The link-card cache stored a LinkCardMetadata object, which does not round-trip through the Redis cache driver — a cache hit came back as __PHP_Incomplete_Class and 500'd the preview endpoint. Cache the plain array (toArray) and rebuild the DTO via a new fromArray(). Primitives round-trip cleanly through every driver. The cache test now asserts a primitive is stored and a hit reconstructs the DTO; the array cache driver used in tests hid the bug because it never serializes.
Compute the bare display host once in LinkCardMetadata (via Laravel's Uri::host + Str::chopStart) and return it as card.domain, so the LinkCard component renders it directly instead of parsing the URL client-side. The component is now purely presentational.
Add config('trypost.security.allow_private_network') (env TRYPOST_ALLOW_PRIVATE_NETWORK, default off) so self-hosted operators can reach their own internal network; only the private-IP rejection is bypassed, scheme/host checks always apply. Add SafeHttpFetcher::guardedRequest() and route the last unguarded user-supplied-URL fetches through it: the Unsplash/Giphy asset import, the API/MCP attach-media-from-URL download, and the OAuth avatar download. Our-own-storage reads (media crop, Bluesky media) are intentionally left unguarded so internal storage keeps working when self-hosted.
RunFetchRssNode, RunWebhookNode and RunHttpRequestNode guarded the initial URL but then followed redirects unguarded, so a public URL could 302 to an internal address. RSS now fetches through SafeHttpFetcher::get() (re-guards every hop); webhooks no longer follow redirects; the generic HTTP request node re-runs the SSRF guard on each hop via a new SafeHttpFetcher::redirectGuardOptions().
Bluesky does not hydrate link cards server-side, so build the app.bsky.embed.external embed at publish time: detect the first URL, scrape its OpenGraph metadata, and re-upload the og:image as the card thumb. Works for web, API and MCP. Adds a posts/link-preview endpoint so the editor renders the card live. The thumb download is SSRF-guarded and does not follow redirects.
- Assert GeneratePostContentRequest rejects a prompt over the shared max, so the
editor's limit is pinned explicitly (not only implied by the create wizard).
- Assert the create wizard accepts a prompt at exactly the minimum length,
complementing the below-minimum rejection.
The counter added earlier drifted from the backend in two ways: it counted
UTF-16 code units over the raw (untrimmed) value, while the backend measures
Unicode characters (mb_strlen) over the trimmed value that is actually sent —
so emoji or trailing whitespace could falsely turn the counter red and block
the button. The 2000 limit was also copied into three places, and the wizard's
frontend `>= 3` minimum had no backend counterpart.
- Add App\Support\AiPromptRules as the single source of truth for the prompt
bounds; both StartPostCreationRequest and GeneratePostContentRequest use it.
- Add min:3 to the create wizard endpoint so front and back agree (the editor's
generate-content flow keeps `required` — it has no counter to mirror).
- Count code points over the trimmed value in AiPostWizard so the counter and
the submit gate match what the backend validates, matching AltTextDialog.
- Cover min/max/boundary in PostAiCreateTest.
- Extract estimatedDecodeMemory() + a MAX_DECODE_MEMORY_BYTES constant so optimizeImage (fallback) and the crop/fit guard (throw) share one estimate instead of duplicating the formula and threshold.
- Correct the cropFailureException docblock: it now covers download, crop, and story-fit failures, not just downloads.
- Add a Facebook crop process-failure test and an Instagram cropped-temp-leak test so the crop path matches the fit path's error coverage.
- Guard MediaOptimizer::fitToCanvas and cropToAspectRatio against huge-dimension sources (getimagesize budget check) so they fail cleanly instead of exhausting GD memory.
- Fit and crop now translate decode/process failures into a clean InstagramPublishException and remove their temp files via finally; the two paths are symmetric.
- publishStory reads the story canvas dimensions via data_get.
- Previews: restore the IG empty-state background, align the autoFitsImage suppression predicate to isImage, drop the explanatory comment, and use single-quote imports.
- Tests: real end-to-end story fit, undecodable/download/container failures, the memory guard (fit and crop), temp-file cleanup, blurred-background pixel assertions, and the aspect-ratio warning suppression.
