Audit found three behaviors with no explicit assertion:
- Job is dispatched with a 10-minute delay (would silently regress if
the duration changed). Uses Carbon::setTestNow + Bus::assertDispatched
inspecting \$job->delay.
- error_context.last_attempt_at is recorded at the moment of failure.
- updatePostStatus does NOT finalize the parent Post while any of its
platforms is in Retrying — covers the central invariant of the
feature (the post must stay Publishing until every platform lands
in Published or Failed).
- A platform currently in Retrying transitions to Published when the
next attempt succeeds — proves the loop terminates.
Before: a scheduled post hitting a platform outage was marked Failed —
user had to manually retry. Now the job reschedules itself for 10
minutes later and the PostPlatform shows status "Retrying". Loops
indefinitely until the platform accepts the post.
- Adds PostPlatformStatus::Retrying (existing string column, no migration)
- PublishToSocialPlatform: PlatformUnavailable catch now calls
rescheduleForRetry() which (a) updates the row to Retrying with
retry_count + next_attempt_at in error_context, and (b) dispatches
itself with a 10-minute delay. updatePostStatus() naturally leaves
the parent Post in Publishing because Retrying is neither Published
nor Failed.
- Same treatment for the retry-refresh edge case (publisher throws
TokenExpired, refresh subsequently fails with PlatformUnavailable).
- i18n + frontend status config updated (en, pt-BR, es) for both
posts.status.retrying and posts.edit.status.retrying.
- Tests: 3 new tests covering the dispatched job, the edge case path,
and retry_count increment across attempts.
Edge case from the prior commits: if the publisher throws TokenExpired
(401 path), PublishToSocialPlatform attempts refreshAccountToken() to
recover. That internally goes through ConnectionVerifier::verify, which
can now raise PlatformUnavailable (5xx). The old catch (\Throwable)
swallowed it but the loop still fell through to markAsTokenExpired —
meaning a transient platform outage during a retry could still flip the
account to expired.
Adds an explicit PlatformUnavailable catch in the retry block: marks
the post failed with category platform_unavailable and breaks before
touching the account status.
Same rule applied to production code in earlier commits now applies to
tests: Http::fake patterns and assertions read from
config('trypost.platforms.*.oauth_api' / '.api' / '.default_service')
instead of hardcoded strings. Hardcoded URLs in tests drift silently
when the config changes.
Also documents the rule in CLAUDE.md under "External Service URLs" so
new code (and tests) start in the right place — only the host comes
from config, path/RPC segments stay inline next to the call.
- TokenRedactorTest (7 unit tests): all four regex patterns, multiple
secrets in one body, null input, and the no-op pass-through case.
Guards against silent regression of the redaction regexes (which
previously drifted across three duplicated copies).
- ConnectionVerifierTest: HTTP 429 during refresh raises
PlatformUnavailableException, not TokenExpiredException. Locks in
the rate-limit-as-transient behavior added when consolidating the
refresh logic.
The previous commits in this PR closed the loophole on the hourly /
daily token-refresh jobs. The same loophole remained on the publish
path: every per-platform publisher (LinkedIn, X, YouTube, TikTok,
Threads, Instagram, Pinterest, Bluesky and their Analytics siblings)
has its own refreshToken() called before publishing a scheduled post,
and all of those treated any non-2xx as TokenExpired — including 5xx.
Result before this commit: a Bluesky outage that coincided with a
scheduled publish would mark the account as expired and fail the post.
Changes:
- Route every refreshToken() in the 16 publisher / analytics classes
through TokenRefreshClient::for(Platform::X)->send(...).
- TokenRefreshClient now also fills platformErrorCode from the HTTP
status and pulls error_description / error.message from the JSON
body, preserving the richer info LinkedIn / X / TikTok / Pinterest /
Threads used to put on their TokenExpiredException.
- PublishToSocialPlatform catches PlatformUnavailableException
explicitly: the post is marked failed (category: platform_unavailable,
with http_status in error_context) but the account stays Connected.
No retry inside this job — the scheduler reattempts the next run.
Test added: publish flow does NOT mark account expired when the
publisher throws PlatformUnavailable. Full suite: 1569 passing.
LinkedInPageAnalytics and MastodonAnalytics were the only two of the 11
files refactored to read OAuth host / default instance from config that
had zero test coverage. Adds smoke tests that assert the HTTP request
hits the configured URL, so a typo in the config key (e.g. linkedin.api
vs linkedin.oauth_api) would now fail loudly.
