Ultimate Guild Invite

A WoW Classic AddOn to help recruit new members into your guild.
Back to Files

UltimateGuildInvite-v2.27.0

File nameFastGuildInvite-UltimateGuildInvite-v2.27.0.zip
Uploader
PmptastyPmptasty
Uploaded
Nov 18, 2026
Downloads
238
Size
7.3 MB
Flavors
RetailMoP ClassicClassic TBCClassicForever
File ID
9690256
Type
R
Release
Supported game versions
  • 12.1.0
  • 12.0.7
  • 12.0.5
  • 5.5.4
  • 4.4.2
  • 3.4.5
  • 2.5.6
  • 1.60.1
  • 1.15.9

What's new

<UGI> Ultimate Guild Invite (formerly FastGuildInvite)

[v2.27.0] (2026-11-18) — an announcement log, templates that confirm they arrived, and chat left alone on retail and WoW Forever

Fixed — templates sent on WoW Forever never arrived, while the sender was told "Sent" (Modules/FGI_TemplateWire.lua, Tests/zz_template_wire_spec.lua)

Vishiswaz (Discord, 2026-11-06, Forever): the sender read "Sent the template" and the recipient got nothing. "Sent" only meant the sender's own client took every piece -- ChatThrottleLib's didSend is C_ChatInfo.SendAddonMessage's local result code, with no receipt -- and whether Forever's server delivers an addon WHISPER to a "First Last" name was never verified. Templates now go on the GUILD addon channel addressed inside the message (G <to> <name> <encoded>, to through Recruiter:ResolveName and fn:fullPlayerName); every other guildmate's client reads the addressee and drops the rest before logging or decoding. The recipient stores it and replies K <sender> <name> on GUILD, and the sender is told "received" only on that reply. The queue's verdict now treats only true as sent: lost, error and suppressed (nil) used to print "Sent" too. The old whispered T form is still accepted from older senders. KNOWN COST: an older recipient drops G, so a sender sees "sent" and never "received"; and a template's few KB now cross the guild channel, read and discarded by every online guildmate's client. No timeout on the confirmation (the operator's no-timer rule): the sent line says a missing "received" line means offline or an older version. Not verified in a client.

Peer Review 797e1ede, same day: the old T form named no addressee and was accepted from any channel, so a T sent on GUILD would land on every trusting guildmate's shelf; it is now taken only when WHISPERED, the only way an older sender ever sent it (new spec). The "Sent" and "Couldn't deliver" lines name the ADDRESSED full name rather than the typed text, so a misspelt or unresolved target that gained the sender's realm is visible (F4, and F4b for the failure line). templateAwaitConfirm and templateConfirmed are in Locale/enUS.lua now, not only as fallbacks. F5: sending refuses a template over 16 KB encoded (MAX_GUILD_ENCODED, "toobig"), since every online guildmate's AceComm reassembles a guild-channel message before UGI reads its addressee; receiving keeps the 64 KB wall for the old whispered form. No whisper fallback for a larger one: which flavours deliver addon whispers is not established, and a real template is a few KB.

Peer Review 703dd2b8 P1: both send sites printed the guild/name line for every refusal, so a "toobig" told a player with a correct name to fix the name. sendRefused in GUI/TemplatesPage.lua now maps Send's second return: "toobig" names the size and the Copy this out box; "nomodule" says part of UGI did not load; "encode" and SP:Encode's own reasons (libs, malformed, serialize, compress) say the template could not be packed (no Copy this out pointer, since that box uses the same encoder). Only "nocomm", "notarget" and unknown reasons keep the name line. Specs in zz_templates_page_spec.lua drive both send sites, the toobig case through the real encoder with a template padded past 16 KB.

Fixed — UGI no longer pulls keyboard focus out of an open chat box (FGI_APICompat.lua, Modules/FGI_Dialog.lua, GUI/UI.lua, Modules/FGI_WhisperTabs.lua, GUI/Tabs/*.lua, Tests/zz_take_focus_spec.lua)

Danne (Discord, Forever): typing /ginvite <name> gave "Interface action failed because of an AddOn". Suspect, not confirmed by a taint log: UGI calling SetFocus on its own edit box while a chat box had focus with name suggestions attached, so the chat box's focus-lost handler (ChatFrameEditBox.lua:414-419 -> AutoComplete_HideIfAttachedTo, AutoComplete.lua:234-244, which writes AutoCompleteBox.parent) ran under UGI's taint. New addon.API.TakeFocus(box): it does nothing while a shown edit box in a DIFFERENT top-level window holds focus, and otherwise focuses as before (a box in the same UGI window still hands focus on, so moving between fields of one form works). Falls back to plain SetFocus on a client without GetCurrentKeyBoardFocus. Twelve call sites go through it: the dialog, the blacklist reason box, the copy popup (two), the whisper thread and its Whisper menu entry, and the New / Add new / Body / row-click paths of the Announce, Filters, Messages and Custom Scan tabs. COST: with one UGI window's box focused, an action in another UGI window no longer moves focus. 18 specs. Not verified in a client.

Fixed — changing the level range while a /who was still waiting ran the old range anyway (functions.lua, Modules/FGI_SettingsProfile.lua, Tests/zz_scan_reset_spec.lua)

fn.setLevelRange and a settings template emptied the query list at once, even with an answer outstanding. The late answer then advanced progress on the new list and, when it was capped, pushed the old range's split halves into it; with builtRange cleared the drift check never fired, so the old range ran on (WoW Forever answers take 30-40 s, so the window is real there). New fn.restartScanForRangeChange: with nothing in flight it resets as before; with a query in flight it leaves the run for the answer to land on and lets the next press's drift check (after the in-flight gate) restart it, marking a run saved before builtRange existed so the check still fires. Found by an agent reading the scan path for the filter-range todo item; not seen in a client.

Peer Review 797e1ede F1: the same edit then restarted or not depending on whether a /who was in flight -- the immediate path reset unconditionally (even when filters own the range and only the strip moved) and reset every Scan Groups bucket, while the deferred path restarted only on a real drift and rebuilt only the sweep's bucket. One rule now: fn.restartScanForRangeChange restarts only when fn.levelRangeDrifted() says the EFFECTIVE range changed, through the new fn.restartDriftedSweep, which fn:nextSearch's drift branch also calls. Two new specs.

New — an announcement log (Modules/Announce.lua, GUI/Tabs/Announce.lua, Tests/announce_log_spec.lua, Tests/zz_announce_log_window_spec.lua)

Vishiswaz (Discord threads 92580907 and 293a7840, 2026-11-18): "Announce log. Timestamped, showing who spammed what announcement and when", "Toggleable. Default: See everyone. An easy to access toggleable checkbox on the same window to filter to self only." A Log button beside Soundboard on the Announce tab opens a window listing, newest first, every post of a known announcement this client saw in guild, officer or public-channel chat: time, poster, announcement name, channel. Only my posts (off by default) narrows it to this character's lines; Clear log empties it.

  • Known means the guild's policy snapshots (rotation members included) and this player's own profiles, matched with the same fold the per-layer sightings use, so a guildmate's copy with a different guild link or colour still matches. A guildmate's personal announcement is unknown here and is not logged, nor is anything in a zone or layer this client was not in. Both said in the button's tooltip.
  • Stored in DB.factionrealm.announceLog, newest 300 kept, so alts on the realm see one log; "mine" is decided when read, with fn.isCommSelf against the reading character.
  • Cost: one fold of each guild/officer/channel line while any announcement exists (the text table is cached for 10 s, as perLayerTexts is). Not measured with /ugi perf.
  • Specs: 8 in Tests/announce_log_spec.lua (real chat events through the module), 4 in Tests/zz_announce_log_window_spec.lua (the window). Whole suite 4962 passed, 0 failed, 11 pending, default and reverse order. Not verified in a client.
  • The window's list is one LibAceGUIWidgets RowList (Time, Poster, Announcement, Channel) refilled with SetData (Peer Review 797e1ede F3): the first build released and re-created up to 300 AceGUI Labels on every logged line while it was open. New Time, Poster, Channel strings.

Fixed — Wingman opened WoW Forever's Group Finder Who list (Libs/LibWho/LibWho.lua, Modules/Wingman.lua, Tests/zz_wingman_who_pane_spec.lua)

MajinDrew (Discord, Forever): Wingman acts on mouse-down, so the click that closes the Group Finder sent a /who while its Who list was still shown, and the answer opened it again. New libWho:WhoPaneOpen() -- the visibility check BorrowWhoList already made, factored into whoHosts / anyPaneVisible so both read one copy -- and Wingman's scan step yields while it is true, falling through to invite/announce exactly as the foreignWhoOwed yield does. No timer. On every flavour the step now also gives way while the Social window is open on its Who tab. Four specs, including one proving the fixture reopens the window without the yield. Not verified in a client.

The yield said nothing, so a player who left the Social window on its Who tab got no scans and no reason (Peer Review 703dd2b8 P4). New whoPaneNotice() beside instancePauseNotice, in its shape: one chat line (L["wingmanMsgWhoPaneSkip"], client locale) on the first skipped search, re-armed by a click that finds no Who list open and by Wingman:Start. Four more specs in zz_wingman_who_pane_spec.lua.

Fixed — UGI's chat filters no longer hand a kept line's text back to Blizzard (functions.lua, Tests/zz_chat_filter_bare_false_spec.lua, Tests/zz_functions_spec.lua)

fn.hideWhisper, fn.hideAwayResponse and fn.hideNotFound answered a line they kept with false, select(3, ...). Every flavour's filter loop replaces the line's arguments with a filter's second return whenever it is truthy (Blizzard_ChatFrameBase/Shared/ChatFrameFilters.lua, ProcessFilters: classic_era, classic and classic_anniversary elseif newArg1 at :156; live and forever if results[2] at :165), so the line went on carrying copies made under UGI's taint, for no gain. They now answer a bare false, which leaves the line as it was. Part of DADDY's report (retail party chat lost in dungeons); these filters are on whisper, away and system lines, not party chat, so this is hardening rather than the fix for that report. A spec comment claimed a bare false blanks the line; no flavour's source on this machine does that, and that spec was rewritten to the bare false contract with the operator's approval (test change 8c9c92851f846cd5). Not verified in a client.

Fixed — the minimap coin is no longer cut at its edges in a square button (FGI_Core.lua, Tests/minimap_icon_spec.lua)

TOGTools (inbox thread 94c58f6e, 2026-11-17): Ultimate Tools' minimap button bag has a Square icons mode that drops LibDBIcon's ring and honours each addon's iconCoords, and the operator's screenshots showed the UGI coin with flat edges there. The crop was TOGProfessionMaster's, { 0.135, 0.825, 0.105, 0.795 }, fitted for the round slot where the ring hides the coin's rim; measured against our own art it cut about 1 px off the right and 2 px off the bottom of the coin. The new spec reads img/minimap/UGI_MMB_Icon.tga, finds the drawn pixels (x 10..53, y 7..52 of 64x64), and pins a square window holding the whole coin with at most 2 px of background on any side; the crop is now { 9/64, 55/64, 7/64, 53/64 }, the 46 px square on the coin. Method and tolerances follow TOGTools' Tests/icon_spec.lua. The coin draws about 4% smaller in the round minimap slot than before. Checked in a client by the operator on 2026-11-17, round minimap and Ultimate Tools' square bag both: "it does" look right.

Then TOGProfessionMaster (inbox thread 53ce750a, 2026-11-18): Ultimate Tools' square bag draws each icon FILE whole and ignores iconCoords, so UGI's coin (44x46 px in a 64x64 file) looked small beside UGR and UTr (56x58). The art is rescaled the way UPM's was: new dev-only tools/rescale-minimap-icon.lua maps the coin box (source edges x 10..54, y 7..53) onto x 4..60, y 2..60, bilinear with premultiplied alpha, written uncompressed 32-bit top-left. The coin is now drawn at x 4..59, y 2..58, the same box as UPM's; iconCoords is the 58 px square on it, { 3/64, 61/64, 2/64, 60/64 }. The tool must be run on the ORIGINAL art from git history, never its own output. Tests/minimap_icon_spec.lua passes unchanged at its own (tighter) limits. Not yet seen in a client, and the upscale has not been checked by eye for blur.

