta: ['Meta (Facebook and Instagram)', 'channels'], instagram: ['Instagram', 'channels'], threads: ['Threads', 'channels'],
   tiktok: ['TikTok', 'channels'], youtube: ['YouTube', 'channels'], linkedin: ['LinkedIn', 'channels'], x: ['X', 'channels'],
-  pinterest: ['Pinterest', 'channels'], bluesky: ['Bluesky', 'channels'], telegram: ['Telegram', 'channels'], reddit: ['Reddit', 'channels'],
-  google_business: ['Google Business Profile', 'channels'], whatsapp: ['WhatsApp', 'channels'], snapchat: ['Snapchat', 'channels'],
+  pinterest: ['Pinterest', 'channels'], bluesky: ['Bluesky', 'channels'], telegram: ['Telegram', 'channels'],
+  google_business: ['Google Business Profile', 'channels'], whatsapp: ['WhatsApp', 'channels'],
   google_ads: ['Google Ads', 'ads'], tiktok_ads: ['TikTok Ads', 'ads'], linkedin_ads: ['LinkedIn Ads', 'ads'], microsoft_ads: ['Microsoft Advertising', 'ads'],
   reddit_ads: ['Reddit Ads', 'ads'], pinterest_ads: ['Pinterest Ads', 'ads'], snapchat_ads: ['Snapchat Ads', 'ads'], x_ads: ['X Ads', 'ads'],
   openai_ads: ['ChatGPT Ads', 'ads'], apple_ads: ['Apple Ads', 'ads'], applovin_ads: ['AppLovin Ads', 'ads'],

package/mcp/queued-image-note.mjs

@@ -0,0 +1,14 @@
+// THE LOGO IS SAID WHERE IT LANDED, ONCE — ON THE QUEUED MCP PATH TOO (journey QA 2026-11-03). The stdio/CLI transport
+// (and /v1 with ?wait=) queues generate_image and then joins two notes: the route's, written BEFORE the render ("The
+// brand's real logo file is laid over the finished picture (bottom right).") and the worker's, written after it
+// ("Your logo was laid over the picture in the top left corner…"). The synchronous route already drops the first when
+// the worker moved the logo (lib/logo-note.mjs dropMovedLogoClaim, 2026-10-19); this path joined both, so one reply put
+// the logo in two corners. Same rule, here, because the CLI copy of this tree cannot import ../lib.
+// tools/journey-qa-1019-check.mjs runs it against the server's own function on the same cases. Pure.
+const ROUTE_CLAIM = /\s*The brand's real logo file is laid over the finished picture \([^)]*\)\./;
+export function joinQueuedImageNote(routeNote, raw) {
+  const lo = raw && raw.logoOverlay;
+  let route = String(routeNote ?? '');
+  if (lo && lo.applied === true && lo.chosen && lo.chosen !== 'asked') route = route.replace(ROUTE_CLAIM, '');
+  return [route.trim(), String(raw?.productNote ?? '').trim()].filter(Boolean).join(' ');
+}

package/mcp/roster-scope.mjs

@@ -52,6 +52,7 @@
 export const NEVER_GATE = new Set([
   'search_meta_ads', 'search_google_ads', 'search_linkedin_ads', 'search_tiktok',
   'search_instagram', 'search_youtube', 'search_reddit', 'search_threads', 'fetch_social_data',
+  'tiktok_ideas', // public TikTok post lookups on our research key + one writing call, never a TikTok connection (2026-11-03)
   'search_meta_places', // place search runs on our data provider and a web search, never on a Meta connection (2026-10-27)
 ]);
 
@@ -104,6 +105,10 @@ export const TOOL_PROVIDER_RULES = [
   [/notion/, 'notion'],
   // Canva (2026-10-27): the user's OWN Canva, offered only to a workspace that connected it.
   [/canva/, 'canva'],
+  // Klaviyo (2026-11-03): the brand's OWN email marketing, offered only to a workspace that connected it.
+  [/klaviyo/, 'klaviyo'],
+  // Mailchimp (2026-11-03): the brand's OWN email marketing, offered only to a workspace that connected it.
+  [/mailchimp/, 'mailchimp'],
   // ── posting-only channels ──
   [/youtube/, 'youtube'],
   [/google_business|business_location|^list_business_(categories|attributes)$|^business_google_updated$/, 'google_business'],
@@ -111,6 +116,13 @@ export const TOOL_PROVIDER_RULES = [
   [/onedrive/, 'microsoft_onedrive'],
   [/bluesky/, 'bluesky'],
   [/telegram/, 'telegram'],
+  // WordPress, Discord and Slack publishing (2026-11-03): the brand's OWN site and webhooks. Discord/Slack posting rides the
+  // same connector row as their alerts (a second, announcement webhook), so the provider is the bare name.
+  [/wordpress/, 'wordpress'],
+  [/discord/, 'discord'],
+  [/slack/, 'slack'],
+  // Whop (2026-11-03): the brand's OWN whop, offered only to a workspace whose whop installed Hermoso.
+  [/whop/, 'whop'],
   // TIKTOK IS THREE CREDENTIALS, AND ONLY FIVE TOOLS RIDE THE POSTING ONE (2026-10-29). Until this date the bare rule sent
   // every remaining tiktok name to 'tiktok', so 41 Business, analytics and creator-marketplace tools were offered by the
   // POSTING login while their routes spend the TikTok Ads advertiser token or its account-holder grant: a TikTok Ads-only
@@ -126,13 +138,12 @@ export const TOOL_PROVIDER_RULES = [
 ];
 
