The transfer-target check previously matched only against
CONFIG.ROLE_TO_PING_ID, so a member with one of
CONFIG.ADDITIONAL_STAFF_ROLES (a recognized staff role everywhere else
in the bot, including requireStaffRole and the messages.js claimer-DM
path) was rejected as a transfer target. Switch to isStaff() so the
transfer-target gate matches the rest of the codebase's staff
definition.
Also:
- Reject bots as transfer targets (guildMember.user.bot).
- Reject self-transfer (transferring to interaction.user.id) — the
rename + DB write would no-op but the log line claimed a transfer
that didn't happen.
- Resolve the target member cache-first to avoid an unnecessary REST
round-trip when the GuildMembers intent has the user cached.
Two real bugs in handleTransfer plus a class issue across all the
channel-mod handlers.
/transfer didn't rename
handleTransfer set claimedBy but never called enqueueRename, so the
channel name stayed at whatever the previous claimer left it as.
/claim (applyClaim in handlers/buttons.js) does the rename via
makeTicketName + STAFF_EMOJIS; /transfer now does the same, plus
writes claimerId (was only writing claimedBy). Uses
'escalated-claimed' state when tier >= 1, 'claimed' otherwise.
DiscordAPIError 10062 (Unknown interaction)
handleAdd / handleRemove / handleTransfer / handleMove / handleTopic
all called interaction.reply() at the end after awaiting one or more
channelQueue ops. Those ops serialize behind any pending rename or
move on the same channel — easily exceeding Discord's 3s interaction-
token window. The reply then 404s with code 10062. Production logs
showed handleRemove failing this way (the visible 'Remove user
error: DiscordAPIError[10062]' lines); transfer had the same pattern.
Switch each handler to deferReply() up front + editReply() at the end
+ editReply() in the catch (with .catch(() => {}) to swallow the rare
case where even the deferred reply context is gone).
handleTransfer keeps the up-front isStaff role check as a reply()
because that path is synchronous and the token is fresh.
After the previous TICKET_BUTTON_HANDLERS gate, ticket creators and
/add'd members were locked out of every ticket button — including
close_ticket on their own ticket. Add a PUBLIC_TICKET_BUTTONS set so
the close flow (close_ticket / confirm_close / confirm_close_with_email
/ confirm_close_no_email / cancel_close) skips the staff check.
Claim, escalate, and de-escalate remain staff-only. The 60s
FORCE_CLOSE_TIMER countdown, the transcript archive, and the optional
customer-closure email all continue to fire on the existing
runFinalClose path — nothing about the close behavior changes, only
who is allowed to click the button.
cancel_close is intentionally public too: anyone in the channel can
abort a pending close, including the original setter, staff, or the
creator. The pendingCloses entry stores who set it, but the abort path
doesn't gate on that — kept permissive to match the rest of the close
flow.
The /escalate slash command never had a reason option in its definition
(commands/register.js only takes a 'level' option), so handleEscalate
hardcoded reason=null. The escalate button path passed null explicitly.
The log line wrote it verbatim as "Reason: null" on every escalate.
Remove the dead surface:
- runEscalation signature drops the reason parameter.
- The customer-facing email body drops the conditional reason suffix
(`reason ? `\n\nReason: ${reason}` : ''`) — always-false branch.
- The logging-channel post drops "\nReason: ${reason}".
- handleEscalate drops the `const reason = null;` line and the call-site arg.
- handleEscalateButton (handlers/buttons.js) drops the trailing `null` arg.
If we ever want to capture a reason, the slash command would need a
StringOption('reason') and an escalate-modal for the button path —
neither exists today.
Closes the remaining non-broccolini interaction paths after the prior
TICKET_BUTTON_HANDLERS gate. After this commit, every bot interaction is
staff-only except the panel buttons (open_ticket / open_ticket_thread /
open_ticket_channel) and their ticket-creation modal submit — those have
to stay public because they're how members and customers open tickets.
Specific changes:
- handlers/commands/index.js: handleCommand no longer has the
`!== 'help'` carve-out. /help now goes through requireStaffRole like
every other slash command. Non-staff get the same ephemeral
"only available to the support team" reply.