Review follow-ups:
- verifyMastodon was the last hardcoded host left after the PR moved
LinkedIn/YouTube/Bluesky to config. Adds trypost.platforms.mastodon
.default_instance (env MASTODON_DEFAULT_INSTANCE) and reads from it.
- refreshToken() docblock now declares @throws PlatformUnavailableException
(the whole point of the PR was missing from its contract).
- Strip the new explanatory comments inside catch blocks and tests —
rationale lives in the commit / PR, not inline. The two comments
inside empty `catch (TokenExpiredException) {}` blocks stay because
there the comment is the only thing telling the reader why the
exception is swallowed.
When a provider's API was down (5xx, timeout, DNS), the hourly
RefreshSocialToken job and daily VerifyWorkspaceConnections job were
treating it as "token revoked" and emailing the user to reconnect.
Bluesky going offline triggered false-positive disconnect notifications
because Bluesky access tokens are short-lived (2h) so every hourly
refresh failed during the outage.
- New PlatformUnavailableException: API unreachable / 5xx, transient.
TokenExpiredException stays for 4xx (token is provably bad).
- New TokenRefreshClient: normalizes failure semantics for OAuth
refresh HTTP calls across all providers. Takes a Platform enum so
typos fail at compile time and the user-facing label comes from
one source.
- ConnectionVerifier: all 8 refresh*Token methods route through the
new client. Hardcoded OAuth URLs (LinkedIn, YouTube) and Bluesky's
default PDS host moved into config/trypost.php alongside the
existing per-platform entries.
- RefreshSocialToken job: PlatformUnavailableException → log warning
and stop. Do NOT markAsTokenExpired, do NOT notify the user. Next
scheduled tick retries.
- VerifyWorkspaceConnections job: PlatformUnavailableException from
the inner refresh propagates and is treated as a transient skip.
The post editor lost the Pinterest board picker during a UI rewrite,
causing scheduled posts to fail in production with 'Pinterest board_id
is required'. This restores the picker and locks the contract with
validation + tests so the regression cannot happen silently again.
Backend:
- PostController: pinterestBoards is now Record<account_id, Board[]>
(mirrors the TikTok creator-info pattern); supports multi-account.
- UpdatePostRequest: 'platforms.*.meta.board_id' rule + after-validator
rejects Publishing/Scheduling Pinterest posts without board_id.
Frontend:
- PinterestSettings.vue: Combobox board picker with empty-state warning;
emits update:meta with board_id.
- ScheduleTab / PostEditorSidebar / Edit pass pinterestBoards down by
social_account_id.
Tests (6 new):
- UpdatePostRequestTest: rejects publishing/scheduling without board_id
across pin/carousel/video pin; allows draft without board_id;
pinterest error doesn't block sibling platforms in multi-platform.
- PinterestPublisherTest: publisher throws for carousel + video pin
when no board_id (existing image-pin case kept).
1542 tests passing.
- StripeEventListener::handleSubscriptionCreated nulls account.trial_ends_at
when a Stripe subscription is created. Prevents 'Trial' badge from
lingering for users who convert mid-generic-trial to paid.
- Drop unused trialDays global Inertia prop (no frontend consumers after
/subscribe redesign).
Tests added (10 new, 0 regressions, 1533 total):
- AccountTest: isOnTrial + activeTrialEndsAt across 4 scenarios
(no trial, generic only, subscription only, both — subscription wins)
- StripeEventListenerTest: subscription created clears generic trial
- TrialMiddlewareAccessTest: trialing-with-card subscription passes
- BillingControllerTest: index exposes onTrial/trialEndsAt for the 3
trial states (generic-only, subscription-only, paying); subscribe
page no longer exposes trialDays prop
- BillingController::index reads onTrial from Account::isOnTrial() (covers
generic trial without a Stripe subscription) and falls back to
account.trial_ends_at when no subscription exists. Vue page already had
the badge + 'Trial ends' UI wired — just needed the right props.
- Drop default trial_days from 8 to 7 for consistency with messaging.
New signups land on a 7-day generic trial (Cashier trial_ends_at) without
a Stripe customer or subscription. Account is on Starter plan limits during
the trial. After 7 days, EnsureAccountReady redirects to /subscribe per the
existing flow.
- CreateUser sets account.trial_ends_at and plan_id = Starter
- EnsureAccountReady allows access when subscribed() OR onGenericTrial()
- Account::isOnTrial() includes generic trial check
Existing users unaffected: paying users have a subscription;
never-paid users continue redirecting to /subscribe.