Fixed — UGI no longer hooks the chat box on retail (FGI_Core.lua, Tests/zz_chat_focus_release_spec.lua)

DADDY (Discord, retail): party chat went missing in dungeons. fn.hookChatFocusRelease put an OnHide ClearFocus on every chat edit box, so Blizzard's focus-lost handler (ChatFrameUtil.DeactivateChat) ran inside UGI's code; it was already skipped on WoW Forever for the same reason, and it was UGI's only code inside Blizzard's chat box on retail. It is now skipped on retail too. The top suspect, not confirmed by a taint log or in a client; DADDY was asked for one on 2026-11-08 and has not sent it. KNOWN COST, stated plainly: the v2.2.4 stale-focus fix this hook carried was WRITTEN FOR RETAIL, so it is now off on retail and Forever and runs only on the Classic flavours. On retail a UGI hotkey pressed right after typing in chat may type into the chat box instead, which is the symptom v2.2.4 fixed. Kept anyway because a lost party chat is the worse failure; revisit if a taint log clears the hook.

[v2.26.2] (2026-11-16) — announcers claim their turn before posting

Fixed — two announcers no longer post a guild announcement in the same second (Modules/Announce.lua, Modules/FGI_AnnounceSync.lua, Modules/FGI_Recruiter.lua)

Vishiswaz (Discord, 2026-11-15, Forever, both on v2.26.1 under Wingman): he and his officer posted the guild's ad, ticked All, to General in the same second. layerDueIn staggers each announcer's turn by LAYER_STAGGER (8 s) per place in the line-up, but that only says when a post MAY go out; Wingman posts on the next click after it, so an earlier announcer who clicked late could click in the same second as the next one, before either had seen the other's line. Guild chat had no stagger at all. Suspected from the code and the reporter's answers (two players, ticked All, Wingman, no fresh login or zone change, same version); not reproduced in a client. The operator, 2026-11-15: "we need to coordinate before announcements if the guild policy stuff is enabled. we can do it as urgent as they are small messages."

TURN CLAIMS: when a guild announcement ticked All comes due and the announcer line-up holds more than this character, the click claims the turn instead of posting: new sync kind K <id> <place> <rank> <name>, sent at ALERT priority (Sync:PushClaim), received into Announce:NotePeerClaim. A click at least CLAIM_SETTLE (3 s) later posts, unless a better claim for the same place arrived. Better means higher on the guild's announcer priority list, the line-up AnnouncerRank already staggers turns by (the operator, 2026-11-15: "isn't there a priority list already built in?"), so a higher announcer's claim wins even when it arrives later. Each claim carries its sender's OWN place in its own line-up, and the comparison reads only the carried (place, name): Peer Review a9133d0a H1 found that ranking a peer by the receiving client's line-up let two clients that disagree (each holding the baton as itself, or each not yet seeing the other online) both rank themselves first. A claim lapses after CLAIM_TTL (30 s), under the 60 s floor, so a claimant who stops clicking hands the turn on and no claim outlives its round; an announcer whose own claim lapses without its ad getting out (no echo at that place; for Guild/Officer chat, no post sent) sits out 30 s more before claiming that place again (M2: one whose posts never reach the channel stays due and would hold everyone off; R1: a pause on every lapse cost the announcer who posted fine ~33 s a round). While its claim lives, a client compares with the rank the claim carried, not its line-up's current one (R2). A client whose line-up holds only itself does not claim but still holds off for a claim it heard (M1). Guild and Officer chat claim under @<channel key>. Send re-reads the claim at the click, beside the seen-ad check, and drops the item when the player has changed zone since claiming (L1); the holdGuild top-up gather sends no Guild claim (L3). The claiming click prints new announceMsgClaiming, and a click held by another announcer's claim prints new announceMsgClaimHeld (L2); both enUS only, other locales fall back. New Recruiter:AnnouncerLineup() returns the line-up and this character's name in it; AnnouncerRank now reads it. Not claimed: the designated announcer on an announcement not ticked All, and a soundboard press that bypasses the time gates (FR-017). KNOWN COSTS: a claim slower than 3 s to arrive still lets two post; a client older than v2.26.2 neither claims nor honours claims, so every announcer on the list needs this version; and each post goes out one click and about 3 s after its turn. Specs in Tests/announce_turn_claim_spec.lua. Not verified in a client.

[v2.26.1] (2026-11-14) — rotation cooldowns across a handover, blacklist removals that reach guildmates, and Wingman's ON line

Fixed — Wingman's ON line no longer claims a level sweep when Scan is off (Modules/Wingman.lua)

Vishiswaz (Discord, 2026-11-14): with Wingman set to announce only, the ON message still read "level sweep 10-60". Wingman:PrintScope listed the active filters, the selected custom scans and the default sweep without asking whether the Scan step runs. All three now read doScan() first, and with Scan off the line ends at the steps ("Wingman ON · announce"), with no "your current filters" fallback either. Testing-mode still shows. Two specs in Tests/wingman_prompt_spec.lua.

Fixed — a guildmate's announcement post restarts the activity count (Modules/FGI_AnnounceSync.lua)

Vishiswaz (Discord, 2026-11-14): a guild-chat rotation on a 1-hour interval, ticked Guild and not All, posted Message 1 from him; he logged off three minutes later and his officer, now the designated announcer, posted Message 2 at once. Both on v2.26.0. A peer's post (S/T, and the A/B login catch-up) moved lastPosted and the cursor but left msgsSinceLastPost alone, so the activity bypass kept counting guild chat from the RECEIVING client's own last post, often hours of it. Once the 60 s floor passed, rotationDue (and IsEligible for a plain guild announcement) read the bypass as met. mergeOne and mergeRotation now zero the count whenever they advance a record. Two specs in Tests/announce_rotation_sync_spec.lua. That this was the cause depends on the rotation having an activity threshold set, which is asked of the reporter; not seen in a client. If it has none, the remaining suspect is the two PCs' clocks: S/T still carry the sender's time() (KNOWN COSTS in Modules/Announce.lua).

Fixed — a blacklist removal reaches guildmates at once, and closes their open kick prompt (functions.lua, Modules/FGI_DeltaSync.lua)

Vishiswaz (Discord, 2026-11-14): a blacklisted player joined, both officers got the kick prompt, one pressed Remove, and 23 s later the other's Remove posted a second "removed from the blacklist" officer line. An add was live-pushed (fn:blackList → PushName(BLACKLIST)) but fn:blacklistRemove left its tombstone to the next catch-up, so the guildmate's copy was still listed and their prompt still up. fn:blacklistRemove now live-pushes the tombstone under both stored spellings (PushName(TOMBSTONE, …); UGR entries stay local as before), and Bridge:ApplyEntries(TOMBSTONE) calls the new fn.recheckKickPrompt, which closes the FGI_BLACKLIST prompt when its head is no longer listed and moves to the next queued name. Two specs reversed ("removals sync by catch-up" in zzzz_blacklist_kick_popup_spec.lua and zzzz_blacklist_kick_displaced_spec.lua), two added for the receiving side; support/addon.lua's broadcast stub now records the sync item of each push.

[v2.26.0] (2026-11-12) — announcements every listed announcer can post, the invite box on a whispered phrase, and scans that follow your level range

Changed — the Scan timer goes up to 2 minutes on WoW Forever (functions.lua)

Danne (Discord, 2026-11-11, Forever) saw "Gave up on 1-60 after 13s" on about 80% of Scan clicks, with results arriving 30-40 s after the click: the server answered after the give-up deadline, past even the old 30 s ceiling. The operator, 2026-11-12: "make it go up to 2 mins for now, we may make it more". New fn.getGiveUpMax() (120 on Forever, 30 elsewhere, read per call) replaces the load-time addon.GIVEUP_MAX; fn.getGiveUpAfter and the Settings slider (GUI/SettingsPanel.lua) read it. With "Wait the full timer between scans" on (the default) the same number also spaces the player's own searches, so a long timer is a slow scan by the player's own choice. Spec in Tests/zz_settings_panel_spec.lua; Tests/zz_settings_profile_spec.lua lists the new read as a clamp, not a side effect.

Changed — changing the level range restarts the scan run in progress (functions.lua)

Jedric (Discord, 2026-11-11, Forever) set 10-30 and kept scanning 40-40: a run in progress keeps the queries it was built with, even across a reload (savedScan), and populatePassQueue refills only an empty list. The operator, 2026-11-12, asked whether a range change should reset it: "yes". New fn.setLevelRange(lo, hi) saves the range and, when it changed, calls fn.resetScanPosition; the Scan tab's wheels (GUI/Tabs/Scan.lua) and the legacy window's spinners (GUI/LegacyMainWindow.lua) go through it. A settings template writes the range by key, so SP:Apply (Modules/FGI_SettingsProfile.lua) calls fn.resetScanPosition when it changed either bound, and only then, so a template applied at every login keeps the saved run (Peer Review dd17d014). Specs in Tests/zz_settings_panel_spec.lua and Tests/zz_settings_profile_spec.lua.

A range set by filters (the level-range priority on filters, or an enforced guild filter) changes without going through either. The operator, 2026-11-12, asked whether editing a filter's level range should restart the run too: "yes". Rather than hook every writer (filter edits, a filter switched on or off, the priority setting, a guild filter arriving by sync), fn.freshPassQueue records the range the default sweep was built with (a third return value; the caller that installs the queue stores it in search.builtRange, so the Scan tab's preview of the next pass stays side-effect free), and fn:nextSearch calls new fn.levelRangeDrifted() before it fills the list: when the effective range (fn:getEffectiveLevelRange, falling back to the strip range) no longer matches, it calls fn.resetScanPosition and builds the new range. The built range rides in savedScan.range, so an edit made between sessions still restarts a resumed run. A pass of custom scans only records no range and never restarts this way, and with Scan Groups the check runs only while the standalone bucket (the one holding the sweep) is active, since ScanGroups.BuildBuckets builds every bucket at once and a group bucket has no range. There a drift calls new ScanGroups.RebuildStandalone(), which rebuilds and re-activates only that bucket, keeping the groups already run this pass, and rebuilds with the exclude set the bucket was built with (kept on the bucket) so a group changed mid-pass neither re-runs nor skips its scans; it is also checked the moment AdvanceBucket reaches that bucket, so no query from the old range goes out first (Peer Review 72d55727). Specs in Tests/zz_scan_reset_spec.lua.

New — "everyone on the list" for each guild-policy duty list (Modules/FGI_Recruiter.lua)

Vishiswaz (Discord, 2026-11-07) asked that every online announcer be prompted to turn Wingman on for zone-channel announcements, not only the top of the announcer list. The operator, 2026-11-10: "it's not only officers, anyone can be on that announcer list" and "we need a new check box for each of the 3 guild policy settings to turn it on for everyone on the list, and have it off by default".

Three policy fields, recruiterEveryoneOnList, inviterEveryoneOnList and welcomerEveryoneOnList (Recruiter.EVERYONE_KEYS), each a toggle under its own list in Settings > Guild policy, stored only when on so an older saved policy keeps its fingerprint. For the inviter and welcomer, on, amDesignateFor lets every available member the list covers act (listed, an alt of a listed account in this guild, or the assignee, through the election's own eligible gate), so CanInvite / CanWelcome open to all of them. The election (GetDesignated*) and the per-layer line-up (AnnouncerRank) are unchanged. The one-line automatic officer notice (CanPostAutoNotice, the fallOpen caller) ignores the setting and keeps one poster. New Recruiter:AmIOnDutyByList(role) answers "does the list put me on duty": false with no list, in the warm-up, and while the list falls open.