- broccolini-discord.js: the signature_modal_* modal-submit handler now
calls requireStaffRole before writing to StaffSignature. /signature
already gates the modal display via the slash-command staff check;
this is defense in depth against directly crafted submissions.
- handlers/buttons.js: cancel_delete_tag moved out of
FREE_BUTTON_HANDLERS and gated alongside confirm_delete_tag::*. The
dialog is only shown ephemerally to the staff who triggered
/response delete, so non-staff can't reach it in normal flow; gating
keeps the button surface consistent.
Kept public (by design — these are the customer entry points):
open_ticket / open_ticket_thread / open_ticket_channel buttons
ticket_modal / ticket_modal_thread / ticket_modal_channel submits
handleButton routed claim_ticket, close_ticket, confirm_close /
confirm_close_with_email / confirm_close_no_email, cancel_close,
escalate_ticket, escalate_to_tier2, escalate_to_tier3, deescalate_ticket,
and confirm_delete_tag::* straight to their handlers without any staff
check. Any non-staff member with View Channel on a ticket — the ticket
creator themselves, or anyone /add'd to it — could click those buttons
and mutate ticket state (claim, escalate, close, delete saved-response
tags).
The slash-command dispatcher in handlers/commands/index.js already
calls requireStaffRole before invoking any handler; the button
dispatcher needed the same gate. Now:
- confirm_delete_tag::<name> → requireStaffRole, then proceed.
- TICKET_BUTTON_HANDLERS dispatch → requireStaffRole, then proceed.
- FREE_BUTTON_HANDLERS (open_ticket* panel buttons, cancel_delete_tag)
remain ungated — those are public-facing by design.
requireStaffRole replies ephemerally ("This command is only available
to the support team (<@&role>)") and returns true when the caller
should bail, matching the slash-command behavior.
QUAL-004 handlers/messages.js — DM-on-customer-reply now reads
guild.members.cache.get(claimerId) first and only falls back to
guild.members.fetch on cache miss. Avoids a REST round-trip per non-staff
reply on busy tickets. GuildMembers intent already keeps the cache warm.
QUAL-005 handlers/buttons.js (runFinalClose) + handlers/commands/close.js
(finalizeForceClose) — close paths now $unset welcomeMessageId alongside
the status: 'closed' write. Stops a stale message-ID from carrying into a
future reopen on the same Gmail thread, where escalation's "edit welcome
buttons" path would silently fail trying to fetch a message in a deleted
channel.
QUAL-007 services/configPersistence.js — writeEnvFile mismatch error now
includes the missing/extra key sets, not just count vs count. Saves the
operator from guessing which key vanished after a partial write.
QUAL-008 utils.js stripEmailQuotes — replaced order-dependent first-match
loop with an earliest-match-across-all-markers scan. The previous code
could truncate at a late "_____" signature underline even when an earlier
"On X wrote:" reply header was the real cutoff. New test in
tests/utils.test.js exercises the dual-marker case.
QUAL-010 broccolini-discord.js — moved `let httpServer / internalServer /
appReady` declarations from after the ready handler to before it. Same
runtime behavior (module-load completes before ready fires asynchronously),
but the read order now matches the assignment order.
SEC-002 routes/internalApi.js — POST /restart now goes through a tighter
2/min limiter on top of the shared 10/min internalLimiter. Defense in
depth in case INTERNAL_API_SECRET ever leaks; an attacker with the secret
can no longer crash-loop the container.
Skipped: QUAL-009 (re-checked the regex; ^\s*\n* → \n is already
idempotent — the audit finding was incorrect).
vitest run: 88/88 (one new test for QUAL-008).
The 1028-line handlers/commands.js bundled escalation logic + force-close
flow + /response tag CRUD + /panel + /signature + context-menu handlers +
several config-toggle slash commands. After the dispatch-table refactor it
was still a god module. Split into handlers/commands/ with one file per
topic; require('./commands') resolves to handlers/commands/index.js
(handlers/commands.js is removed).
