ReturnUrlComponent stored ?returnUrl= whenever it contained "http", and
layout.context sent the browser there after login or sign-up. That made the
login an open redirect, and a javascript: value could run script on the
Postiz origin. returnUrl is now kept only when it is an http(s) address on
the app's own origin, checked when it is stored and again when it is read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The HubSpot app's redirect URL (/oauth/hubspot/callback) dropped people on
the bare calendar, so nothing said the install worked or what to do next.
It now lands on /launches?hubspot=installed, where HubspotInstalled opens a
modal with the next steps in HubSpot (Agent Builder > MCP Servers > Postiz
MCP) and then clears the param. returnUrl brings someone who has to sign in
first back to it. Two new strings in en and the 14 Lingo locales.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The frontend build fetched Plus Jakarta Sans from Google Fonts at build
time, and a few CI builds failed with "Can't resolve
'@vercel/turbopack-next/internal/font/google/font'". The font is now
loaded with next/font/local from woff2 files in the repo, so the build
no longer depends on Google.
The files are the variable fonts from tokotype/PlusJakartaSans commit
18d1cd2f (the same files google/fonts ships, SIL OFL 1.1, license
included), converted to woff2. All four layouts share one 500-700
upright + italic definition.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Since early October the Graph API rejects total_video_impressions on the
video_insights edge with "(#100) The value must be a valid insights
metric", on reels and feed videos alike and on every API version. The
metric is requested together with the others in one call, so the whole
request fails and videoPostAnalytics returns [] for every Facebook video
and reel post. Production logs show 100+ such rejections per hour.
Drop the metric and its Impressions case; the remaining metrics
(total_video_views, total_video_reactions_by_type_total and the reels
ones) are still accepted individually and together.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
VK asked integrations to move from the vk.com hosts to vk.ru
(api.vk.ru, id.vk.ru) by 2025-09-30. Switch the authorize, token,
user_info and API method URLs; the post releaseURL links stay on vk.com.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Pinterest's pin creation endpoint intermittently answers with
{"code":2787,"message":"Sorry! Something went wrong on our end."}.
It was unmapped in handleErrors, so the post failed immediately as a
non-retryable BadBody with the 'Unknown Error' placeholder, even though
identical requests succeed minutes later.
Map it to a retry (3 attempts, 5s apart) with a curated message so a
short Pinterest hiccup no longer fails the post, and if it persists the
user sees what actually happened instead of 'Unknown Error'.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The connect command input had a controlled value without onChange,
which made React log a console error when opening the Telegram
connect modal.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Instagram rejects a media container when a collaborator handle cannot be
tagged, with error_subcode 2207018 ("Invalid user id"). handleErrors had no
branch for it, so it fell through to the generic BadBody fallback and the
customer's failure email read "An error occurred while posting to Instagram:
Unknown Error", with nothing pointing at the collaborator tag.
Map 2207018 to a bad-body message naming the three real causes: wrong handle,
private account, or collaborator invites disabled. instagram.standalone
delegates to this handleErrors, so both channel types are covered.
Meta's own error_user_msg is not surfaced instead: it comes back localized to
the connected account's language (Hebrew, in the reproduction), so it is not
usable as user-facing text.
Verified against a live Instagram channel by creating a media container with
the customer's exact handle: pre-fix "Unknown Error", post-fix the new message.
The video is already live when thumbnails.set is rejected, but the post
ended in ERROR with no release URL, so users re-uploaded duplicates. The
provider now throws PublishedWithError with the video id and URL; the post
activity persists them through PostsService.updateResolvedRelease (the
existing org-scoped repository write, state untouched) and rethrows
BadBody, so every workflow version still marks the post ERROR while the
calendar can link to the live video.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The youtube.thumbnail error mapping said 'Your account is not verified',
which customers read as their Postiz account/channel connection - one
customer disconnected and re-added their YouTube channel trying to fix
it. The video is actually live on YouTube in this scenario; only the
thumbnails.set call is rejected because custom thumbnails require a
phone-verified YouTube channel.
The new text states the video was published, names the real requirement
(phone-verified channel, youtube.com/verify) and tells the user how to
set this video's thumbnail manually in YouTube Studio.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Meta returns subcode 1772107 ("Instagram Comment Not Added") from the
first comment call after the post itself was published. It fell back to
"Unknown Error", which hid it from the calendar tooltip and suggested
the whole post failed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The frontend proxy stored the team-invite token (org cookie) and the
selected-org cookie (showorg) for only 15 minutes. An invitee who opened
the invite link but finished signing up later lost the invite, even
though the link itself is valid for 2 days since #1643, and a member who
joined via the link fell back to the default org after 15 minutes.
Complementary to the invite token expiry extension in #1643: the org
cookie now lives 2 days to match the link, and showorg gets the same
365-day lifetime the backend already uses for it.
On a same-plan Monthly/Yearly switch the stored tier already matches
before the webhook, so the first poll wrote a stale snapshot that
reverted the period toggle and could restore cancelAt. showConfirmedPlan
now waits until both the stored tier and period match the change; the
webhook writes tier, period and cancelAt in one update.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The in-place plan change only updates Stripe, the stored plan follows
with the webhook, so refetching the credits right away still returned
the old plan's. The billing cycle card now takes its plan from the
stored subscription and, after a plan change, polls /user/subscription
until it matches, then writes the three credit counts and the
subscription into the cache together. Any failure or a webhook slower
than about 10 seconds leaves the previous values in place.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>