The regenerate job replaced the media item without carrying its meta, silently dropping the user's alt text (and slide metadata) from the persisted post.
Copy meta from the freshly-locked post row (not the pre-render snapshot) inside the transaction, so an alt edit made while the multi-second render runs is kept rather than overwritten.
Publishing:
- Only send alt text for images (isImage guards on LinkedIn, X, Discord, Mastodon); never inject altText into video/document payloads.
- X sets alt via a best-effort media/metadata call so a metadata failure no longer blocks the tweet.
Validation:
- Validate media alt_text with a closure on media.*.meta so width/height/duration/slide_* survive a post update (Laravel's excludeUnvalidatedArrayKeys was stripping them).
- Add ALT_TEXT_MAX_LENGTH constant, a proper string-type error, and a localized attribute name.
Media attach (REST + MCP):
- Support per-image alt on attach-media-from-url via structured urls: [{url, alt?}] and on the MCP upload tool via an optional alt; alt is stored only for images.
- Carry submitted meta onto hosted external-URL media so alt is no longer dropped.
Composer:
- Alt-text dialog disables Save and reddens the counter over the limit, counting code points of the trimmed value to match the backend.
- Autosave shows 'Saved' only on a successful response; the lightbox alt overlay renders for images only.
Adds unit, feature, MCP, and browser tests covering every path above.
Centralize the per-platform mb_substr truncation that every publisher was
repeating into MediaItem::altTextFor(Platform), delete each publisher's
private altFor() helper, and clarify the Platform::altTextMaxLength()
docblock so it doesn't imply Instagram's documented 1000-char cap is a
guess. Add the assertions review flagged as missing: LinkedInPage/
InstagramFacebook alt-text caps, non-string and literal-"0" alt_text
normalization, altTextFor() truncation/unsupported-platform behavior,
Mastodon's no-description-part case, and the public API's accept/reject
path for media.*.meta.alt_text.
Two hardening fixes for the paid first month:
- FirstMonthCheckoutDiscount throws when the paid first month is enabled
but STRIPE_FIRST_MONTH_COUPON_ID is unset, instead of silently charging
every new customer the full price with no discount.
- Guard workspace store() with the same active-subscription check create()
already applies, so a direct POST can't bootstrap a second billable
workspace and inflate checkout quantity past the fixed first-month coupon.
Brand autofill on the workspace settings page already captured the site logo
and rendered a preview beneath the URL, but the update flow never persisted it.
The store flow attached it via LogoAttacher; the update flow was missing all
three legs: the form field, the request rule, and the controller attach.
- BrandTab: add logo_url to the useForm payload so autofill can set it and the
form submits it.
- UpdateWorkspaceRequest: validate logo_url (nullable url) — FormRequest strips
any unvalidated key, so without a rule it was silently dropped.
- WorkspaceController::updateSettings: pull logo_url out of the validated data
(it is not a column) and attach it through LogoAttacher, mirroring store.
- Extract ContentLanguageOption into @/types and use it for availableContentLanguages
across BrandForm, BrandTab, Brand, Create, and the LanguagePicker options, so the
englishName field that drives search is visible to TypeScript instead of being
dropped silently by the pass-through prop types.
- Add BrandAnalyzerTest: assert the language schema enum equals the full 15-language
set (guards against it shrinking back to a hardcoded subset behind ::fake()) and
that instructions() lists every code.
- Isolate the "LLM language wins" autofill test by declaring the page as `en` while
the LLM returns `de`, so it actually proves mergeLlm precedence instead of both
paths agreeing.
- Cover SetLocale's cookie side effect: the default locale cookie is set on an
invalid/absent cookie and left untouched for a valid locale.
- Brand-analyzer prompt now lists every supported language instead of only
en/pt-BR/es, so onboarding autofill can detect the 12 added languages. The
backtick-formatted list is built in BrandAnalyzer::instructions(), keeping the
Blade clean and the enum free of prompt presentation.
- Translate the delete-confirmation keyword for el/ja/zh/ar (the four locales
that still shipped the English "delete").