The LinkedIn Page connection has a two-step OAuth: first the
`callback` stashes the Socialite user in `linkedin_page_pending` and
redirects to the page picker, then `select` finalizes by writing the
chosen organization to social_accounts. The pending payload was missing
`approved_scopes`, and both finalize paths (`update` for reconnect,
`updateOrCreate` for first connect) never wrote the `scopes` column.
Result: every LinkedIn Page account had `scopes = NULL` in the DB,
the publish-time scope check saw `w_organization_social` as missing
and blocked every post with 'Missing permissions. Please reconnect
your account.'
Fix: stash `approved_scopes` in the session payload, then in both
finalize paths persist it with the same comma-split treatment used by
the LinkedIn personal controller (the LinkedIn-OpenID provider has the
same separator quirk — granted scopes come CSV-joined inside a
single Socialite array element).
Test: `linkedin page select splits comma-separated approvedScopes
before saving` covers the persist + split path.
Adds an explicit test asserting that when a workspace has multiple
LinkedIn users (e.g., the owner plus a client teammate), syncing
tokens from one user's accounts never touches accounts admin'd by a
different LinkedIn user — even though all live in the same workspace.
The behavior is already correct by virtue of the
`admin_user_id` / `platform_user_id` match in the synchronizer's
query, but the property is now part of the test contract instead of
emergent.
LinkedIn and Pinterest's OAuth providers return the granted scope list
joined by comma (LinkedIn) or space-in-one-element (Pinterest), but
Socialite's scope splitter doesn't match either, so 'approvedScopes'
lands as a single-element array containing the whole list:
LinkedIn: ['email,openid,profile,r_basicprofile,w_member_social']
Pinterest: ['boards:read boards:write pins:read pins:write user_accounts:read']
That breaks the publish-time scope check in PublishToSocialPlatform
(array_diff does exact string compare), surfacing as
'Missing permissions: w_member_social. Please reconnect your account'
even though the scopes were actually granted at the provider.
Fix is inline at each callback — re-split before saving. Each provider
has its own quirk (LinkedIn = comma, Pinterest = space), so each
controller handles its own separator.
Tests added: callback splits the joined approvedScopes into individual
tokens for both providers.
Adds a combobox-style filter to the posts index toolbar so users can
narrow All / Scheduled / Posted / Drafts views by one or more labels.
- `PostController::index` accepts `?labels[]=<id>` and applies
`whereHas('labels', whereIn(...))` (OR semantics across selected labels).
Workspace labels are exposed to the page (sorted by name) and the
selected set comes back under `filters.labels`.
- New `LabelFilter.vue` component reuses the existing Popover + Command
pattern (matching `FontPicker` in the Brand settings page). Trigger
renders the selected `LabelBadge`s inline (mirroring how each post row
already displays its labels): 1-3 shown directly, 4+ shown as the
first three plus a "+N" overflow indicator. Clear button has a
tooltip and `cursor-pointer`, and stops `click`/`pointerdown`/
`mousedown` so it doesn't reopen the Popover.
- Existing search debounce is shared with the new label watcher via a
single `buildFilterUrl` helper. URL is updated with `preserveState +
replace` so the back stack stays clean.
- i18n in en / pt-BR / es: `filter_by_label`, `label_search_placeholder`,
`no_labels`, `clear_label_filter`.
Tests: 4 new index tests covering the labels prop exposure, single-label
filter, multi-label OR filter, and blank-id sanitization. Full suite:
1509 passed, 2 skipped, 0 failed.
Five small fixes scoped to the publishReel method:
1. Replace generic \Exception on media download failure with a typed
FacebookPublishException(ServerError). The generic exception was
landing in the \Throwable catch in PublishToSocialPlatform with
category 'unknown', defeating the whole point of the social
exception hierarchy.
2. Replace handleApiError($startResponse) with a direct
FacebookPublishException throw when video_id/upload_url are missing.
The previous code passed a successful HTTP response into a method
built for error responses — fromApiResponse would fall into the
default arm and surface 'An unknown Facebook error occurred.'
ironically reintroducing the same bad UX we just spent the day
fixing.
3. Drop the four redundant Log::error calls before handleApiError.
FacebookPublishException::fromApiResponse already pulls the FB
error code/subcode/message into platformErrorCode + userMessage,
and the downstream catch in PublishToSocialPlatform::handle logs
the exception anyway (Nightwatch picks that up). Same pattern as
the X cleanup in PR #29.