For the ANNOUNCER the setting is a master switch for a new All column on the Announce tab (policy editors only, beside Guild). The operator, 2026-11-10: "I simplified it with 2 states 1. Designated Announcer Only 2. Any announcer on the designated announcer list, IF that setting is enabled", "i don't care what channel its in", unticked by default ("they need to turn it on when they need it"). The tick is allAnnouncers on the personal profile, copied into the guild snapshot by setGuildAnnounceMark (GUI/Tabs/Announce.lua); a row not ticked Guild refuses it with a chat line, since the library's RowList cannot disable one cell. The toggle is AnnounceTab.SetAllAnnouncers, so Tests/zz_announce_policy_spec.lua pins it; Tests/policy_sync_spec.lua pins that the tick and the setting survive a peer's ApplyPolicy. Modules/Announce.lua: new allAnnouncersFor(profile) (tick AND Recruiter:EveryoneOnList("announcer")); publicHeld now holds a guild announcement for anyone but the designated announcer unless allAnnouncersFor, on EVERY channel, Guild and Officer included (a personal announcement keeps Guild/Officer ungated); the rotation readout and the round's per-place offer gate their rank the same way, and so do the zone timers (new turnRank, read by layerDueIn and PlaceTimers, so a member sees no turn they cannot take); HasPerLayerAnnouncement became HasAllAnnouncersAnnouncement (any channel). amDesignateFor does not widen the announcer role, so CanPostPublic stays the designated announcer's. Wingman's duty prompt (onDutyByList) offers a listed non-designated announcer Wingman only when some guild announcement is ticked All. KNOWN COST: since v2.16.0 every listed announcer posted every guild announcement in public channels with no switch; after this update that needs the tick plus the setting, so guilds relying on it go back to the designated announcer until an officer ticks. A guild announcement's Guild-chat posts, ungated until now, become the designated announcer's unless ticked. Mixed versions do NOT fail safe in public channels: a client older than v2.26.0 ignores the tick and keeps the v2.16.0 rule, so a listed announcer on an old build still posts every guild announcement in public channels, ticked or not (doc: the v2.16.0 rule this entry replaces; not verified in a client). FOR THE PLAYER NOTES: say that every member on the announcer list needs v2.26.0 for the All column to take effect. Tests/announce_per_layer_spec.lua fixtures carry the tick and the setting; four examples that pinned "no switch" or ungated Guild chat were rewritten, each with the reason. For the welcomer, the guild-chat stand-down in Modules/FGI_Welcome.lua (peerWelcomed) still drops a name a peer has just welcomed, so a guild welcome normally goes out once; the tooltip says so (Peer Review W1, thread 9ca40714). Specs in Tests/recruiter_spec.lua ("everyone on the list"). Not verified in a client.

New — on WoW Forever, a whispered invite phrase opens the /ginvite box (Modules/FGI_PhraseInvite.lua)

The operator read WOWForeverGuildInviter (Wreck), a Forever inviter addon that queues a player who whispers an exact phrase. It found no way past Forever's invite block: every one of its invites calls C_GuildInfo.Invite and raises ADDON_ACTION_FORBIDDEN, its right-click "Invite to Guild" included (an entry it adds itself with Menu.ModifyMenu; Forever's stock player menu has no guild invite -- Blizzard_UnitPopup/Mainline/UnitPopupMenus.lua). The operator, 2026-11-06: "there are plenty of addons that do that, whisper on invite is the one i use in classic. we could do that and have it pop up the copy box as another optional feature."

Two settings under Settings > General > Scanning, beside "Open the /ginvite box after each whisper" and hidden wherever addons can invite: "Open the /ginvite box when someone whispers an invite phrase" (DB.global.phraseInvite, off by default) and "Invite phrases" (DB.global.phraseInviteList, one per line, default ginv and guild inv). A whisper counts when it CONTAINS a phrase, in any case, as plain text rather than a Lua pattern. It was whole-message at first; the operator, same day: "it doesn't have to be excactly the phrase, but incliudes it? like if my phrase is inv and the user types invite me, it should fire". The "Invite phrases" tooltip warns that a short phrase also matches inside other words. A sender gets one box per session (PI.opened), so a phrase whispered again or spammed does not reopen it (self-audit F1, Peer Review fc522293). The box opens through addon.ChatMenu.GuildInvite, the right-click Guild Invite's own path, so a blacklisted sender and a reached recruitment goal are refused there and the invitedForAccept stamp credits a join from the pasted line. No box for a guildmate, without a guild, with Let UGI manage whispers off, or in the first ten seconds after login (replayed whispers, as Modules/WhisperAlert.lua). Both keys are in the settings-template catalogue. New Tests/zz_phrase_invite_spec.lua; the module is in all seven TOCs and in Tests/support/addon.lua. Not tried in game.

Fixed — a combined welcome joins names in the game's language, not always in English (Locale/*.lua)

Adrenalin (Discord, 2026-11-10): "the 'and' isn't affected by different languages." A combined guild welcome joins names with welcomeNameSeparator, welcomeNameFinalJoin and welcomeNameSerialJoin, read from the CLIENT locale (fn.chatLoc), and only Locale/enUS.lua defined them, so every client posted "A, B, and C". The ten non-English game-client languages now define all three: deDE " und ", esES/esMX " y ", frFR " et ", itIT/ptBR " e ", ruRU " и ", koKR " 및 " (the list "and", which does not change with the name's final sound), and zhCN/zhTW the enumeration comma "、" with "和". None uses a serial comma. English keeps "A, B, and C", the serial comma added in v2.14.5 at Vishiswaz's request. The UI-only locales (Arabic, Hindi and the rest) are not client languages, so chat never reads them. New Tests/zz_welcome_joiners_locale_spec.lua. Not seen in a client.

Changed — /ugi announce, /ugi invite and /ugi nextsearch send an owed welcome first (FGI_Core.lua)

Adrenalin (Discord, 2026-11-10): "How can i send the guild chat welcome after someone joined with a macro (not wingman)? /ugi announcement doesn't work." A welcome (and a decline whisper) waits in the click pump until a click drains it (fn.pumpClick, functions.lua). The keybind buttons and the tray's buttons all call it first; the three slash commands went straight to their action, so a macro built on them could never send a welcome. Each now calls fn.pumpClick() before acting, as FGI_BindBtn_* do. A typed or macro'd slash command is the player's own key or button press. Also corrects the /ugi invite comment, which said the keybind routes through the slash (it has used its own button since v2.2.3). New examples in Tests/zz_slash_misc_spec.lua. Not seen in a client. Not explained: Adrenalin says the horn did not send the welcome while the announcement picker beside it did; both call the pump (Modules/compactFrame.lua :772-774 and :1254-1256), so that is not reproduced (not verified whether the welcome's delay had passed).

Changed — the CurseForge page says near the top that UGI needs companion addons (docs/Curseforge_Description.html)

Opakrake_TTV (Discord, 2026-11-08, WoW Forever) only learned UGI needs Ace3 and its libraries from the game's addon list after a restart. The page's Requirements section listed them all, but at the bottom, after the slash commands. A one-line note under the title now says companion addons are needed and points to Requirements for manual installs and WoW Forever.

Changed — the whisper tray's reply box lets go of the keyboard after a sent reply (Modules/FGI_WhisperTabs.lua)

rocky (Discord, 2026-11-08): "perhaps mimic blizzard behavior by sending focus out of the textbox so you don't put wwwwww in the textbox after hitting enter". The reply box's Enter sent the line and cleared it but kept focus, so the next movement keys typed into it. A sent reply now also calls ClearFocus, as Blizzard's chat box closes after Enter; a refused send keeps focus and the text. The existing example in Tests/zz_whisper_tabs_spec.lua now asserts both. Not seen in a client.

Changed — the main window's Push policy icon glows while there are unpushed changes (GUI/MainWindow.lua)

Vishiswaz (Discord, 2026-11-07): "I should get a glow to see it without having to mouse over and check". The icon on the Announce, Filters and Eligibility tabs said so only in its tooltip; Settings already showed a red asterisk and a warning line. MainWindow:RefreshPushIcon now shows a pulsing red border (an ADD-blend UI-ActionButton-Border with a looping Alpha animation) while addon.IsPolicyDirty() holds and the icon is shown. It refreshes on tab switch, window open, guild-data events, and now also on AceConfigRegistry's ConfigTableChange for UGI, which a push, a received policy, a Guild tick and the Settings clear-list buttons fire. A plain value change inside Settings does not fire it (AceConfigDialog fires it only on group selection), so that case shows on the next tab switch. New examples in Tests/zz_mainwindow_handlers_spec.lua. Not seen in a client.

Fixed — a guild rotation no longer posts inside its cooldown for a guildmate who just logged in (functions.lua, Modules/FGI_AnnounceSync.lua, Modules/FGI_Baton.lua)

Vishiswaz (Discord, 2026-11-07): a guild-chat rotation on a 1-hour cooldown posted at 4:14 from him and at 4:59 from Gummi, who was offline at 4:14 and logged in afterwards. A client that logs in learns the guild's last posts from the login catch-up (R, answered by A/B/C whispers), and that catch-up fired on a fixed 8-second timer after PLAYER_LOGIN. Every reply passes the trust gate (fn.isCommTrusted -> Bridge:IsTrustedSender -> LibGuildRoster IsInGuild), which during the roster-building window can read a real guildmate as a stranger, so the replies were dropped and nothing asked again. The DeltaSync bridge had already moved its own catch-up off a timer for exactly this (Modules/FGI_DeltaSync.lua, "The roster-building window"); these two had not. New fn.whenRosterReady(owner, callback) runs the callback when LibGuildRoster reports OnRosterReady (at once if it already has) and keeps listening, because the library fires it again for a new guild after a guild change (Peer Review fc522293, R2), and both modules' login step (Sync:LoginCatchup) now waits on it; the baton's re-assert of a permanent opt-out moves with it. A module loaded without functions.lua (only the specs that load one file alone) skips the catch-up. New Tests/zz_when_roster_ready_spec.lua. That this was the cause is our best reading of the report, not seen in a client: the policy did reach both guildmates (their screenshots show all three guild announcements and the same countdown).

Changed — on WoW Forever the Mode tooltip says why Whisper only is the only mode (GUI/UI.lua, GUI/Tabs/Scan.lua)

Players kept asking on Discord where the other invite modes went (Perplexion, 2026-11-07, among others), and Vishiswaz suggested the tooltip answer it. New UI.InviteModeTooltip(allModesTip) returns the new inviteModeTipForever string on Forever (Blizzard blocks addon guild invites there; invite whoever replies with /ginvite or right-click Guild Invite; whisper messages are on the Messages tab) and the all-modes text everywhere else. The Scan tab's Mode hover reads it. English only for now: other locales fall back to the English line. New examples in Tests/zz_forever_invite_mode_spec.lua.

Fixed — a whisper with GUILDLINK is no longer refused on clients that have no Guild Finder (Modules/FGI_Placeholders.lua)

Found answering Adrenalin (Discord, 2026-11-07, WoW Forever). fn:msgMod printed "Unable to create guild link" and returned nil whenever no link resolved, so on a client that can never make one (no club-finder API on the Classic flavours; WoW Forever has the API but C_ClubFinder.IsEnabled() is false) every whisper whose message carried GUILDLINK was dropped, on every send. The broadcast path already fell back to <Guild Name>. New fn.guildLinkPossible() answers whether a link can be made at all; where it cannot, the whisper path now puts the guild's name in GUILDLINK's place (with no-break spaces, as GUILD does) and the byte estimator counts it the same way (the no-break spaces only since the follow-up below). The refusal is kept where a link is possible but not cached yet (retail before the listing loads), since opening the guild window fixes it. resolveGuildLink now also skips the club-finder call on Forever. GUILDLINK is parked on a sentinel across the GUILD pass in fn:msgMod too, so the fallback name is not rescanned. My example in Tests/zz_placeholders_guildlink_spec.lua asserted the refusal on a client with no club-finder API, which pinned the defect; it is rewritten, with new examples for Forever, the fallback and the estimator. C_ClubFinder is added to .luarc.json. Not seen in a client.