- Make the language and font comboboxes RTL-correct (logical ms-* instead of
physical ml-*), and let the language combobox be searched by English name via
a visually-hidden label (ContentLanguage::options() now exposes englishName).
- Correct the ContentLanguage class docblock: the enum is also the source of
truth for the UI locales' text direction.
Tests: SetLocale middleware dir/RTL, isRtl and the full 15-language
englishName/label match arms, LLM language detection beyond en/es/pt-BR, and
store-path persistence of a non-default content language plus rejection of an
unsupported one.
Extend the homepage-language detection dataset to assert all 15 supported
languages resolve from their <html lang> subtag, and pin the primary-subtag
matching so a malformed tag ("english") no longer resolves via a two-letter
prefix.
Register the 12 additional languages (fr, de, it, nl, pl, el, ja, ko, zh,
ru, tr, ar) as available UI locales so the language switcher and the API
accept them, keeping the set in lockstep with the ContentLanguage enum.
- config/languages.php lists all 15 UI locales with their native names.
- ContentLanguage::isRtl() drives the document `dir`; SetLocale shares it
to the Blade root and HandleInertiaRequests shares it to Inertia, and
app.ts mirrors it on SPA navigations so RTL locales lay out correctly.
- dayjs imports the 12 new locales so dates localize instead of falling
back to English.
- A LocalizationParityTest guards against key drift: every locale must
ship every base translation file with exactly the keys of lang/en.
The new content-language options were hand-duplicated across request
validation, the UI picker, and homepage detection, while the brand
analyzer's structured-output enum and the AI image prompt's language
name still only knew about en/pt-BR/es. That left autofill unable to
detect the new languages and made image text fall back to English for
them.
Introduce App\Enums\Workspace\ContentLanguage as the single source of
truth and derive every site from it:
- Store/UpdateWorkspaceRequest validate against ContentLanguage::values()
- BrandAnalyzer's language enum uses ContentLanguage::values()
- AiImageClient::languageName() resolves via the enum's englishName()
- HomepageMetaExtractor detects through ContentLanguage::fromHtmlLang()
- BrandForm consumes availableContentLanguages from the backend, like
availableFonts/availableImageStyles, instead of a hardcoded list
Also fix two labels: nl "Nederlandse" -> "Nederlands", zh -> "中文".
refreshThenVerify only guarded refreshToken(); the verify that followed was
outside the try, so the sub-commit window where a lock-skipped refresh reloads a
not-yet-persisted token surfaced as a 401 there and falsely disconnected a
still-usable single-use-refresh_token account (X/LinkedIn). The refresh and the
verify that follows it now share one recovery: on a TokenExpiredException,
reload and, if a concurrent refresh has since persisted a fresh access_token,
verify with it instead of giving up.
TokenRefreshClient classified every non-5xx/429 refresh failure as a dead
token. Meta returns rate-limit (code 4/17) and transient (code 1/2) errors as
HTTP 4xx with type OAuthException, so a throttled proactive refresh was
disconnecting still-valid Instagram/Threads tokens — and the wider 24h/15-min
refresh cadence raised the odds of hitting it.
Classification now keys on error code 190 (the signal the publish exceptions
already use): only a genuine 190 disconnects; every other Meta 4xx is treated as
transient (PlatformUnavailable) and retried next cycle. The same over-broad
type-based check in verifyInstagram/verifyFacebook/verifyThreads is replaced
with the shared Meta\GraphError helper so verify and refresh agree.
A null token_expires_at drops an account from every refresh path (the
cron's whereNotNull filter and the is_token_expired / is_token_expiring_soon
checks all treat null as "nothing to do"), so the token silently lapses.
Threads could persist null two ways: the long-lived exchange failing at
connect (kept the ~1h short-lived token) — now fails the connect instead;
and a refresh response omitting expires_in — now defaults to 60 days for
both Instagram and Threads, matching the X refresh convention.
Extension-model tokens (Instagram/Threads) can't be refreshed once they
expire, so the shared 30-minute cron window left only a ~15-minute buffer
against queue backlog on the default queue — and a lapse forces a full
reconnect. Rotating platforms go through verify() and won't rotate a
still-valid token, so they keep the tight 30-minute window; extension
platforms now get a 24-hour lead via a per-model query.