4. Stream the upload body via fopen() resource instead of
file_get_contents(). Eliminates loading the whole video into memory
for large reels.
5. Replace \@unlink with unlink + Log::warning. Surfaces temp-file
cleanup failures instead of silently leaking files.
Tests:
- Strengthened the missing-upload_url test to assert the exception
message and class.
- Added a typed-exception test for the media-download failure path
(would have caught the regression where we used a generic
\Exception).
- Full suite green: 1505 passed, 2 skipped, 0 failed.
Previous attempt at the hosted-file flow (`file_url` in JSON body) hit
the rupload.facebook.com validator with HTTP 400:
'HeaderValuePredicate: Header Offset not convertable to unsigned long'
Facebook's rupload endpoint requires the `Offset` and `file_size` headers
regardless of whether the upload is local or hosted-file. The docs
describe the hosted-file path without them, but in practice rupload
rejects requests that lack them.
Switching to the well-documented local-file flow:
- Download the media to a temp file via Http sink (already a pattern
used in XPublisher for media downloads).
- POST raw bytes to upload_url with three headers:
- Authorization: OAuth {token}
- Offset: 0
- file_size: {actual bytes}
- Use mime_type from the media item as the Content-Type.
- Always cleanup the temp file in a finally block.
Tests updated to fake the media-download GET (returning bytes that the
publisher then re-uploads) and to assert on the Offset/file_size/
Authorization headers being present and correct. The existing
"cleans up temp files after reel upload" test now actually exercises
its assertion since this code path creates temp files again.
Full suite: 1504 passed, 2 skipped, 0 failed.
Production was failing every Facebook reel publish with the cryptic
'Video Upload Is Missing' / error_subcode 1363130. Root cause: our
'transfer' phase was making up its own API contract that doesn't exist.
The Reels publishing API (per Meta docs) is a 3-step flow:
1. start → POST /{page_id}/video_reels {upload_phase=start}
returns {video_id, upload_url}
2. transfer → POST {upload_url} on rupload.facebook.com
Header: Authorization: OAuth {token}
Body (JSON): {"file_url": "https://..."}
3. finish → POST /{page_id}/video_reels {upload_phase=finish, ...}
Our code was doing:
2 (broken) → POST /{video_id} on graph endpoint
Body: {video_file_chunk: '<some URL>', access_token: ...}
`video_file_chunk` is not a real parameter; the graph endpoint accepted
the request (returning 200) but nothing was actually uploaded, so the
finish phase reported 'Video was not uploaded'.
Changes:
- Capture upload_url from the start response and use it for transfer.
- POST to that upload_url with file_url in JSON body and
'Authorization: OAuth ...' header (the format docs.facebook.com
documents for the hosted-file flow).
- Bail with handleApiError if the start response is missing video_id
or upload_url so we don't silently no-op like before.
Tests:
- Existing 'can publish reel' test updated to include upload_url in the
start mock and assert the transfer call actually POSTs to rupload
with file_url + OAuth header (would have caught this bug before
shipping).
- New 'fails reel publish when start does not return upload_url' test
for the safety bail-out.
- Pre-existing 'cleans up temp files after reel upload' test updated
to match the new mock pattern.
Full suite green: 1504 passed, 2 skipped, 0 failed.
Production was returning HTTP 413 with empty body on the first APPEND
segment of every video upload to X v2 — surfacing to users as 'An
unknown X error occurred.'
Empty-body 413 is the classic signature of an edge/CDN rejection: the
X gateway is denying the request before X's application code sees it.
The X v2 reference docs say 'max chunk size: 5MB', but every canonical
reference uses 1MB:
- X's official Python quickstart: `chunk_size = 1024 * 1024`
- X's official JavaScript quickstart: `const chunkSize = 1024 * 1024`
- twitter-api-v2 (the most-used Node SDK, used by Postiz et al.):
`chunkSize: number = 1024 * 1024`
5MB plus multipart-form overhead apparently exceeds an undocumented
edge limit. 1MB is the empirically safe size everyone converges on.
Changes:
- XPublisher chunked APPEND now uses 1MB chunks. An 8MB video uploads
as 8 segments instead of 2; more roundtrips but actually succeeds.
- Set explicit Content-Type on each chunk attach (matches the simple
upload path in the same file).
- XPublishException::fromApiResponse maps HTTP 413 to
ErrorCategory::MediaFormat with the message 'Media chunk rejected
by X (payload too large).' so we don't surface 413 as 'unknown' if
it ever recurs.