Changed — the /who debug output names the Who windows each search holds shut (Libs/LibWho/LibWho.lua)

MajinDrew (Discord, 2026-11-07, WoW Forever, v2.25.0): under Wingman the Group Finder's Who list opens on every click showing the results of UGI's own search, while the debug log shows UGI claiming that same answer; scans from the >> button do not open it. So the Group Finder heard an answer BorrowWhoList should have kept from it, and nothing UGI logged said which hosts the borrow held. With /fgi debug on, each search now prints how many Who hosts it held, their names, and whether a Who list was open. Diagnostic only; the cause is not found yet.

Fixed — joining a guild mid-session now syncs its lists (Modules/FGI_DeltaSync.lua)

Self-audit F2 (Peer Review fc522293): three pieces of code waited for the guild roster. The DeltaSync bridge kept its own registration (Bridge:ArmRosterReady) and ran its catch-up only on the first build, behind a catchUpArmed flag. LibGuildRoster builds again after a guild change, so a player who left a guild and joined another in one session never reconciled the new guild's invite, blacklist and policy lists until the next login. ArmRosterReady now goes through fn.whenRosterReady, and Bridge:OnRosterReady runs RunCatchUp on every build (it keeps its own guards: initialised, sync on, in a guild); the flag is gone. Held packets still replay after it, through the same trust gate. My example in Tests/zz_deltasync_bridge_spec.lua asserted the opposite ("a rebuild must not run a catch-up nobody armed") and is rewritten to the new contract. Tests/policy_sync_spec.lua, which loads the bridge without functions.lua, stands in the helper's contract. Checked and left alone: Recruiter:WarmupRemaining asks "is the roster built now?" for the designate hold and is not a wait, and FGI_AnnounceSync needs no hold for live S/T/L packets dropped before the build, because its catch-up R goes out after the build and the replies carry every cooldown still running. The baton's catch-up repeats cleanly on a second build (ReassertNeverOnDuty does nothing while already stepped aside). Not seen in a client.

Fixed — follow-ups from Peer Review on self-audit 1930a360 (functions.lua, Modules/FGI_DeltaSync.lua, Modules/FGI_Placeholders.lua, GUI/UI.lua)

  • F4, a catch-up broadcast on every same-guild rebuild. LibGuildRoster resets and rebuilds not only on a guild change but whenever IsInGuild() reads false (OnPlayerGuildUpdate and OnGuildRosterUpdate, LibGuildRoster-1.0.lua:2336-2343 and :2365-2366, read by Peer Review), so the every-build rule above would re-broadcast the item hash list after any momentary false. fn.whenRosterReady now runs its callback once per guild, by GR:GetHomeGuildKey(), remembered per owner so a re-arm does not forget it; an unknown key always runs. Its new fresh argument starts the memory over, and only Bridge:ScheduleCatchUp (the login catch-up) passes it. Held packets are replayed on EVERY build through their own registration (Bridge:ReplayHeld), since a once-per-guild wait would strand packets held by exactly such a rebuild. Whether IsInGuild() really flickers false on any client is not verified. ReplayHeld resolves the guild scope itself before judging, because on a same-guild rebuild it runs from its own registration and CallbackHandler promises no order against GuildScope's (Peer Review, thread a1278738, L1).
  • F1, the whisper byte estimate. fn:estimateSentBytes ran a second GUILD pass after restoring the link, rewriting the word GUILD inside a cached link or a guild name, and counted the guild name's spaces as one byte where fn:msgMod sends the 2-byte no-break space. It now makes fn:msgMod's single pass. The estimator examples compare against fn:msgMod's output instead of hand counts, with a new one for a link containing GUILD.
  • F2. UI.InviteModeTooltip drops its or allModesTip: FGI's locale table answers a missing key with the key string (FGI_Constants.lua, the key-fallback metatable), so the or could never run.
  • Tests/zz_when_roster_ready_spec.lua, Tests/zz_deltasync_bridge_spec.lua (a same-guild rebuild does not broadcast; a new guild does) and Tests/policy_sync_spec.lua's stand-in follow the new contract. Not seen in a client.

Fixed — on WoW Forever a /ginvite you type is no longer blocked as an addon's (FGI_Core.lua)

Danne (Discord, 2026-11-08, WoW Forever): with UGI enabled, typing /ginvite raised the client's "an addon blocked an action" pop-up with no Lua error; disabling UGI fixed it. Since v2.2.4 UGI hooks every chat edit box's OnHide and calls ClearFocus there, to release a stale focus claim that stopped the hotkeys working after typing in chat. The hook runs as UGI's code, so the ClearFocus runs Blizzard's OnEditFocusLost inside UGI's taint, and that calls ChatFrameUtil.DeactivateChat (forever Blizzard_ChatFrameBase/Shared/ChatFrameEditBox.lua:414-420), which rewrites the chat box's active-window state with the taint on it. Forever protects guild invites (the operator's own test: a line put into chat by code is blocked, one the player types goes through), so the player's next typed /ginvite inherits the taint. The install moves into fn.hookChatFocusRelease, still run at file load, and returns at once on Forever. New Tests/zz_chat_focus_release_spec.lua. That this hook is the cause is our best reading, not seen in a client. KNOWN COST: on Forever the stale focus claim can come back, so a UGI hotkey pressed straight after typing in chat may type into the chat box.

Fixed — the chat guard no longer compares a chat value before asking whether it may (FGI_APICompat.lua)

Found investigating DADDY's report (Discord, 2026-11-08, retail): party chat stops showing in the chat frame once a dungeon starts, only with UGI enabled. API.RegisterChatEvents' wrapper asks canaccessvalue of every CHAT_MSG_* argument, but first tested each with v ~= nil. The Warcraft Wiki's Secret Values page: "Tainted code is not allowed to compare or perform boolean tests on secret values", an "immediate Lua error", while "type(secret) returns its real type"; and a chat payload is secret not only in lockdown but "when the player is on a communication-restricted map such as a dungeon or raid". The test is now type(v) ~= "nil". This is NOT shown to be the cause of DADDY's report: an error in our own handler should not stop Blizzard's chat from showing a line, and the taint log asked of him is still what will name the culprit. Not seen in a client.

Fixed — a refused baton message no longer shows a BugSack error (Modules/FGI_Baton.lua)

A player (Vishiswaz, Discord, 2026-11-07) got AceCommQueue-1.0's one-time "message NOT delivered ... no callbackFn was supplied (prefix=FGIBaton queue=FGIBaton GUILD)" error. AceCommQueue reported it twice to our inbox (threads 9eed15d0 and a2e9b100). send() passed onSendResult only when a verdict context was given, so the catch-up request R, the role request Q and the S/P catch-up replies went out with no callback, and on a refusal the library reported the drop itself, by design. My own example in Tests/baton_roundtrip_spec.lua asserted that R carried no callback, so the suite pinned the defect. Every send now passes onSendResult; a send with no context is counted in Sync.quietRefusals and otherwise ignored, since each is recovered by the next login or broadcast. Also fixed: a refused assignment (G) told the player it was a "step back"; it now says "assignment". Which refusal the client gave is not known: AceComm drops the result code.

[v2.25.0] (2026-11-06) — the WoW Forever invite box on your terms, and editor rows that no longer overlap

Changed — on WoW Forever the Scan list's invite no longer opens the /ginvite box unless you ask for it

v2.24.0 opened the copy box after every Scan-list whisper on Forever (fn:invitePlayer, invite type 3), and six Discord threads from five players asked for it to stop (TheThiccExpress, Jedric twice, Pockit, Akranaz twice). The operator, 2026-11-06: "that should be off by default and the user has to turn it on through a setting. most folks are not going to want that box to pop up every whisper", and "we need to leave the 'manula' invite that opens the box alone, that should always work. this is only for popping the box up on a sent invite whisper." New DB.global.foreverInviteBox, default false, in Settings > General > Scanning as "Open the /ginvite box after each whisper", hidden wherever addons can invite (API.AddonGuildInviteAllowed). It gates only the box after the Scan list's whisper, which covers a row's Invite, the Invite button, the compact window and Wingman. The right-click Guild Invite (ChatMenu.GuildInvite) and the API.GuildInvite fallback still open the box every time. Carried by settings templates in the scan group. Tests/zz_chat_menu_spec.lua: the default is off, unset and off whisper with no box, on whispers then opens it; the chat-menu examples for the right-click box are unchanged and pass. Not checked in a client.

Changed — the /ginvite copy box opens where you last dragged it

Akranaz on Discord (2026-11-06): the box always opened in the middle of the screen. Every box is a fresh pooled AceGUI Frame, and a fresh Frame is centred (AceGUIContainer-Frame.lua ApplyStatus), so a drag never stuck. The operator: "yes, it should remember where it was dragged, we need to add that to the SV file as well so it lives through /reload and exit/enter". addon.Dialog takes a new optional spec field position (a key); on close it saves the frame's top and left to DB.char.frames.dialogPositions[key], and the next dialog with that key opens there, after its size is fitted. Only top and left are kept, not AceGUI's status table, because that also restores a width and height and this dialog sizes itself. Saved on close rather than by hooking the title bar, because the Frame is pooled and a hook would follow it into another addon's window. A spot off the screen (a smaller resolution or UI scale since) is ignored. UI.ShowCopyLine passes copyLine, so the /ginvite and /gremove boxes share one spot, and /ugi resetWindowsPos clears it with the main window's. Specs in Tests/zz_ui_helpers_spec.lua and Tests/zzzz_core_spec.lua. Not checked in a client.

Changed — a template the recipient drops now says why, with debug on

Vishiswaz on Discord (2026-11-06, WoW Forever): sending a settings template printed "Sending the template ..." and "Sent the template November 6 2026 to Gumshots Snowbound" on the sender (screenshot), and the recipient, in the same guild on the same version, saw nothing. Wire:OnCommReceived drops a template silently on six branches (from ourselves, sender not a guildmate, unknown verb, malformed, too big, undecodable), so nothing could say which fired. With /fgi debug on, each now prints a [Template] line, and the arrival itself is logged; with debug off a stranger still gets silence, so the drop cannot be turned into a chat line anyone can trigger. New examples in Tests/zz_template_wire_spec.lua: the debug lines, and a WoW Forever round trip (regional names on, "First Last" names, a target typed bare, lower-cased or with a realm, an AceComm sender carrying a realm). The Forever round trip passes, so FGI's own addressing and trust checks deliver there offline; the cause is not found. Not checked in a client.

Fixed — the Announce editor's New button covered the Act box at narrower window sizes

Vishiswaz on Discord (2026-11-06, screenshot on v2.24.0). Row 1 of the Announce editor (Type, Name, Channels, CD, Act) was chained left to right at fixed widths while New and Save hang off the right edge. v2.23.0 added New beside Save without raising AnnounceTab.MIN_WIDTH (840), so near the floor the two ends met and New sat on the Act box. Row 1 is now anchored from both ends: Act, CD and Channels hang off New, and the Name box stretches between the Type dropdown and Channels, so nothing can run under the buttons at any width. MIN_WIDTH goes to 910 so the Name box keeps at least 100 units. New Tests/zz_announce_strip_layout_spec.lua, the Announce twin of the Scan strip's layout spec, fails on the old anchors (New over Act by 40px) and passes on the new ones. It renders the tab 56 units narrower than MIN_WIDTH, the main window's border and tab group derived from Ace3 (Frame content inset 17 each side, TabGroup border 1 and content 10; Peer Review traced it on thread c9c78c91). The first cut used 80, read off a 1.37-scale screenshot in screen pixels. The Profile Name label now follows the stretched name box's right edge instead of keeping 140 wide, which could run its hover box under the Channels label. The idle hint over the editor ("Press New to write an announcement...") is the v2.23.0 editor cover and is drawn over the fields on purpose. Not checked in a client.