Layout:
helpers.js — requireStaffRole, fetchLoggingChannel
(cross-submodule, kept here to avoid cycles with index.js)
escalation.js — runEscalation, runDeescalation, handleEscalate, handleDeescalate
(run* are still exported via index.js for handlers/buttons.js)
close.js — handleForceClose, handleCancelClose, handleCloseTimer
+ finalizeForceClose / postTranscript (timer callback)
response.js — handleResponse + send/create/edit/delete/list subcommands
+ handleAutocomplete (only /response autocompletes)
panel.js — handlePanel, buildPanelButtonRow, handleSignature
contextMenu.js — handleCreateTicketFromMessage, handleViewUserTickets
index.js — dispatch tables, handleCommand/handleContextMenu, plus the
short-and-not-thematic handlers (notifydm, add, remove,
transfer, move, topic, staffthread, pinmessages, gmailpoll,
help) and the public re-exports.
No behavior change — every imported name, every Discord call, every DB
write, every embed, every reply payload preserved verbatim. Public surface
of require('./commands') is still { handleCommand, handleContextMenu,
handleAutocomplete, runEscalation, runDeescalation }.
Largest single module is now index.js at 299 lines; others are 33–214.
QUAL-006 store ticket.creatorId on creation; legacy split-pop returned the
message ID for discord-msg-* tickets, breaking transcript DM, close
log, and channel rename for context-menu-created tickets. Adds the
field to the Ticket schema and writes a one-shot backfill script
(scripts/backfill-creatorId.js, dry-run by default).
QUEUE-001 add enqueueOverwrite + enqueueTopic to services/channelQueue.js
(chain on renameChains alongside enqueueMove). Migrate handleAdd /
handleRemove / handleMove / handleTopic so permissionOverwrites,
setParent, and setTopic no longer race pending renames or sends.
handleMove now uses the existing enqueueMove. Initial overwrites in
handleTicketModal stay inline; channel doesn't exist yet so no race.
DISCORD-001 replace ephemeral: true with flags: MessageFlags.Ephemeral across
broccolini-discord.js, handlers/sharedHelpers.js, handlers/buttons.js,
handlers/commands.js. runDeferred opts now take { flags } directly.
SEC-003 /gmailpoll min interval is 30s. Drop the 5s/10s slash-command
choices and clamp Math.max(30000, ms) in handleGmailPoll for
defense in depth.
QUAL-001 upgrade silent .catch(() => {}) on the lastActivity updateOne in
handlers/messages.js to log via logError, so transient Mongo errors
surface in the debug channel instead of disappearing.
QUAL-002 drop await from logError/logWarn calls in services/staffThread.js
and services/pinMessage.js — fire-and-forget per CLAUDE.md hard rule.
QUAL-003 wrap stray setTimeouts (handleConfirmCloseRequest force-close timer,
runFinalClose channel-delete + overflow-cleanup, checkAutoClose
delete-after-email) in trackTimeout via lazy require so they clear
on shutdown.
Each customId now maps to a named handler in one of two tables:
FREE_BUTTON_HANDLERS (open-ticket panel, tag-delete cancel — no ticket
lookup) or TICKET_BUTTON_HANDLERS (anything fired inside a ticket channel
— the dispatcher does the lookup once before delegating). The dynamic
`confirm_delete_tag::*` id is matched by prefix.
To find a button's logic, search handle<Name>Button or handleTagDelete*.
Other cleanups in the same pass:
- Move findTicketForChannel and runDeferred from handlers/commands.js to
the new handlers/sharedHelpers.js so both files share one source of
truth. runDeferred now also calls logError(verb, ...) — was logged ad
hoc in buttons.js, missing in commands.js. Strictly additive.
- Hoist three inline `require('../services/...')` calls (staffThread,
pinMessage, debugLog) to top imports.
- Collapse escalate_to_tier2 and escalate_to_tier3 into one
handleEscalateButton(interaction, ticket) that derives the tier from
customId. Same for confirm_close / confirm_close_with_email /
confirm_close_no_email — one handleConfirmCloseRequest deriving
sendEmail from customId.
- Decompose the 156-line handleConfirmClose into runFinalClose +
buildTranscriptText + formatDateForTranscript + renderTranscriptHeader
+ dmTranscriptToCreator + postCloseLogEntry. Each piece is testable in
isolation.
- Decompose handleClaim into applyClaim + applyUnclaim.
- Extract buildOpenTicketModal() and postTicketWelcomeEmbeds() so the
ticket-creation modal flow is readable top-to-bottom.