Test: unit test covering the 413→MediaFormat mapping. Full suite green
(1503 passed, 2 skipped).
Three orthogonal fixes that together close the gap where social tokens
were silently aging out without ever being refreshed, then dying at the
provider when the refresh_token also got revoked.
The original failure mode: a user's X token expired because the hourly
proactive-refresh cron's smart `verify()` skip-logic kept saying 'token
still works, no need to refresh', and once the token actually expired,
the cron's WHERE clause excluded it from future runs. By the time anyone
noticed, the refresh_token at X was also gone.
(C) ConnectionVerifier: rename private `refreshTokenIfNeeded` →
public `refreshToken`. Callers that want the smart 'try
access_token first' behavior keep using `verify()`. Callers that
want a proactive refresh (the cron) call `refreshToken` directly.
(B) RefreshExpiringTokens command: drop the
`where('token_expires_at', '>', now())` filter. Already-expired
tokens now get a last-chance refresh attempt before the
refresh_token also dies at the provider. Status filter
(`Connected`) still excludes accounts already marked TokenExpired.
(D) RefreshSocialToken job: switch from `verify()` to
`refreshToken()`, and on `TokenExpiredException` call
`markAsTokenExpired` so the user is notified immediately. The lock
+ transition detection in markAsTokenExpired prevents notification
spam if subsequent cron passes also fail.
Tests:
- 3 new tests for RefreshSocialToken (calls refreshToken not verify,
marks TokenExpired on TokenExpiredException, logs warning on other
errors)
- Updated RefreshExpiringTokens test to assert already-expired tokens
are now dispatched (was previously asserted as 'should NOT')
Fills gaps surfaced in the code review:
- 3 new `markAsDisconnected` tests (mirroring the existing TokenExpired
trio). Previously the method had zero coverage despite this PR
changing its lock key.
- 1 test confirming `markAsTokenExpired($msg, notify: false)` skips
notification dispatch (used by VerifyWorkspaceConnections batch path).
- 1 test confirming `disconnected_at` is preserved when already set
(the `?? now()` branch).
- 2 end-to-end tests (one per method) that DO NOT fake the queue, so
`SendNotification::handle()` runs synchronously. Asserts the
Notification row is actually created in the DB with the correct
i18n-substituted title/body, that NotificationCreated event is
dispatched, and that AccountDisconnected mail is queued. Catches
bugs where the i18n placeholder keys (`:platform`, `:account`)
silently fail to substitute.
These tests use Http::fake, model factories, and DB — by Laravel/Pest
convention that's a feature test, not a unit test. Moving them to the
Feature suite to match the convention.
No code changes. Tests still pass: 1493 passed, 2 skipped, 0 failed.
Extends the X-only fix to LinkedIn, LinkedInPage, Pinterest, Threads, and
TikTok. Each publisher's `refreshToken` now throws `TokenExpiredException`
directly when the OAuth refresh endpoint rejects the refresh_token, instead
of routing through `handleApiError` -> `<Platform>PublishException::fromApiResponse`.
Why: `fromApiResponse` is designed for the publish API response shape and
only converts certain status codes to `TokenExpiredException` (e.g., 401).
OAuth refresh failures typically return 400 with an `error`/
`error_description` (or platform variant) body and were falling into the
generic 'Unknown' bucket, masking the real cause and skipping the proper
`markAsTokenExpired` + user-notification flow.
Dropped the redundant `Log::error` before each `throw`: the downstream
catch in `PublishToSocialPlatform::handle` already logs the exception
(captured by Nightwatch), and the rich context now lives on the exception
itself (`message` = provider's `error_description` / `error.message`,
`platformErrorCode` = HTTP status).
Tests: one regression-style test per publisher confirming refresh rejection
becomes `TokenExpiredException` instead of a generic publish exception.
Three related fixes for the failure mode where a scheduled post errors out
as 'An unknown X error occurred.' when a social account's refresh_token
was already invalidated by the provider:
1. **PublishToSocialPlatform**: fail-fast when account status is
`TokenExpired`. Previously the job tried to publish, the publisher
internally tried to refresh, the provider rejected the rotated
refresh_token, and the failure surfaced as a generic 'unknown' error
instead of a clear 'reconnect your account' signal.