The extend-while-valid path is shared by Instagram and Threads via
Platform::extendsAccessTokenOnRefresh(); add the Threads sibling of the
Instagram job test so both extension-model platforms are locked down.
Refs #126
Cold review caught a regression from the two previous commits. Instagram and
Threads use long-lived tokens refreshed by EXTENDING the access_token itself
(grant_type=ig_refresh_token / th_refresh_token) — they have no separate
refresh_token and CANNOT be refreshed once expired. The anti-over-rotation rule
("only refresh a token once it's actually expired") is right for rotating
single-use refresh_token platforms but wrong for these: it left IG/Threads
tokens to lapse, after which the extend call fails and the account disconnects
(~every 60 days).
Gate the anti-rotation on the platform's refresh model:
- Platform::extendsAccessTokenOnRefresh() — true for Instagram/Threads.
- SocialAccount::needsProactiveTokenRefresh() — expired for rotating platforms,
OR expiring-soon for extension platforms (restores isTokenExpiringSoon).
- RefreshSocialToken extends (refreshToken) extension-model tokens while still
valid, and verifies (access-token-first) rotating ones.
- All 23 publisher/analytics pre-checks now use needsProactiveTokenRefresh().
Tests: proactive job extends a still-valid Instagram token; a model test covers
the rotating-vs-extension branching; existing X/LinkedIn anti-rotation tests
are unchanged.
Refs #126
Every publisher and analytics service proactively refreshed the token when
it was expired OR merely "expiring soon" (within 15 min), calling
refreshToken() directly. For X (and other single-use-refresh providers)
that rotated a perfectly valid access_token whenever an operation ran in the
token's final 15 minutes — the same needless rotation that breaks the
refresh_token chain and disconnects accounts.
Narrow every pre-check to refresh only when the token is actually expired. A
still-valid token is used as-is; if it expires mid-operation the existing
reactive retry (PublishToSocialPlatform) refreshes and retries.
- Drop `|| is_token_expiring_soon` from all 23 publisher/analytics pre-checks.
- Remove the now-unused `isTokenExpiringSoon` accessor (no references remain
anywhere in the repo).
- The reactive retry path (AbstractLinkedInPublisher::retryWithRefresh) and
the expired-token path are unchanged.
Refs #126
X OAuth2 refresh tokens are single-use: each refresh rotates the pair and
invalidates the previous refresh_token, and reusing a rotated one kills the
whole family. Three things made this fragile and disconnected accounts far
more often than necessary:
- The proactive refresh job called refreshToken() directly, bypassing the
access-token-first guard in verify() and rotating on every run.
- RefreshExpiringTokens used a 2h window on an hourly schedule — equal to the
2h access-token lifetime — so every X account was rotated every hour even
while its token was still valid.
- A single 4xx refresh failure disconnected the account without checking
whether a concurrent refresh had already persisted a working token.
Changes:
- RefreshSocialToken now routes through verify() (access-token-first), so it
only rotates when the access_token is actually invalid.
- Shrink the proactive window to 30m and run the command every 15m, so the
window still covers the run interval but rotation happens near real expiry.
- verify() tolerates the lost-rotation race: on a 4xx refresh, reload and
verify with a concurrently-refreshed token before marking TokenExpired.
Refs #126
trackBeginCheckout was defined but never called, so the begin_checkout
(GTM/dataLayer) and checkout.started (PostHog) events never fired. Wire
it into the onboarding Connect submit handler, before the redirect to
Stripe, and pass the workspace plan from the controller so the event
carries plan name + interval.
When a workspace is created without an explicit content_language,
`data_get($data, 'content_language')` returns null, array_filter drops the key,
and the column falls back to its DB default of 'en'. So AI content and
notifications came out in English even when the app was running in another
locale.
Default to `app()->getLocale()` instead, so a new workspace inherits the user's
current language. An explicit content_language still wins.
Story images that aren't 9:16 were clipped by Instagram. They are now
fitted onto a 1080x1920 canvas — the image is contained and a blurred,
darkened copy of itself fills the letterbox gaps, so nothing is cropped
and the background color adapts to the image.