 // The provider a tool needs, or null when this module cannot attribute it. null ⇒ NEVER dropped (property 2).
-// ── PROVIDERS HERMOSO DOES NOT OFFER, AND AS OF NOW WILL NOT (2026-09-03) ────────────────────────────────
-// "make sure posting to reddit or snapchat are never offered, never described, never wasting context in a tools
-// list". Reddit organic posting needs Reddit's API approval we do not have; Snapchat posting is not even built. Both
-// have ADS connectors that are live and are NOT in this set (`reddit_ads`, `snapchat_ads`). A tool whose provider
-// is here is not registered on any roster, not listed by find_tools, not offered by the Studio agent, and not
-// counted. Gating at call time was not enough: a disabled tool still costs its schema in every canon and its name
-// in every listing. If Reddit ever approves the app, remove it from this set and the whole surface returns.
+// ── PROVIDERS HERMOSO DOES NOT OFFER: REDDIT AND SNAPCHAT ORGANIC POSTING, PERMANENTLY ────────────────────────
+// Permanently not offered (owner 2026-11-03: "reddit posting and snapchat posting are never happening … they should
+// never be in the static list or anywhere"). Their OAuth descriptors, gates, tiles, routes and tools were DELETED that
+// day; this set is the BELT that stays, so a tool named for either provider that ever reappears is not registered on
+// any roster, not listed by find_tools, not offered by the Studio agent and not counted. The ADS connectors are live
+// and are NOT in this set (`reddit_ads`, `snapchat_ads`). Never remove an entry from this set.
 export const UNOFFERED_PROVIDERS = new Set(['reddit', 'snapchat']);
 export function toolUnoffered(name) { const p = toolProvider(name); return !!p && UNOFFERED_PROVIDERS.has(p); }
 
@@ -143,7 +154,7 @@ export function toolUnoffered(name) { const p = toolProvider(name); return !!p &
 // the Studio agent's schema and executor) asks THIS function, so the answer cannot differ between them.
 // The set mirrors lib/multi-account.mjs MULTI_ACCOUNT_PROVIDERS, which the npm package cannot import;
 // tools/multi-account-arg-check.mjs fails when the two differ, and DERIVES the tool split below from server.js.
-export const MULTI_ACCOUNT_CHANNELS = Object.freeze(new Set(['tiktok', 'x', 'youtube', 'threads', 'bluesky', 'telegram', 'reddit', 'pinterest', 'instagram']));
+export const MULTI_ACCOUNT_CHANNELS = Object.freeze(new Set(['tiktok', 'x', 'youtube', 'threads', 'bluesky', 'telegram', 'pinterest', 'instagram']));
 // Filed under a multi-account channel, but routed through a connection that holds ONE grant per brand, so `account`
 // would do nothing there: these two Pinterest reads ride the ad-account (pinterest_ads) row. (The TikTok Business and
 // Ads tools are no longer filed under 'tiktok' at all — only the five posting tools are — so they need no entry.)
@@ -227,7 +238,14 @@ export const SHOPIFY_STORE_TOOL_RE = /shopify/;
 export const toolNeedsShopifyStore = (name) => SHOPIFY_STORE_TOOL_RE.test(String(name || ''));
 export function toolHeldForNoShopifyStore(name, conn) { return !!(conn && conn.shopifyStore === false && toolNeedsShopifyStore(name)); }
 export const SHOPIFY_STORE_NOTE = 'works only for a Shopify store that installed Hermoso from the Shopify App Store, and this account has no linked store. An existing Hermoso account cannot be linked to a store from here; a merchant who did install it opens Hermoso once from their Shopify admin (Apps \u2192 Hermoso).';
-export function toolWithheldByServer(name, conn) { return !!(conn && ((conn.withheld instanceof Set && conn.withheld.has(String(name || ''))) || toolHeldForNoShopifyStore(name, conn))); }
+// ── A TOOL THAT SHIPS DARK (2026-11-03, owner: "reangle_video can't stay offered while it fails its one promise") ──
+// The opposite polarity of `withheld`: a tool here is OFF unless the server's providers read names it in `flagsOn`
+// (server.js reangleOffered(), env REANGLE_VIDEO=on). FAIL CLOSED: no read, a failed read or an older server all keep
+// it hidden, so the docs build, a CLI with no session and an unreadable store never offer a capability that is not
+// ready. The route and its job types refuse it for free as well, so a stale roster cannot spend.
+export const FLAGGED_OFF_TOOLS = new Set(['reangle_video']);
+export function toolFlaggedOff(name, conn) { const n = String(name || ''); return FLAGGED_OFF_TOOLS.has(n) && !(conn && conn.flagsOn instanceof Set && conn.flagsOn.has(n)); }
+export function toolWithheldByServer(name, conn) { return toolFlaggedOff(name, conn) || !!(conn && ((conn.withheld instanceof Set && conn.withheld.has(String(name || ''))) || toolHeldForNoShopifyStore(name, conn))); }
 export function toolHeldBackByConnectors(name, conn) {
   if (toolWithheldByServer(name, conn)) return true;              // the server says this capability is not offered now
   if (!conn || !conn.readOk) return false;                       // property 1 — fail OPEN on an unreadable store