2. **XPublisher::refreshToken**: when the OAuth endpoint rejects the
refresh_token (typically because it was rotated/revoked at X), log the
raw response and throw `TokenExpiredException` instead of falling
through to `XPublishException::fromApiResponse` which expects the
tweet-API response shape (`type`/`title`/`detail`) and treats
OAuth-style responses (`error`/`error_description`) as 'Unknown'.
3. **SocialAccount::markAsTokenExpired**: dispatch an in-app + email
notification (`Type::AccountDisconnected`) when an account
transitions from `Connected` → `TokenExpired`, mirroring the
existing pattern in `markAsDisconnected`. Wrapped in a lock to
prevent duplicate notifications on concurrent transitions. Accepts an
optional `notify: false` so the batch verifier
(`VerifyWorkspaceConnections`) can suppress per-account
notifications and rely on its summary email.
Wire `value`, `currency`, and `transaction_id` from Stripe Checkout Session
into the `purchase` dataLayer event so GTM can fire Google Ads Conversion
Tracking with accurate per-plan revenue and deduped transaction IDs.
- `BillingController::checkout` adds `{CHECKOUT_SESSION_ID}` to success_url
- `BillingController::processing` retrieves session lazily (closure prop,
so polling partial reloads don't re-hit Stripe) and exposes
`conversion` with `value`/`currency`/`transaction_id`
- `Processing.vue` forwards `conversion` to `trackPurchase`
- `useTracking.trackPurchase` pushes `conversion_value`,
`conversion_currency`, `conversion_transaction_id` to dataLayer +
PostHog
Two more pieces in app/Ai/ that were registered but never reached by any
production caller, same audit as the PlatformRules cleanup.
DebugGeminiRequest middleware: zero references — not registered in the
laravel/ai middleware pipeline, no agent imports it. Pure dev-time
scaffolding that got left behind.
ExtendedGeminiProvider: registered via Ai::extend('gemini', ...) so the
class IS resolvable when AI_DEFAULT=gemini, but its only override —
defaultImageOptions — is unreachable in practice:
- All text agents (PostContentGenerator, PostContentHumanizer,
PostContentReviewer, PostContentStreamer, BrandAnalyzer) only call
text generation, never defaultImageOptions.
- Image generation (AiImageClient) hardcodes OpenAI's `gpt-image-2`
model, so the Gemini provider isn't on the image path either.
Removed:
- app/Ai/Middleware/DebugGeminiRequest.php
- app/Ai/Providers/ExtendedGeminiProvider.php
- tests/Unit/Ai/Providers/ExtendedGeminiProviderTest.php
- Ai::extend('gemini', ...) wiring + configureAi() method and its four
imports (ExtendedGeminiProvider, Ai, GeminiGateway, Dispatcher) in
AppServiceProvider.
The whole app/Ai/PlatformRules/ system was registered in AppServiceProvider
but never read in production code. Confirmed via grep across app/, resources/,
config/, database/, tests/ — only consumers were:
- AppServiceProvider::configurePlatformRules() registering everything in the
Registry (which nothing then called)
- tests/Feature/Ai/PlatformRules/RegistryTest.php asserting the registrations
it had just performed
The actual AI text agent (PostContentGenerator) reads the per-platform caps
straight from the Platform enum (maxContentLength / recommendedAiContentLength)
and feeds them into the prompt template — bypassing this layer entirely.
Removed:
- app/Ai/PlatformRules/ (12 files: Contract, Registry, 10 Rule classes)
- tests/Feature/Ai/PlatformRules/RegistryTest.php
- 11 use imports + boot() call + configurePlatformRules() method in
AppServiceProvider
- Platform::contentOverflow returns int (0 = fits) via max(), drops the
nullable-int + ternary pattern; HasSocialHttpClient and the validation
rule updated to compare against 0 instead of null.
- Drop the hardcoded 63206-char limit to 10000 across all entry points
(Platform enum, MCP CreatePostTool/UpdatePostTool, FacebookRules, all
three FormRequests). Facebook's API accepts up to 63206 but nobody
writes 63k-char posts and emoji-heavy content risks overflowing the
TEXT column's 65535-byte ceiling.
- Compiled i18n JSON regenerated with the content_exceeds_platform key.
Threads posts over 500 chars were saved + scheduled successfully and only
failed inside the publish job. The frontend already showed the 537|500 badge
but `canSchedule` ignored content length, so Schedule and Post Now stayed
enabled. Backend `UpdatePostRequest` only capped at 63206 (Facebook's max),
not per-platform.