Fixed — the Filters editor's Min RIO M+ label ran under the Raid label

Checking the other two tabs that gained a New button in v2.23.0. New Tests/zz_editor_strip_layout_spec.lua renders the Filters and Messages strips at their minimum width and at 1400 and asserts no two controls in a row overlap. New does not squeeze anything on either tab. It did find, on Filters, that the "Min RIO M+" label is 30 wider than its box so the text fits, while the Raid box started only 14 past it, so the label's hover box sat 16px under the Raid label at every width. The Raid box now starts 32 past the RIO box. The spec tolerates 10px of overlap for the deliberate dropdown anchors (Filters' race dropdown 10px into the class one, Messages' kind dropdown 8px into the name box), which cancel the dropdown template's empty margin. Not checked in a client.

[v2.24.0] (2026-11-05) — every zone's announcement timer, the WoW Forever invite box from the Scan list, and the compact window following the text scale

Fixed — a guild announcement read "Ready" for a member whose clicks could not post it

Discord, Vishiswaz (2026-11-05), with a screenshot: "General with no needs (guild)" read Ready on the Announce tab, still Ready straight after a /reload with five minutes left on it, and "the actual announcement doesn't seem to be going out with wingman". The round (gatherEligible) holds a public channel for a member who is neither the designated announcer (Recruiter:CanPostPublic) nor in the per-layer line-up (Recruiter:AnnouncerRank nil -- which includes the login warm-up after every reload). The readouts did not apply that gate: IsEligible's per-layer branch needs a rank, so without one it fell back to the guild-wide record, which a per-layer post never stamps, and answered Ready. So the tab, both horn tooltips and Wingman's GetNextReadyIn all said there was work while every click posted nothing.

  • New publicHeld(profile, channelKey, canPublic, rank) in Modules/Announce.lua is the one rule; gatherEligible uses it in place of its inline test, and plainReadiness and RotationStatus now skip the channels it holds. A profile whose every ticked channel is held returns a new held flag (5th return of plainReadiness, 4th of ProfileReadyIn, 7th of RotationStatus, and held on each GetStatus row).
  • The Announce tab, StatusReadout (both horns) render it as "another announcer's turn" through the new Announce:HeldText(); new locale key in Locale/enUS.lua, other locales fall back to it.
  • Three examples in Tests/announce_per_layer_spec.lua.
  • NOT THE CAUSE OF THIS REPORT, or not all of it: Vishiswaz replied that he is the guild's #1 announcer, and the warm-up ends as soon as the roster is ready, so this mismatch covers at most the first seconds after a reload. The gap it closes is real for every member off the line-up; why a #1 announcer's General row reads Ready while Wingman posts nothing is still open. Not checked in a client.

Fixed — a Guild post held for a welcome could starve every other channel

Found by reading while chasing Vishiswaz's report above; whether it is his cause is not known. While a guild welcome is owed, Wingman passes holdGuild and Announce:Send posts the round's non-Guild items first. A round down to its held Guild items returned "held-for-welcome" without finishing, and a new round is only gathered once the old one has finished -- so a General (or any other) post that came due meanwhile was not gathered until the welcome went out. Send now tops a held round up with what gatherEligible finds due on the other channels; the held Guild items stay as they were, so nothing is posted twice. New example in Tests/zz_wingman_welcome_priority_spec.lua (not run against the old code to watch it fail).

Changed — Wingman's auto-stop asks Announce who may post on a public channel

Wingman:OffDutyUnderPolicy worked out "may I post on a public channel" itself (CanPostPublic, or HasPerLayerAnnouncement with an AnnouncerRank), beside the round's own per-channel publicHeld rule in Modules/Announce.lua. Two copies of one rule, so a change to the per-layer turn could leave Wingman stopping an announcer the round still posts for, or running for one it holds. The rule now lives in Announce:CanPostPublicSomewhere, built from the same two inputs publicHeld reads, and Wingman calls it. No behaviour change. New example in Tests/announce_per_layer_spec.lua checks the answer against what Send posts; Tests/wingman_prompt_spec.lua's stand-in now answers the new method. Self-audit finding F1, Peer Review thread 643b528c.

Fixed — an announcer out of a guild announcement's channel was told it was another announcer's turn

publicHeld decided "this public channel is mine by the per-layer turn" through layerPlace, which needs the channel joined. So a non-designated announcer standing outside General saw "another announcer's turn" on the tab and horn, and Send reported the round as held, when the real reason was not being in General. It now asks only whether the announcement is the guild's (isPolicyAnnounceId), and IsEligible reports the unjoined channel as "not-joined", so the readout says "not in that channel". Rotations already did this. The designated announcer was never affected, so this does not explain Vishiswaz's "Ready but nothing posts". New example in Tests/announce_per_layer_spec.lua, which fails with the old line restored (checked: 72 passed, 1 failed, that example). Raised by Peer Review, thread 643b528c.

New — hover a guild announcement's Status to see every zone's timer

Discord, Vishiswaz (2026-11-05): "If there's a timer, there should be a way to see its status. Period." A guild announcement on a public channel is timed per place (each zone's General is its own channel), but the Status cell shows only the place the player stands in. New Announce:PlaceTimers(profile, now) lists every place this session holds a sighting for, read from the same sightings / layerDueIn the cell uses, each marked here when the player stands in it. The Announce tab's Status column shows it on hover (onCellEnter, LibAceGUIWidgets RowList MINOR 30) in his own shape (same thread): "tooltip shows "Current Zone (Hold Shift to see all)" and if you hold shift it will show the cooldowns for all zones not currently ready". A MODIFIER_STATE_CHANGED frame repaints the open hover when Shift goes down or up. Personal announcements keep one timer per channel and show no hover. Places are known only once seen or posted this session or restored from the last one, so a zone nobody has posted in is not listed. Two examples in Tests/announce_per_layer_spec.lua. Not checked in a client.

Fixed — the text and widget scale now reaches the compact window and its announcement picker

Discord, Pockit (2026-11-05): "Is the text scaling meant to work w/ th ecompact window? If so, it sadly does not". The slider is LibAceGUIWidgets' scale, which grows only what the library draws; the tray (Modules/compactFrame.lua) and the soundboard picker are plain frames with fixed sizes. Both are now frame:SetScaled to W:GetScale() -- the library's case against scaling a whole window (a big window leaving the screen, pixel-fitted columns) does not fit two small fixed-layout bars -- through W:OnScaleChanged, keeping each frame's top-left corner where it is on screen (points are in the frame's own scaled units, so they are re-anchored at offset * old / new and re-saved). Before the login restore only the scale is set, so the library's ADDON_LOADED restore cannot save the file-load centre over the player's position. KNOWN COST: on a library older than MINOR 49 the scale arrives at PLAYER_LOGIN, possibly after the tray's restore, and the first restore of a session can then sit off by the scale factor (clamped to the screen). No spec: whether the harness frame model scales GetLeft/GetTop was not checked. Not checked in a client.

Changed — WoW Forever: the Scan list's invite opens the /ginvite copy box too

Discord, AyoAdrian (2026-11-05), then the operator: "of course it should, any invite function in Forever needs to pop open the copy box". On Forever the Scan list runs only Whisper only, and until now that sent the whisper and stopped, so a recruit found by a scan could be invited only from a chat reply. fn:invitePlayer's Whisper only branch now opens the same one-line box after the whisper when API.AddonGuildInviteAllowed() is false; invitedForAccept was already stamped there, so a join from the pasted line is credited. That covers the per-row Invite, the Invite button, the compact tray's invite and Wingman's invite step, which all go through it. The box itself moved to one function, ChatMenu.ShowInviteLine, used by the chat menu as well. API.GuildInvite, the guard for any direct caller, now opens that box too instead of printing, keeping the chat line only for a load with no box. KNOWN COST: with Wingman driving invites on Forever, every invite click opens a box to paste. Specs: a new example in Tests/zz_chat_menu_spec.lua, and Tests/zz_apicompat_guild_spec.lua's Forever example split in two for the box and the no-box line. The box and the pasted /ginvite line are the ones the operator confirmed in a Forever client for v2.23.0; this change only opens them from more places.

New — the scan queue's copy popup has a "/ginvite lines" tab

Allrithin on Discord (2026-11-05), then the operator: "just make it another tab on that popup window, so you have the list as is, and one with the /guildinvite problem solved". UI.GetCopyTextForTab("scan") now returns a third value, two views of the queue: the bare names as before, and each name as /ginvite <name> spelled through fn:formatNameForRetailAPI, as the one-line invite box spells it. UI.ShowCopyAllPopup(title, text, tabs) shows two or more views as buttons above the box, the selected one disabled, and hides them again for a single-view list; the bottom-row Copy icon passes the views through. Every flavour gets the tab, since /ginvite works on all of them. KNOWN COSTS: a line pasted from this list stamps nothing, so a recruit who joins from it is not credited as an accepted invite, unlike the one-line box; and whether the chat box sends more than the first line of a multi-line paste is not known. Specs in Tests/zz_ui_helpers_spec.lua. The operator confirmed both tabs in a client (2026-11-05, screenshot: "validated it's working").

Fixed — the Addon compartment switch turned itself back on

Found by Ultimate Guild Roster (inbox thread f136a043) while building its own minimap button from FGI's. Retail and WoW Forever only. Turning Settings > General > Appearance > "Addon compartment button" off called LibDBIcon's RemoveButtonFromCompartment, which sets DB.global.minimap.showInCompartment to nil (Libs/LibDBIcon-1.0/LibDBIcon-1.0.lua:673); the AceDB default is true (FGI_Core.lua:611), so the switch read back as on and the entry came back at the next login. applyCompartment (GUI/SettingsPanel.lua) now writes false after the removal. The stand-in library in Tests/zz_settings_panel_spec.lua now clears the key the way the real one does, and the example checks the switch stays off. Not checked in a client.

Fixed — the Scan tab's Invite mode tooltip showed a setting name and the wrong scope

Found answering Avxryj on Discord (2026-11-05). The tooltip's last line read "Per-character (DB.global.inviteType).": an internal name shown to players, and wrong, since DB.global is account-wide. The string is a locale key carried by ~30 Locale/*.lua files, every translation ending with the same name, so GUI/Tabs/Scan.lua now cuts the line holding DB.global.inviteType at display time; the key and every translation stay matched. Not checked in a client.

Changed — the text and widget scale defers to the one LibAceGUIWidgets saves

LibAceGUIWidgets (inbox thread 01be743b, unreleased MINOR 49) now saves the suite's single UI scale itself. fn.applyUIScale(atLoad) skips the login call when W:HasSavedScale() is true, so UGI no longer sets the size for every suite addon by loading last; the slider and a settings-profile import still apply. On an older library HasSavedScale is nil and nothing changes. New example in Tests/zz_window_backdrop_spec.lua. The library then made the scale per addon by default (MINOR 50, same thread): HasSavedScale is true only while the player has typed /lagw scale shared, so by default UGI applies its own saved size at login exactly as before. The slider shows UGI's own saved size first and the library's live value only when none is saved; a draft that read W:GetScale() first was reverted for this.