No behavior change. handleButton + handleTicketModal exports preserved;
24/24 modules load clean (sharedHelpers.js is the new one).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Each slash command and context-menu entry is now its own named function;
handleCommand looks the name up in COMMAND_HANDLERS and delegates. Finding
where a command lives is now "grep handle<Name>" instead of scrolling 600
lines of sequential ifs.
Other cleanups in the same pass:
- Restore ActionRowBuilder, ModalBuilder, TextInputBuilder, TextInputStyle
to the top-level discord.js destructure. ActionRowBuilder was used at
runtime by /response delete and /panel but had been dropped from the
imports based on a stale "unused" diagnostic — those paths would have
thrown ReferenceError. Hoisting the previously-inline /signature
imports as well.
- Hoist `logTicketEvent` to the top imports (was inline-required at three
callsites).
- Extract findTicketForChannel(), runDeferred(), and fetchLoggingChannel()
helpers — replaces the lookup-then-defer-then-try-catch boilerplate
repeated in nearly every branch.
- Pull the force-close timer body into finalizeForceClose() and
postTranscript() so the timer registration is one line.
- Pull the panel button-row construction into buildPanelButtonRow().
- Split /response into its own RESPONSE_SUBCOMMANDS dispatch + per-sub
handlers.
- Consolidate the duplicated transcript date formatter into one local fmt().
No behavior change. All 23 modules still load clean; handleCommand,
handleContextMenu, handleAutocomplete, runEscalation, and runDeescalation
exports are preserved (handlers/buttons.js imports the last two).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Rename CONFIG.TRANSCRIPT_CHAN -> CONFIG.TRANSCRIPT_CHANNEL_ID and
CONFIG.LOG_CHAN -> CONFIG.LOGGING_CHANNEL_ID across 9 callsites so
CONFIG keys match their .env names — no more "grep .env, find nothing"
for new readers
- Replace handlers/commands.js#hasStaffRole with utils.js#isStaff
(was a verbatim copy)
- Delete utils.js#enforceEmbedLimit and its 2 callsites; both inputs are
bounded well under the 6000-char Discord embed cap, so the trim was
defensive code that never fired
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Remove no-op log stubs (logGmail, logAutomation, logSecurity, logSystem)
and ~17 callsites; dead counters in tickets.js and gmail-poll.js go too
- Dedup three near-identical Gmail send paths into sendThreadedEmail helper
- Drop dead Mongoose fields: broccoliniTicketId, lastSyncedBroccoliniArticleId,
renameCount, renameWindowStart, reminderSent, staffChannelId,
unclaimedRemindersSent, lastMessageAuthorIsStaff
- Drop dead config fields and their .env.example entries
- Inline api/botClient.js (3-line wrapper, 2 callers)
- Trim unused exports across utils.js, tickets.js, configSchema.js, debugLog.js
- Fix handlers/messages.js to use isStaff() — old partial check ignored
ADDITIONAL_STAFF_ROLES, so those members were treated as customers
- Drop unused deps p-queue + dotenv-expand; move mongodb to devDependencies
Net: -583 LOC source + -57 LOC lockfile. All 23 modules load clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- secondary rename-bot token was set as RENAME_TOKEN in .env but utils/renamer.js reads RENAMER_BOT; silently no-op'd every rename (host .env renamed separately)
- services/tickets.js canRename gutted to an always-ok shim; Mongo 2/10min per-channel gate is redundant since renames flow through RENAMER_BOT's own bucket. Ticket.renameCount / renameWindowStart remain as orphan fields (no migration)
- handlers/buttons.js + commands.js: drop the four "Channel renamed too quickly" else-branches and the rename-countdown label suffix; replace .catch(() => {}) with .catch(err => logError('rename', err)...)
- services/channelQueue.js: executeRename falls back to channel.setName(currentName) when renamer throws err.fallback === true (401/403/429); classifies non-fallback errors as renameQueue:token/permission (401/403) or renameQueue:secondary-bot ratelimited (429)
- utils/renamer.js: on 401/403 throw err.fallback=true immediately; on 429 respect retry_after up to 2000ms then throw err.fallback=true
- docs: align CLAUDE.md, docs/api/DISCORD_API_VALIDATION.md, docs/architecture/CRITICAL_FILES_AND_HOW_IT_WORKS.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