- Add `Platform::contentOverflow()` as the single source of truth and reuse it
from `HasSocialHttpClient::validateContentLength` (publish-time).
- New `ContentFitsPlatformLimits` rule applied to the `content` field on
`App\\UpdatePostRequest`, `Api\\UpdatePostRequest`, and `Api\\StorePostRequest`
via `Rule::when(...)` so drafts are not blocked.
- Rule dedupes per platform (two Threads accounts -> one error) and reports
the platform label, hard cap, and overage via i18n.
- Edit.vue feeds `contentLengthOverflows` into `canSchedule` and lists each
offending platform in `postActionTooltip` using the existing
`getPlatformLabel` resolver.
## Photo carousel support
- Adds `ContentType::TikTokPhoto` enum case (max 35 photos, 1:1 aspect,
supportsImage true, supportsVideo false) and JS mirror in content-type.ts.
- Variant pill picker (Video / Photo carousel) at the top of TikTokSettings,
mirroring the Instagram pattern. Wired through ScheduleTab to the parent
editor's existing update:platformContentType emit.
- i18n keys for variant_label / variant.video / variant.photo in en/pt-BR/es.
- Publisher: split buildPostInfo into buildVideoPostInfo (uses `title`,
TikTok cap 2200 chars) and buildPhotoPostInfo (uses `description`, cap
4000 chars; omits Duet/Stitch/AIGC since they don't apply). Removed the
no-longer-needed queryCreatorInfo() call from publishVideo/publishPhotos
— its only previous consumer (silent privacy_level fallback) is gone.
## UX Content Sharing API compliance
Per TikTok review feedback citing
https://developers.tiktok.com/doc/content-sharing-guidelines#required_ux_implementation_in_your_app
Point 1 — already satisfied (creator_info fetch + nickname display).
Point 2/4 — Music Usage Confirmation declaration is now always visible
in TikTokSettings; text changes between "Music Usage Confirmation" and
"Branded Content Policy and Music Usage Confirmation" based on toggle
state. Previously the entire `<p>` block was conditional on a brand
toggle being selected, hiding the baseline declaration.
Point 2b — privacy_level may not have a default. UI was already correct;
backend hardened: UpdatePostRequest now requires meta.privacy_level for
tiktok platforms when status is publishing/scheduled (via withValidator);
TikTokPublisher::resolveRequiredPrivacyLevel throws TikTokPublishException
(ContentPolicy category) when missing instead of silently falling back to
the creator's preferred level.
Point 2c — interaction settings now condition on content type:
- Photo posts hide Duet/Stitch (they don't apply per TikTok docs).
- Photo posts hide AIGC (also video-only).
- Video posts hide Auto Add Music (photos-only feature).
- Max-duration warning hidden when not a video post.
Source of truth is the user-selected contentType prop, not inferred
from media — ensures the UI reacts immediately to the variant pill.
Point 3a — publish button stays disabled when Disclose toggle is on
without a sub-selection (already the case via tiktokComplianceValid).
The disabled tooltip now uses the verbatim TikTok-required text "You
need to indicate if your content promotes yourself, a third party, or
both." instead of the generic "Some platform settings are incomplete..."
when the only blocker is TikTok disclosure incompleteness.
Point 3b — SELF_ONLY (Only me) privacy option is no longer filtered out
when Branded Content is checked. It is rendered disabled with a hover
tooltip "Branded content visibility cannot be set to private." plus a
persistent amber warning paragraph below the dropdown. When the user
toggles Branded Content while privacy is SELF_ONLY, the privacy clears
and a vue-sonner warning toast surfaces the change.
## Cross-cutting
- New `resources/js/enums/platform.ts` mirrors the PHP Platform enum,
used in Edit.vue (tiktokComplianceValid + tiktokDisclosureIncomplete)
and ScheduleTab.vue (all selected*Platforms computeds) to replace
string literal comparisons against `'tiktok'` / `'facebook'` / etc.
- PostPlatformFactory tiktok() state defaults meta.privacy_level to
SELF_ONLY so existing test fixtures keep passing under the new
publisher/FormRequest requirements.
## Tests
- New tests/Unit/Enums/PostPlatform/TikTokPhotoContentTypeTest.php
covering the new enum case (4 tests).
- TikTokPublisherTest: added "video uses title not description" and
"throws when meta.privacy_level missing" regression tests; renamed
two existing tests that depended on the removed silent fallback.
- New tests/Feature/UpdatePostRequestTest.php with 3 tests covering
the FormRequest's privacy_level enforcement (publish-rejected,
publish-passes, draft-allowed).