[v2.23.0] (2026-11-05) — WoW Forever invites and kicks through a copy box, choose your main, guild policy as text, and editors that say what you are doing

Changed — WoW Forever: every Guild Invite opens "/ginvite Name" to copy, in a small copy box

Asked by Guz on Discord (thread 2ef34cfb). The operator tested it first ("lets do a test, have the right click UGI invite selection have that copy popup") and a /ginvite line pasted into chat with Ctrl+V and sent by the player went through on Forever (2026-11-05: "this works! finally, something we can do!!!"), where every route an addon drives was refused. So the v2.22.0 greyed-out entries are live again:

  • ChatMenu.GuildInvite, on Forever, opens /ginvite <name> (the name through fn:formatNameForRetailAPI) to copy instead of calling API.GuildInvite. Its refusals run first (blacklist, recruitment goal, testing mode). Nobody knows whether the line is sent, so it records no attempt and no pending invite, only the invitedForAccept stamp: a recruit who joins from the pasted line is credited as accepted, and an unsent line's stamp expires with the retention window (Peer Review ba763bea C1, which found the join going uncounted). Every entry calls it, so the chat menu, the legacy dropdown (addon.InitMenu), the whisper-tab name menu and the Group Invite panel all get the box. The test-only ChatMenu.CopyGuildInviteLine is gone. API.GuildInvite still refuses on Forever and prints the line, for anything calling it directly.
  • New UI.ShowCopyLine(title, line, hint): a one-line copy box built on addon.Dialog, so it has the frame, fonts, opacity and Escape of UGI's other dialogs and fits its height to its contents. The operator: "lets make the UI match the other UI's visually, and I dont' think it needs to be so 'big' as it will just be one line copy." It opens through Dialog:ShowOrDefer, so a dialog already on screen is not closed under the player (Peer Review ba763bea C2). It has no button of its own: the frame's Close and Escape close it, through a new noButtons dialog option that skips the implicit Okay (operator, 2026-11-05: "we can get rid of the 2nd close button on the left hand side, it's not needed"). The whisper tabs' Copy name uses it too; UI.ShowCopyAllPopup stays for the multi-line tab lists.
  • API.ForeverInviteDisabledText is now API.ForeverInviteCopyText, the tooltip saying what the click does. The whisper-tab menu's enabled / disabledTip machinery is replaced by a plain hover tip, since no entry is disabled any more; the Group Invite panel lists the group on Forever again.
  • Specs: zz_chat_menu_spec, zz_whisper_tabs_spec and zz_apicompat_guild_spec move to the new behaviour, and zz_ui_helpers_spec covers the copy box. Seen in the client only for the chat menu entry with the old big popup; the small box and the other three entries are not checked in a client.

New — choose your main character

Discord, Vishiswaz (2026-11-04): "It should choose our main, and let that be switched"; operator 2026-11-05: "yes". New Settings > Guild > Main character lists this realm's characters (from AceDB, as AltGroups:MyCharacters already reads them) plus Automatic. AltGroups:OwnerKey returns the chosen main when it is one of the set, else the first alphabetically as before; the choice is stored account-wide per normalized realm in DB.global.accountMain (AltGroups:ChosenMain). AltGroups:SetMain refuses a character that is not ours on this realm, publishes the group straight to LibGuildRoster, then calls the sync bridge's quiet RunCatchUp (which honours "disable sync"). Switching is safe for the election because no client computes another account's key: peers read the owner from the published group, and Feed's existing sweep removes the old owner key on every client the group reaches. The owner is what the Eligibility tab names the account line after and what comes first in the account's order (accountOf in Modules/FGI_Recruiter.lua). accountMain joins SP.NOT_A_SETTING, since it names characters a template's recipient does not have. Five new examples in Tests/altgroups_spec.lua. Not checked in a client.

New — a text and widget scale

Discord, Pockit (2026-11-04): "Being able to change the scale of the text so I don't have to change my entire UI scale"; operator 2026-11-05: "this is in libaceguiwidgets, you just need to adopt it". New Settings > General > Appearance > Text and widget scale (0.8 to 2.0, DB.global.uiScale) hands the value to LibAceGUIWidgets' W:SetScale through new fn.applyUIScale, at PLAYER_LOGIN and from the slider; templates carry it (SP.CATALOGUE, with the same live apply). KNOWN COST: W:SetScale is session-global, so the other suite addons' library widgets follow the same factor; so nothing is applied until the player has set a value, and UGI never resets a scale another addon chose. Only what the library draws scales; UGI's own raw frames and AceGUI's stock window chrome do not. Two examples in Tests/zz_window_backdrop_spec.lua. Not checked in a client.

Changed — Quiet Zones: tick several zones at once, nothing pre-selected

Discord, Vishiswaz (2026-10-18, thread 024b5c62): a checklist of the chosen continent's zones; operator 2026-11-05: "don't we already have this with the dropdowns? we just need to make the 2nd one multi-select", and on the pre-selection, "don't do this, the current zone is irrelevant". The Zone dropdown's entries are now checkable and stay open on a click (keepShownOnClick); the ticked zones are held in selectedZones and the dropdown reads the one zone's name or "N zones". Add adds every ticked zone through the existing addZone (static-area and duplicate refusals unchanged), then clears the ticks. The tab no longer pre-selects the player's continent and zone, so getCurrentLocation is deleted. The Zone label's and Add button's tooltips are reworded, which retires their two old locale keys; the translated rows keyed on them were deleted rather than rekeyed, since the English changed meaning. Tests/zz_quiet_zones_form_spec.lua moves to the new behaviour, with two new examples (ticking and unticking, and adding several at once). Not checked in a client.

New — carry the guild policy to another game version as text

Discord, Vishiswaz (2026-11-04, thread df8413f1): "I was using it to switch from one game client (TBC) to another (Forever) so we may want to have that capability"; operator 2026-11-05: "yes". Each game client keeps its own saved variables, so the v2.15.0 "Copy policy from another character" cannot see the other client's records, and the guild sync never crosses clients. The Guild Policy page gains, for editors, "Copy your policy out as text" (Bridge:ExportPolicy, the content leaves through the template encoder SP:Encode, tagged kind = "ugiGuildPolicy", version 1) and "Load a policy from text" (Bridge:ImportPolicy, behind a confirm). The import follows the copy's rule and shares its code, now replacePolicyContent: content only, this record's own setAt / setBy / contentHash / leaves kept, so it reads as unpushed and reaches the guild only on Push, stamped as the editor's own save. The text carries no metadata and any it holds is ignored, which is what keeps a pasted datestamp from outranking the destination guild's policy, the reason the settings templates leave gmPolicy out. Decoding goes through SP:Decode, so the paste box has the templates' size limits. The import is also type-gated (Peer Review 8eea7b24): every leaf must be plain data, and of the same Lua type as the leaf this record already holds under that name, else the whole import is refused ("shape"). gmPolicy has no shipped default, so a leaf the record does not hold is checked only for being plain data. Nine examples in new Tests/zz_policy_transfer_spec.lua. Not checked in a client.

New — History tab: an "Invited by" column

Discord, Vishiswaz (2026-10-18, thread 14fba524): "sync functionality for the History tab, specifically to recover information that individual FGI users might not have, particularly Invited By"; operator 2026-11-05: "yes". Recruit records already sync (guild:recruits, behind "Share recruit history with the guild"), but their by is who OBSERVED the current status, not who invited. The inviter was already recorded and synced elsewhere: fn:rememberPlayer stamps inviter on each DB.realm.alreadySended entry, and guild:alreadySended carries the whole entry. So no new wire field: the History tab gains an "Invited by" column (inviterOf in GUI/Tabs/History.lua) read from that record, which works whether or not recruit history is shared. KNOWN COST: the column goes blank once the anti-spam entry expires (antiSpamExpiryDays), and for entries written before v2.8.0, which carried no inviter; and the tab still lists only the recruits this client holds, so a guildmate's recruit appears only with sharing on. The tab's minimum width goes from 580 to 670 for the new column. The lookup goes by the recruit's store key first, which the same fn:normalizePlayerName made, and only then by the display name (Peer Review 8eea7b24). Two examples in Tests/zz_history_form_spec.lua, one with a realm whose name has a space. Not checked in a client.

Changed — Announce, Filters and Messages: the editor waits for New or a row, and says which

Discord, Vishiswaz (2026-10-18, threads 3569ed8f and 495047a0): the editors greyed out until you start editing or creating an entry, with a New button and a visible creating-or-editing marker, "Sometimes I have thought I was doing one when I was actually doing another"; operator 2026-11-05: "yes". New UI.EditorGate (GUI/UI.lua) lays one dim, click-taking cover over a tab's editor strip while idle, with a hint to press New or click an entry, rather than disabling each control (the strips mix edit boxes, Blizzard dropdowns and checkbox menus, and a control added later would be missed). New, and on Announce the Send and Soundboard buttons, which act on the whole list rather than the form, are raised above it. Once open, a line above New and Save reads "Creating a new ..." in green or "Editing: Name" in gold. Save, and deleting the row being edited, put the cover back, and drop keyboard focus from any box inside the strip, since a Save sent with Enter leaves that box focused under a cover that only takes clicks. Filters and Announce had no New button and get one, left of Save; Messages keeps its "Add new". Filters' reset block becomes clearForm, shared by Save and New. Opening an enforced guild announcement reads as creating, "Save makes your own copy", which is what its Save does. KNOWN COST: after a Save the form no longer stays open for the next entry; New has to be pressed again. Twelve examples in new Tests/zz_editor_gate_spec.lua. Not checked in a client.

Changed — WoW Forever: removing a guild member opens "/gremove Name" to copy

Peer Review lead (thread ba763bea); operator 2026-11-05: "do the same copy box you did for invites ... we should handle them the same", and "we'll let the community test it". On Forever API.GuildUninvite now works exactly like the Forever invite: it opens the one-line copy box (UI.ShowCopyLine) with /gremove Name highlighted and no longer prints a chat line. The blacklist kick prompt's Kick tooltip on Forever now says the click gives /gremove Name to send (reusing the retired chat line's translated string), as the invite entries' tooltips do. NOT VERIFIED that a pasted /gremove goes through on Forever; the pasted /ginvite did. Examples in Tests/zz_apicompat_guild_spec.lua (whose after_each now closes the box) and Tests/zzzz_blacklist_kick_popup_spec.lua.

Changed — Wingman's gap between actions goes up to 30 seconds

Discord, Jedric (2026-11-05, thread e0f88075): "I'd like to do 10 seconds instead of 5, make it seem more natural"; operator 2026-11-05: "We just need to increase the maximum time on this slider to say 30 seconds". Settings > Wingman > Minimum seconds between actions now runs 0.5 to 30 (was 0.5 to 5). Modules/Wingman.lua getMinInterval has no upper clamp, so the slider was the only cap. The randomized gap in the same request was not built. One example in Tests/zz_settings_panel_spec.lua. Not checked in a client.

Fixed — a list sorted by default showed no sort arrow until a header was clicked

Discord, Vishiswaz (2026-10-18; held in a spam filter until 2026-11-04): "Columns in the main window that are sorted by default are not showing their sort method until clicked ... in the Blacklist window, the default sort is by Name, but the arrow for it does not appear until clicked." LibAceGUIWidgets' RowList draws the arrow only for its own sortKey (_updateHeaderText), and no tab ever set one: each ordered its rows in buildRows and handed them over unsorted as far as the list knew, so the first click on that column sorted ascending again instead of reversing. The Blacklist, Anti-Spam and Filters tabs now call rl:SetSort("name") and Quiet Zones rl:SetSort("zone") before their first SetData, the order the library documents for a pre-sorted first render. The library's sort is case-insensitive where buildRows compared case-sensitively, so a lowercase name now sits among the others rather than after them. The Filters Name column gains sortValue = rawName or name, because its cell carries a colour code and a "(guild)" or "(v1)" tag that would otherwise decide the order. History, Custom Scan and Messages keep their stored order and have no column sort to show; Announce and Eligibility order their own rows (externalSort). New Tests/zz_default_sort_arrow_spec.lua. Not checked in a client.

[v2.22.0] (2026-11-04) — notes on join, account-wide guild policy, and WoW Forever's invite block

New — notes on join are back, on the click path

Discord, Guz (2026-11-02): "is there a way to add a officer note like in fgi i like to add the date a player joined". v2.13.0 removed the old "Notes on invite" option because it could never act: it wrote from a C_Timer.After(5, ...) after the join, and its guard (if C_GuildInfo and C_GuildInfo.SetNote then return end) was true on every client. C_GuildInfo.SetNote is HasRestrictions = true in every flavour tree, and the same write from a timer is what raised ADDON_ACTION_FORBIDDEN ... SetNote() on TBC/Anniversary (the "SetNote() taint guard" entry, now in CHANGELOG_ARCHIVE.md). Removing the option rather than moving the write was the wrong call; the operator, 2026-11-02: "restore it using the hardware event mechanism for adding the note".

The new Modules/FGI_JoinNotes.lua queues each joiner from FGI_Core's LibGuildRoster OnMemberJoined handler (after the home-guild gate) and writes nothing there. It registers on the click pump (fn.registerClickPump("joinNote", ...), after the welcome and decline-whisper pumps by load order), so the next UGI click writes the oldest joiner's notes through LibGuildRoster's lib:SetMemberNote, the same write and the same player-input rule Ultimate Guild Roster's note editor uses. The pump counts as that click's one owed action, as a welcome or decline whisper does. Combat holds it like the other pumps.