package/mcp/static-batch.mjs

@@ -5,7 +5,7 @@
 // different selling angle with its headline and CTA on the image, in 1:1, 4:5 or 9:16. We had every part of that
 // (draft_brand, the product library, the batch planner the web Studio fans into up to 30 ranked statics) but an agent
 // had to stitch it together itself: plan_variations (capped at 8, plans only), then one generate_image per variant —
-// and a CLI generate_image held past 60s died client-side while the render still billed. This is the one-call shape.
+// and a CLI generate_image held past 60s timed out client-side while the render still billed. This is the one-call shape.
 //
 // Everything here is PURE so tools/static-ads-batch-check.mjs runs the real functions. Byte-twin: cli/mcp/static-batch.mjs.
 
@@ -144,7 +144,10 @@ export function staticAdPrompt(v, { logo = 'server', productPhoto = false, layou
     x.supporting ? `supporting line "${String(x.supporting).trim()}"` : '',
     x.cta ? `call to action "${String(x.cta).trim()}"` : '',
   ].filter(Boolean);
-  if (words.length) lines.push(`On-image text, spelled exactly as given and nothing else: ${words.join(', ')}.${L ? '' : ' Bold ad typography with a clear hierarchy, crisp and correctly spelled.'} A finished, ready-to-run static ad.`);
+  // TEXT SLOTS (2026-11-03, catalog audit §3): every slot carries its literal
+  // quoted words, and type under ~2% of the picture's height turns to gibberish, so no line is set that small and a line
+  // set small is 4-6 words (a longer one breaks onto two lines rather than shrinking).
+  if (words.length) lines.push(`On-image text, spelled exactly as given and nothing else: ${words.join(', ')}.${L ? '' : ' Bold ad typography with a clear hierarchy, crisp and correctly spelled.'} No line of text is smaller than about 2% of the picture's height (smaller type turns to gibberish); a line set in small type holds at most 4-6 words, so a longer one breaks onto two lines instead of shrinking. A finished, ready-to-run static ad.`);
   if (L) lines.push('Show the product once (one unit as it is sold) unless the layout is a flat lay or a set of its range; any person has at most two hands and every hand in frame belongs to a visible arm.');
   if (productPhoto) lines.push('The first reference image is the REAL product: keep its shape, colours and label exactly.');
   // THE LOGO IS LAID ON, NEVER PAINTED (2026-10-06): 'overlay' (and the old 'attached' / 'server') add nothing here. The

package/mcp/tool-hints.mjs

@@ -128,6 +128,19 @@ export function finishedRenderKind(tool, r) {
   return '';
 }
 
+// ── A FINISHED RENDER ON A WIDGET HOST NEEDS NO OTHER PICTURE (2026-11-03) ───────────────────────────────────────────
+// Measured on ChatGPT: after generate_image returned a finished Fernfizz ad, the reply showed ChatGPT's OWN image-search
+// result beside it (a 3D-model page from an unrelated site, rendered as a "search image" chip), because the model wanted
+// a picture in its answer and our card was the only one it had been told not to repeat. The card already shows the
+// result, so the reply says so and asks for no other picture. Agent-facing, not user-facing. Pure, no imports.
+export const WIDGET_RESULT_NOTE = (kind) => `Note for the assistant, not for the user: the Hermoso card in this reply already shows the finished ${kind === 'video' ? 'video' : 'image'}. Show no other picture with it, no image search and no web images, because none of them is this result.`;
+export function withWidgetResultNote(tool, r, widgetHost) {
+  if (!widgetHost) return r;
+  const kind = finishedRenderKind(tool, r);
+  if (!kind) return r;
+  return { ...r, content: [...r.content, { type: 'text', text: WIDGET_RESULT_NOTE(kind) }] };
+}
+
 // ── OUT OF CREDITS ON A HOST THAT FORBIDS SELLING CREDITS (2026-09-30, OpenAI's plugin guidelines) ─────────────────
 // "Plugins may: explain that a certain feature is not available with the user's current plan or entitlement; link to
 // an informational page … Plugins may not: link directly to a checkout … or promote upgrades." ChatGPT is not offered