The fit happens at publish time (the hosted copy lives in social-crops/),
and the post editor preview now renders the same blurred-background fit
for stories, so the user sees exactly what will publish. The aspect-ratio
warning is suppressed for story images since they're auto-fitted.
Instagram only — Facebook stories are video-only in our flow.
MediaAttacher::download() only caught RuntimeException, but a connection-level
failure throws Illuminate\Http\Client\ConnectionException (extends Exception, not
RuntimeException) — e.g. a slow/unreachable proxy hitting the 20s timeout. On the
API hosting path that propagated as a 500 (not the promised 422), skipped the
batch rollback (orphaning an already-hosted item), and leaked the temp file.
Catch Throwable so any fetch failure returns null → the caller rejects cleanly
with 422 and rolls back. Also hardens the existing MCP/REST attach-from-url paths
(a timeout there now reports a failed URL instead of 500).
- External URL that downloads (200) but isn't a supported media type → 422,
nothing persisted (the type-rejection branch, distinct from a download 404).
- Mixed batch: an already-hosted item + an external URL both succeed → both kept
in order, only the external one creates a Media row.
Final-review follow-ups:
- MediaAttacher::resolveInlineMedia now deletes the media it hosted in this call
when any item fails, so a mixed [good, bad] batch no longer orphans the good
item's Media row + file while the request is correctly rejected with 422. Makes
the create/update media resolution truly all-or-nothing.
- PostMediaRules: keep source/source_meta on both contracts (the API previously
passed them through with no item rules — don't silently drop them) so the media
item shape is uniform; only id/path/url differ by contract.
- Make MediaAttacher::fetchToWorkspace private (no external callers).
- Test the partial-batch rollback (no Media, no post persisted).
Cold-review follow-ups on the PR:
- Trim the oversized docblocks/inline comments added across the API controller,
MediaAttacher, Post, the publish job, and the X publisher to one line (keeping
the @param/@return array-shape annotations).
- XPublisher::chunkedUpload now accepts ?string $mediaCategory and only sends
media_category when present — getMediaCategory() can return null, so the strict
string param was a latent TypeError (unreachable on X today, removed anyway).
- Fix MediaAttacher docblocks: the file imports Type as MediaType, so the
@param array<Type> annotations didn't resolve — now array<MediaType>.
- Tests: cover the failed() job hook genericizing a raw error, and X failing
cleanly (XPublishException) when media can't be downloaded.
After the failure-email sanitization, any non-SocialPublishException is shown to
the user as a generic line. The deliberate, user-facing throws in the X,
Facebook, Threads, and LinkedIn (+Page) publishers were plain \Exceptions, so
they'd have been genericized too — losing actionable messages like 'Unsupported
media type for Facebook' or 'X posts require either text or media'.
Convert those throws to their platform's SocialPublishException with a vetted
userMessage + category, so the publish failure surfaces a specific, safe message
(and the catch-all's generic line is reserved for genuinely unexpected errors).
Left as plain \Exception on purpose: the download helpers in Bluesky and Mastodon
(caught internally, returned as null — they never reach the email), and the
shared content-length guard in HasSocialHttpClient (no per-platform exception in
the trait, message is path-free, and it's already validated upstream).
The post-failure email renders each platform's stored error_message verbatim.
On an unexpected publish error the job's catch-all (and the job-failed hook)
stored the raw exception message — for a PHP TypeError that includes the server
file path (.../releases/<id>/app/Services/Social/XPublisher.php on line 130),
which then reached the customer's inbox.
Only our own SocialPublishException carries a vetted, user-facing message.
Every other throwable (engine errors like TypeError, or library exceptions such
as Intervention's decoder errors which embed temp paths) is now replaced with a
generic line; the raw detail stays in the logs. Token-expired and rescheduled
paths are unchanged.
Note: publishers that still throw plain \Exception for user-facing reasons
(some X/Facebook/Threads/LinkedIn cases) will now show the generic message for
those; converting them to typed SocialPublishExceptions to restore specific
copy is a worthwhile follow-up.