Full Pest suite: 1490 passed, 2 skipped (pre-existing).
TikTok's photo endpoint caps the 'title' field at 90 UTF-16 runes but
allows 'description' up to 4000. Today the publisher always sends the
caption in 'title', which would silently fail for any photo caption
longer than 90 chars. Split buildPostInfo into buildVideoPostInfo
(uses title) and buildPhotoPostInfo (uses description, omits Duet/
Stitch/AIGC since they don't apply to photos).
Adds the enum case TikTokPhoto = 'tiktok_photo' with maxMediaCount 35,
supportsImage true, supportsVideo false, 1:1 aspect ratio. The publisher
service already calls /post/publish/content/init/ correctly — this
enum case is the missing data layer that lets validation, UI, and
defaults treat photo posts as a first-class content type.
Replaces the vague "intro → development → conclusion" guidance in the
generator prompt with an explicit per-slide arc: open with a specific
real urgent problem, develop the idea while creating swipe motivation,
prove with concrete evidence (result/before-after/BTS/learning), and
close with a single specific next action. Includes a slide_count to
role-distribution table (2 through 6+ slides).
Adds a `role` enum field to each slide in the carousel JSON schema
(hook, development, proof, cta) so the structure is enforced via SDK
not just prompt text. Field is additive — handleCarousel still reads
only title/body/image_keywords, so consumers are unaffected.
- TemplateImageGenerator->render() now returns a typed array shape
{path: string, source_meta: array} instead of a one-off RenderedSlide
DTO. Single internal callsite, no need for a dedicated class.
- Drop the `?? 'en'` fallback on $workspace->content_language in 6
callsites: the column has a NOT NULL default of 'en' at the DB level,
so the null coalesce was dead code.
- WorkspaceFactory now seeds content_language, brand_tone, brand_font
and image_style explicitly so make() (no DB persist) produces a
complete model — DB defaults aren't applied until create().
Convert all 'foo.'.$bar style concatenations to "foo.{$bar}" interpolation
in event broadcast channels (PostCreated, PostDeleted, PostCommentCreated,
PostPlatformStatusUpdated, NotificationCreated) and the related event
tests.
Document the convention in CLAUDE.md so new code follows the pattern.
Core changes:
- Replace Unsplash slide pipeline with gpt-image-2 via Laravel AI SDK.
New AiImageClient builds prompts from a Blade template seeded by the
workspace's ImageStyle enum, content language, brand color (mapped to a
human-readable name via HexColorName helper) and brand description.
- Drop Template B from TemplateImageGenerator: every slide now renders as
Template A (full-bleed photo + bottom gradient + white/grey overlay).
Removes renderTemplateB, roundCorners, blendHex, ensureContrast and the
closing-slide pipeline.
- StreamPostCreation creates the Post directly and dispatches
PostCreationReady with post_id; the wizard kills its preview step and
redirects straight to the post editor on completion. Finalize endpoint
removed.
- New Workspace.image_style enum field with an 8-option visual picker
shared by /workspaces/create and /settings/workspace/brand via a single
BrandForm component (autofill is a prop). 8 sample webp thumbs ship
under public/images/branding/image-styles/.
- Media items gain optional source ('ai'|'unsplash'|'giphy') and
source_meta (recipe needed to regenerate AI images later); the gallery
picker tags Unsplash/Giphy attachments.
- Brand-color autofill: new CssColorFrequencyExtractor parses every
hex/rgb/hsl value in the homepage CSS, clusters perceptually similar
shades in CIE LAB (Delta E 76 < 12), filters neutrals and returns the
most frequent cluster. Solves Tailwind/utility-CSS sites where no
semantic --primary variable is exposed.
- Credits: gpt-image-2 metered at 15 credits/image (low quality default).
- Layout: AuthSplitLayout right column is sticky/h-svh so the form
textarea growth no longer stretches the marketing slider.
- i18n cleanup: localized labels follow the no-em-dash convention.
The event was dispatched by StripeEventListener but had no consumers
on either side (no Event::listen registration in PHP, no Echo listener
in JS, no subscriber on the users.{owner_id} channel). Removed along
with its tests; the listener no longer references it. Cleans up the
StripeEventListenerTest to match (drops Event::fake(SubscriptionCreated)
in five negative-path tests, replaces with Bus::assertNotDispatched(
TrackBilling) which is the actual cascade-side-effect we care about).