  • One note per click, officer first. Whether a second SetNote inside one hardware event is refused has never been measured, and Blizzard's own editors make exactly one per OK (Peer Review 16c30580, F2), so with both notes on, the officer note is written on one click and the public note on the next.

  • A joiner whose roster row has not arrived yet is kept, not dropped. A join reported by the system message reaches LibGuildRoster before its roster row and guid, which come on the next roster update. The first cut removed the queue entry before trying it, so a click in that window lost the note for good (Peer Review 16c30580, F1). Now a member who left is dropped, one with no guid or no client row yet is kept for a later click, and one still not ready ten minutes after the join (checked on a click, the decline whisper's ceiling) is given up.

  • The client row is found by guid, the key LibGuildRoster's own EditMemberNote uses, not by a name comparison that realm spelling could miss (Peer Review 16c30580, D1). LibGuildRoster 1.8.0 (MINOR 40, inbox 2bcc0109) adds lib:GetClientNotes(name), the same read, and it is used when present; FGI's own row walk is now only the fallback for an older installed library, with its own example in the spec.

  • Where it is offered: lib.CAN_WRITE_NOTES (Classic Era, TBC/Anniversary, Mists). The settings are hidden on retail, WoW Forever, Wrath and Cata, where the library offers no note writes.

  • Permissions: lib:CanEditMemberNote -- Blizzard's own rule: CanEditPublicNote() for the public note, CanViewOfficerNote() and CanEditOfficerNote() for the officer note.

  • Never overwrites: the client's roster row is read on the click and a note that is no longer empty is left alone, so two officers with the option on do not overwrite each other, or a person.

  • Templates: NAME (without the realm) and date() codes, expanded against the join time rather than the click's, then cut to the client's 31 characters on a UTF-8 boundary. A template date() refuses goes out as typed instead of erroring inside the click.

  • Settings and defaults: the pre-v2.13.0 keys (setNote, noteText, setOfficerNote, officerNoteText), both off, officer template Joined %Y-%m-%d. A character that had the option on before v2.13.0 still has true saved and finds it on again. The four keys are back in the template catalogue (Modules/FGI_SettingsProfile.lua).

KNOWN COSTS. The queue is session-only: a joiner not yet written when you /reload or log out gets no note. Notes wait for a click on a UGI button; an officer who is not recruiting writes nothing until they press one. Each click serves one owed action, welcomes first and decline whispers second, so a steady stream of those makes join notes wait behind them (Peer Review 16c30580, F3). SEEN IN GAME on Classic Era (operator, 2026-11-03): an officer note Joined: 11/03/2026 was written on a new member at the next UGI click, with no Lua error. Not yet seen on TBC/Anniversary or Mists, and whether SetNote spends the click's hardware event, so the same click's own /who or invite is refused, was not reported.

New spec Tests/zz_join_notes_spec.lua (17 examples, real LibGuildRoster and the harness's recorded SetNote): nothing is written from the join, the click writes the dated note to the right member, one note per click with the officer note first, a join reported before its roster row survives the click and is written once the row arrives (and is given up after ten minutes), filled notes and missing permission are respected, a member who left is skipped, combat holds the write, and the template expansion edge cases.

Fixed — a character with no guild policy never fetched it from guildmates

Discord, Pockit (2026-11-02, retail): after pushing the guild policy from the main, their own alts in the same guild still had none of it, while other members it was pushed to were fine. The alts are on the other faction (a cross-faction retail guild), so each has its own factionrealm store and its own policy, empty until something fills it. On login the bridge advertises its item hashes (Bridge:GetMyHashes), and peers offer only the items a hash list names (P2PSession OnHashListReceived); the bridge's hasMissingItems is false, so nothing ever asks for a missing item. A client holding NO policy advertised nothing for it, so no peer offered one: it could only get the policy from a live push it was online for. Any member offline at the push, with no other character sharing the store, stayed without it.

A client with no policy record now advertises NO_POLICY (Bridge.NO_POLICY). A peer holding a stamped policy sees a different hash and offers it, and the existing query and ApplyPolicy path fills the empty store. Two clients with none advertise the same value, so nothing moves between them, and HasContent keeps an empty client from offering. A record with no canon (saved before the stamp, or an unpushed draft) still advertises nothing, so no draft can be overwritten. A holder that has unpushed edits no longer offers its policy at all (HasContent is false while PolicyHasDraft): OnDataRequest never serves a draft, so the offer used to leave the requester's session waiting out the delivery watchdog for nothing (Peer Review 4e8eb635, P1). KNOWN COST: a character whose only policy is an unpushed draft -- one click in the eligibility table or one keystroke in a policy list creates the record -- still advertises nothing and is not reached by this (G1). The policy_sync_spec example that pinned "advertises nothing when there is no policy" is reversed with the reason recorded, plus one asserting no real version can equal NO_POLICY. Not verified in a client, and one unknown remains: the catch-up's offer and reply legs travel by whisper, and whether a whisper reaches a guildmate on the other faction in a retail cross-faction guild has not been checked.

New — the guild policy is kept once per guild for the whole account

Discord, Pockit (2026-11-03, retail): every character had the guild policy from weeks ago, but none of their own alts had the changes made a few days ago. The fix above covers a character with NO policy; this covers the older one. gmPolicy lives in AceDB's factionrealm scope, so each faction-realm holds its own copy, and an alt caught up only from a guildmate online at its login. The only editor is the player's own GM character on the same account, so it is never online at the same moment as an alt. The operator, 2026-11-03: "we should probably save the guild policy data at the account level ... so that any character in that guild gets the newest policy, and any online character that gets and update, updates it for all characters. We would need to save this for each guild the account happens to be part of".

New Modules/FGI_AccountPolicy.lua keeps DB.global.guildPolicies[<guild key>], the newest published policy of each guild. The live slot stays where every reader looks; the module moves versions between the two:

  • Load (AccountPolicy:Sync), from GuildScope:Resolve once the verdict is settled: the account copy goes through Bridge:ApplyPolicy, the per-leaf merge a guildmate's push goes through, so the newer leaf wins and nothing is blended any other way. The character then reads as a receiver of that version (gmPolicyReceived), which it is.
  • Save (AccountPolicy:Save), after StampPolicySave (this character's own Push) and after an ApplyPolicy that won (a guildmate's push). A draft (PolicyHasDraft) is never saved, and an older live copy never overwrites a newer account copy. A Save before any load loads first, so it merges rather than writes over.
  • Guild key: C_Club.GetGuildClubId(), which the operator checked in game on Classic Era: the same id (486724351) on two characters of one guild. It survives a rename. It is RequiresClubsInitialized and Nilable, so until it answers the key is the guild's name and HOME realm (GetGuildInfo's fourth return via LibGuildRoster's GetHomeGuildInfo, seen populated in game), never the character's realm, which differs across a realm cluster. The entry moves to the id once it answers.
  • Migration, once per guild with no entry yet: every other faction-realm store on the account that holds this guild by name, live or parked by GuildScope, is merged in the same way. Only stores on this character's realm, the guild's home realm or a connected realm (API.GetAutoCompleteRealms), since a guild name is unique only within a realm group. Only stamped records with leaf canons.
  • Bridge:PolicyHasDraft(rec) now takes an optional record, so migration can check another character's copy. A record with no leaf canons answers as a draft there.

KNOWN COSTS. A policy saved before v2.13.0 has no leaf canons and is not migrated; that alt gets it from a guildmate as before. Each game version keeps its own saved variables, so nothing crosses between them, and two WoW licences under one Battle.net login are two accounts. Only the policy moves: messages, announcements and recruit history stay per character. The community id is checked only on Classic Era; on another flavour where it stays nil, the name-and-realm key is used. With the "who is on duty" chat notice switched on, an editor holding an unpushed list edit while a newer account copy loads after warm-up may see the duty change printed twice, once for the account copy and once back to the draft (Peer Review a5aafbb4). Not verified in a client.

GuildScope:Resolve's memoised early return asks AccountPolicy:NeedsLoad() on every roster update and loads when nothing is loaded yet or the club id has answered since, because the settled pass can run before GetGuildInfo names the guild or before the id answers.

Peer Review a5aafbb4 found two defects in the first cut, both fixed before it shipped:

  • R1 -- a login where the club id answered late loaded nothing: the key fell back to the name, the entry was already filed under the id, and migration does not read the account store. While the id is nil, an entry filed under an id whose recorded guild and realm match is now this guild's (entryKeyFor), and the next roster update loads again once the id answers.
  • R2 -- the name-keyed entry was deleted before the club entry was written, so a refused save (a draft) lost it. It is removed only after the club entry's save succeeds.

And one behaviour it confirmed: the merge decides each leaf by canon alone, so a newer account copy replaced an editor's unpushed edit at login without a word. The edited leaves (Bridge:PolicyDraftLeaves, new) are now kept across the load; they still read as a draft, so the account copy is not written until the editor pushes.

Tested, not played. The operator's guild does not use a guild policy, so this ships for the community to confirm. Before that: 100% line coverage of FGI_AccountPolicy.lua, FGI_GuildScope.lua, FGI_DeltaSync.lua and FGI_JoinNotes.lua (1239/1239, desk coverage run 20261103-135505.524); the whole suite green in default, reverse and three shuffled orders (seeds 42, 12345, 1); and three mutation checks -- the account-copy load switched off, the draft restore switched off, the realm filter on migration switched off -- each turned exactly the examples written for it red. The spec's end-to-end block switches AceDB's faction-realm section between two characters of one account, as a login does, and walks the report: the main's push reaching an alt on the other faction, a guildmate's push the alt alone was online for reaching the main, two characters each pushing a different part of the policy, and two guilds on one account kept apart.

The shuffled runs also found three spec files leaning on state an earlier file left behind, all fixed in the specs: zz_queue_seen_spec queued "Alice", whom earlier files leave on the guild roster, and since v2.20.1 the scan drops a guildmate from the queue; zzzz_legacy_window_behaviour_spec's blacklist example ran with the guild scope still stamped by another file's guild, so the kick sweep's resolve swapped the blacklist out from under it; and zz_settings_profile_spec's side-effect parity check saw the once-per-guild account load ride on the auto-kick setter's resolve, so that check now settles the guild before it wraps anything (Peer Review a876a26a) instead of exempting the calls. Tests/support/addon.lua now starts every file with an empty account policy store and an unsettled guild scope, and clears guild members a file left on the roster between files (measured first as a raising guard: 17 files leave their own guild behind, which is only wrong for the file after).

New spec Tests/zz_account_policy_spec.lua (27 examples, real LibGuildRoster and GuildScope): a push is saved under the name-and-realm key; an alt holding an older copy takes the newer one when its guild settles, or on a later roster update when the first pass loaded nothing; a late club id still loads the club entry and loads again when it answers; a name entry survives a refused move; an unpushed edit survives a newer account copy; a draft is never saved; an older live copy never overwrites a newer account copy; a winning guildmate push is saved; another guild never reads the copy; the entry moves to the community id; and migration takes a connected realm's newer copy and a parked copy, ignoring another guild, an unconnected realm and a pre-leaf-canon record.

Fixed — guild policy name lists typed with commas matched nobody

Discord, Bronzefinger (2026-11-02, Classic Era): Wingman would not start and said nobody on the guild policy's lists was available, while the member was on both lists. The screenshot showed the announcer list as Bronzefinger-Pagle,Mayhem and the inviter list as Bronzefinger, Mayhem, each on one line. parsePolicyNames in GUI/SettingsPanel.lua split only on line breaks, so each line was stored as one name that matched no member, and with the v2.21.0 stop-when-nobody-is-available settings on, recruiting was held for the whole guild. Commas and semicolons now separate names too (no character name can hold either). A list already saved with commas is still one bad entry until it is accepted again; the reply to the player says to re-enter and push it. New example in Tests/zz_settings_panel_spec.lua.

Fixed — WoW Forever: no more "blocked from an action" popup on every invite

Blizzard's notes for the WoW Forever 1.60.1 launch build: "C_GuildInfo.Invite is now forbidden for addons and macros. Outgoing guild invites are now throttled to 10 per minute." Live on launch day (2026-11-04): three players on Discord and the operator's own Diagnostics capture hit ADDON_ACTION_FORBIDDEN ... 'Invite()' from FGI_APICompat.lua's API.GuildInvite, on a click. The pcall around the call does not stop the popup, so the only fix is not to make the call.

On gv.isForever, API.GuildInvite now calls neither C_GuildInfo.Invite nor GuildInvite. It prints one chat line telling the player to type /ginvite <name> (Blizzard's own slash handler) and returns false. Callers still record the pending invite, which is wanted: the server answers the player's /ginvite with the same system message, so the outcome is tracked unchanged, and one never sent is swept as Unresolved after an hour like any unanswered invite. Every other flavour is unchanged. New example in Tests/zz_apicompat_guild_spec.lua.

Forever is Whisper only. The operator, 2026-11-04: "for forever only ... the only invite mode is whisper only". Every send path reads DB.global.inviteType (fn:invitePlayer), so the save is what holds: GUI/UI.lua gains UI.FOREVER_INVITE_MODE (3), UI.InviteModeAllowed, and UI.EnforceForeverInviteMode. Both mode dropdowns (Scan tab, legacy window) list only Whisper only there, UI.SetInviteMode saves 3 whatever is picked, UI.SavedInviteMode reads 3, and the mode is put back at login (after the login template, in FGI_Core.lua) and after any SettingsProfile:Apply, so a template made on another flavour cannot switch invites back on. The compact window's invite button, its row icons, Wingman and the keybind all go through fn:invitePlayer, so they whisper too.

Every other invite and kick surface asks one predicate. "Apply this logic to whereever we do invites or kicks" (operator, same day). FGI_APICompat.lua gains API.AddonGuildInviteAllowed() and API.AddonGuildKickAllowed(), false on Forever.

On Forever every guild invite entry is greyed out, kept in place. The operator, 2026-11-04: "in forever only, we need to disable the invite buttons. can we just grey them out so they don't work? so that if we come up with a solution, they are still there? maybe change the tooltip saying that Forever broke them". The chat right-click "UGI - Guild Invite" (SetEnabled(false) on its menu description), the legacy dropdown (disabled + tooltipWhileDisabled, both read by Forever's UIDropDownMenu_AddButton), the whisper-tab name menu (the row disabled and greyed, with an OnEnter tooltip) and the Group Invite panel (lists nobody, shows the reason) all ask API.AddonGuildInviteAllowed and show API.ForeverInviteDisabledText: Blizzard no longer lets addons send guild invites there, type /ginvite and the name in chat. Anything reaching ChatMenu.GuildInvite another way records nothing (no attempt, no invitedForAccept, no pending invite) and gets API.GuildInvite's /ginvite chat line.

Why no addon route is left, all seen in the operator's Forever client on 2026-11-04. The first build opened Blizzard's own Add Guild Member box (CommunitiesUtil.OpenInviteDialog, as Ultimate Guild Roster's Invite Member does) for the player to Accept, at first with the name typed in, then empty, then from the menu entry's own click, then from a plain button and from a button inside the guild window. Every Accept was refused, ADDON_ACTION_FORBIDDEN ... 'Invite()' from GameDialogDefs.lua:945 OnAccept (or :991 EditBoxOnEnterPressed with the name typed in), taint log "blocked because of taint from fastguildinvite - Invite()": StaticPopup_Show writes dialog.which with the opener's taint (StaticPopup.lua:405) and the Accept reads it (:699). LibGuildRoster's own Invite Member was refused the same way; the one invite that went through was Blizzard's own Invite Member, clicked by the player. A /ginvite line put into the chat box by code, with Enter pressed by the player, was refused too (SlashCommands.lua:1091 via ChatFrameEditBox.lua:271). Guild Finder links are out: C_ClubFinder.IsEnabled() is false there, and invite links are for communities only (GameDialogDefs.lua:1446). So API.OpenAddGuildMember and the box-already-up check are deleted, and the "Type X into the Add Guild Member box" locale key with them. API.GuildUninvite (the blacklisted-member Kick button) prints /gremove <name> or the guild window instead of calling C_GuildInfo.Uninvite. Not verified that Forever refuses Uninvite: Blizzard's notes name only Invite, and the forever tree's GuildInfoDocumentation flags neither (it does not show the Invite forbid either); kicks follow on the operator's instruction. /gremove's spelling is a client string that is not in the source.

New Tests/zz_forever_invite_mode_spec.lua; Forever examples in Tests/zz_apicompat_guild_spec.lua (no box opens, only the /ginvite line; the tooltip text) and Tests/zz_chat_menu_spec.lua (the entry greyed out with its tooltip, enabled off Forever, a click doing nothing, and the shared invite recording nothing). The greyed-out entries are not checked in a Forever client yet.

Locale/enUS.lua: the L["Copy name"] line had the next comment run onto it with no line break; split.

New — "Copy name" in the whisper window's name menu

Discord, Pockit (2026-11-04): "can we at least get a copy user name option from the UGI chat window box. special characters ... are a real pain to GInvite". With Forever invites now typed by the player, the whisper-tab menu (Modules/FGI_WhisperTabs.lua) gains a Copy name row, before Cancel, on every flavour: it opens UI.ShowCopyAllPopup with the name highlighted for Ctrl+C (an addon cannot write the clipboard), spelled through fn:formatNameForRetailAPI so it is what /ginvite takes. New example in Tests/zz_whisper_tabs_spec.lua.

Fixed — a guild announcement's channel cooldown was lost on /reload or relog

Discord, Vishiswaz (2026-11-04): "I have an announcement set to go to General chat and the cooldown on it goes away upon relog/reload." For an announcer, an enforced guild announcement on a public channel is timed only by Modules/Announce.lua's per-place sightings (layerDueIn), and that table was memory only -- so after a reload the place read as never posted and came due again at once, against the file's own header ("/reload doesn't reset cooldowns"). New Announce:SaveSightings writes it to DB.char.announceSightings (key -> epoch) at AceDB's OnDatabaseShutdown, registered in fn:initDB beside the other last-writes of a session, and Announce:RestoreSightings reads it back at init, newer stamp winning. Entries older than a day are dropped both ways. The meaning is unchanged: per place, and still merged with guildmates' posts. Four examples in Tests/announce_per_layer_spec.lua. Suspected cause, not confirmed for this report: whether Vishiswaz's announcement is a guild one (this path) or a personal one (already saved per character) is not known yet. Not checked in a client.

Fixed — Eligibility rank numbers sat under the wrong column title

Discord, Vishiswaz (2026-11-04), with a screenshot: the Announcer numbers read as if they were under "Inviter". GUI/Tabs/Eligibility.lua's rank columns were justify = "RIGHT", and LibAceGUIWidgets' RowList always left-justifies header text (LibAceGUIWidgets-RowList.lua:824-826), so each rank was drawn at the right edge of its column, a full column-width from its own title and next to the following one. The rank cells are now justify = "LEFT", under their titles. Not checked in a client yet.

[v2.21.1] (2026-10-31) — coverage catch-up, and a legacy guild set that was never adopted

Fixed — a guild never got back legacy data whose verdict had to wait for the roster

Found by the coverage catch-up. When a character joins a guild while the live set belongs to another guild, and the only parked set is an unstamped legacy one, the verdict on whether it is this guild's rests on the policy author, and that waits for a built roster. GuildScope:Resolve in Modules/FGI_GuildScope.lua still swapped the other guild's set aside and made this guild the owner, so on the next resolve the owner was in scope and the legacy set was never looked at again: a guild whose author it was never got its legacy policy and lists. A waiting verdict now sets guildScopeLegacyWait, and the next resolve judges the parked set again; when it is this guild's, it is loaded with what this guild wrote meanwhile merged in (the new loadOver, now shared with the sister-guild case, which also keeps a value set meanwhile where the loaded set has none). Only the shared lists are merged key by key: a policy is one authored record, so the loaded policy wins whole rather than being mixed with one saved while the roster built (found writing the self-audit; the first cut of loadOver unioned every table). What this guild wrote while waiting is kept wherever the loaded set has nothing for that key -- read off the parked set itself, since swap fills every key it is not given with an empty default and a test of the live value dropped the guild's own messages (Peer Review 262c69d8, a). A wait whose parked set has gone clears instead of re-resolving for ever. Three new examples in Tests/zz_guild_scope_spec.lua; the file is at 169/169. Not verified in a client.

Changed — README and CurseForge page: Ace3 is a manual install on WoW Forever

Discord, Danne (2026-10-31, WoW Forever): UGI and every library except LibLocaleOverride reported a missing dependency after a clean CurseForge reinstall. LibLocaleOverride is the only one with no ## Dependencies: Ace3 line, and Ace3 has no Forever release yet (operator, on Discord: "Ace3 doesn't have a Release version in Forever yet. You have to manually install it from CF"). Installing Ace3 into the Forever folder fixed it. Both player docs said CurseForge installs every dependency for you; they now name Ace3 on Forever as the exception, and why: CurseForge installs only release versions by default, and Ace3 publishes only alpha builds for Forever so far (operator, 2026-10-31: "it has to be installed by hand UNTIL they issue a release version ... CF only installs release versions by default, and Ace3 is only issuing -alpha builds right now"). Drop the note once Ace3 ships a Forever release.

Changed — every file of FGI's own code at 100% line coverage

Measured in six chunks through the harness's coverage.lua: every root file, all 38 Modules/, all 19 GUI/ files (the twelve under GUI/Tabs/ had never been measured), FGI's own Libs/LibWho and Locale/summary.lua. Not measured: the vendored libraries (LibCompress, LibDeflate, LibDataBroker, LibDBIcon) and the translation tables. What the gaps were, and what closed them:

  • FGI_AnnounceSync.lua: mergeLayer returned a false no caller reads; removed.
  • FGI_GuildScope.lua: the sister merge guarded two states its own swap rules out, folded into one condition; examples for an unprefixed owner stamp and a parked non-sister set; and the legacy-wait defect above.
  • FGI_Recruiter.lua: the alt taking its main's turn in the announcer line-up.
  • GUI/SettingsPanel.lua: v2.21.0's three stop toggles, the inviters-only sync setting and the three "nobody available" status lines had no example.
  • GUI/Tabs/Blacklist.lua: Import UGR, shown only with Ultimate Guild Roster loaded.
  • GUI/Tabs/Scan.lua: the held count on the scan line.
  • functions.lua: the bare SendChatMessage fallback. Every flavour tree on the reference drive has C_ChatInfo.SendChatMessage, but the Wrath and Cataclysm Classic clients are older builds than any of them, so it was kept and specced rather than deleted.
  • Locale/summary.lua (56%): each flavour's class and race tables are built at load, so only Classic Era's were reached. Tests/zz_summary_flavours_spec.lua reloads it as every flavour. It pins which race key sits on which side under which id; that those ids are a real client's is not checked.

Suite 4675 passed in default and reverse order.

Then the shuffled orders, which nobody had run on this work: seeds 12345 and 42 and the reverse order passed, seed 1 failed two examples, both test isolation and neither a product defect. Tests/wingman_prompt_spec.lua replaced C_Timer with a stub that had no NewTimer, and a DeltaSync session an earlier file left open called it from inside this file's roster publishes; the stub now has one. Tests/zzzz_blacklist_kick_popup_spec.lua seeded its blacklist before publishing its roster, and when an earlier file had left the live set stamped to another guild that publish's guild-scope resolve parked the seeded blacklist; the spec now puts it back once the guild is settled. After both fixes, the default order, reverse order and seeds 1, 12345 and 42 all pass at 4676.

Older versions (129 to v2.21.0) are in CHANGELOG_ARCHIVE.md.