Updated 2026-11-10: operators can set SHIPCONTENT_BILLING_BYPASS_USERS to a comma-separated list of owner user UUIDs on the app server and on the workers that open a billing session. A listed owner does not have to choose a plan. In enforce mode, that owner also bypasses plan limits, usage slots, schedule checks, a changed quote (PRICE_CHANGED) and an insufficient credit balance. The global billing mode, credit lots, ledger balances and other accounts stay unchanged. A bypassed credit reservation in enforce mode is stored as shadow, pins no lots, writes no spend and does not queue auto-refill. Its saved cost and the authorization result include enforcementBypassed: true. The reservation JSON reports mode enforce when that flag is set. A plan-limit, usage-slot or schedule result includes enforcementBypassed: true only when that check would have blocked the action. The balance sets unlimited to true and locked to false, the same unlimited flag as a credit exemption. With no paid plan, the plan line is Unlimited credits. A paid plan keeps its price line. The credit balance shows Unlimited and the credits used without charges. The page hides See plans, Change plan, Cancel plan, Buy credits and Auto-refill, and hides every notice carried on the balance. get_credit_balance still returns those saved notices. Credits used without charges (exemptSpend) still counts only committed credit-exemption reservations, so a bypass-only account shows 0. Studio and a No-limit spending budget do not check that balance. Unset, empty, or any item that is not a UUID disables the bypass for every account. The server compares lowercase UUIDs.
Updated 2026-11-04: each action shows its credit cost before you approve it. Each action rounds up to whole credits. The hosted service uses Stripe live mode and charges real money. Test mode is for a local server or a sandbox. Image quality, video resolution, reference videos and the Motion AI model each have their own price, so Studio, Quick AI and the editor show the credits that the server charges. Pro monthly and annual version 2 include 1,000 credits a month. Version 1 keeps 2,000 for current subscribers. A GPT Image request also pays for its input: each reference image of an edit and the prompt text are extra lines of the quote, and the credits are their total rounded up once. A Motion AI request has one fixed price, the most that one request can cost. Credit costs use the units request, output, second, character, image and byte. See make a Studio project and media.
Updated 2026-11-02: Free starts once for each email address, even after an account is deleted or the same person signs up again. If the email already used Free, the Plans page shows "Free was already used with this email." and offers Pro only; the start request returns FREE_ALREADY_USED. Password signup accepts only the plain form of an address: it rejects + tags for Gmail, iCloud, Outlook, Hotmail, Live, Fastmail, Me and Mac addresses, dots in Gmail addresses, and temporary-mail domains. Google sign-in uses the plain Gmail form and also rejects temporary-mail domains.
Updated 2026-10-23: every account that exists before billing starts gets one year of early access. It runs until one year after the owner signed up. Early access has the Pro limits and credit bundles, and no monthly credits. Billing shows Early access and the end date, and the paywall does not apply during that time. The account gets one credit grant when its email is verified. The grant has the same amount and twelve-month expiry as the Free grant, and it happens once. Early access is not a Free plan, and it is not in Checkout or the Portal. A paid Stripe plan replaces early access while it is active. After the end date the account has no plan. Accounts with an internal or other operator plan do not change. An early access account can subscribe to Pro and can buy credit bundles. Its first bundle purchase creates the Stripe customer first, through the same saved request as a Pro Checkout. If that request has an uncertain result, use Check plan request; no second customer is created. Updated 2026-10-28: a Free account buys credit bundles the same way. The one-time migration assigns early access only to accounts that existed before plan selection. New accounts must choose Free or Pro, including when billing enforcement is off, except an owner listed in SHIPCONTENT_BILLING_BYPASS_USERS (see the 2026-11-10 note).
Updated 2026-10-19: Billing reads the current billing status again by itself; the page has no Refresh button (see the 2026-10-24 note). Use Check purchase to recover the saved purchase; it does not start another payment. Workspace credit history still requires access to that workspace.
Updated 2026-10-23: a database works in one Stripe mode, test (sandbox) or live. It starts in test mode. An operator changes the mode once, before any Stripe data exists, with pnpm billing:ops stripe-mode. The key in STRIPE_SECRET_KEY must have the same mode. If the modes differ, Billing is not configured: it sends nothing to Stripe and saves nothing, and the webhook returns an error so Stripe sends the event again later. The Billing page shows Sandbox · No real charges in test mode only.
Open Account → Billing. Billing uses verified Pro monthly and annual prices. In test mode it uses a separate Stripe sandbox and no real money moves. In live mode it charges real money. The owner account pays for all its workspaces.
Updated 2026-10-24: Billing in the Account menu opens the Billing page of the current workspace. Outside a workspace, for example on /workspaces, it opens /billing, which goes to the Billing page of the last opened workspace. If that workspace has not finished setup, it goes to setup first. Use Billing to recover a saved payment request while plan selection is required.
Updated 2026-10-24: Billing is one page. The billing dialog and Manage billing are removed, and every billing action is on the page. The cards, from top to bottom, are:
- Notices: shown only when there is something to say. They carry the Stripe return message, Sandbox · No real charges, "Payments are not available right now." when Stripe is not configured, "Plan change needs payment" with Update payment method, credit warnings and the result of the last action.
- Plan: the plan name and one line: price and Renews or Cancels date for a paid plan, "Free until {date}" for early access, "Free plan · Ends {date}", or "Unlimited credits". See plans (Change plan with a paid plan) opens the picker with a Monthly and Yearly switch, the current plan and the offer. It shows Save {n}% when both prices exist. The button is Subscribe or Switch to {name}. Cancel plan opens the Stripe Portal for an active paid plan that still renews. A pending plan request or a scheduled change shows a row with Check plan request or Remove scheduled change.
- Credits: the available credits, split into Plan credits (with the reset date), Added credits (with the expiry date) and Other credits, and the credits held for running jobs. Buy credits opens the bundles and Buy. Check purchase and Auto-refill are here too.
- Usage: a meter for scheduled posts, X posts, connected channels, workspaces, storage and team seats, as "{used} of {limit}" or "{used} · No limit". A feature that the plan does not include shows "Not on your plan".
- Payment: shown when the account has a Stripe customer. Invoices and Update payment method open the Stripe Portal. The page has no card details.
- Credit history: one row for each credit event, with a plain label, the date and the signed amount. Show more reads older rows.
The page reads again after an action and when another part of the app reports a billing change. When the window gets focus, it reads the credits again, and the billing status only if the last read is more than a minute old. It has no Refresh button. An unlimited account sees Unlimited and the credits used without charges. It has no See plans, Buy credits or Auto-refill.
Updated 2026-10-24: a credit history row is one credit event. A spend that draws on several credit sources shows as one row with the summed amount. Each entry from get_credit_balance keeps its id, kind, amount and time, and now also has source and action. See get_credit_balance.
Updated 2026-10-27: the Billing screens are on for every signed-in owner. Every new account chooses Free or Pro before it uses the product, except an owner listed in SHIPCONTENT_BILLING_BYPASS_USERS (see the 2026-11-10 note). Existing accounts do not change.
Updated 2026-10-19: replaying an unchanged cancellation for a subscription that never had a paid invoice does not create a history conflict. It grants no credits or renewal grace. Unexpected later paid facts for that same terminal subscription are held for review. A stored cancellation reason and paid cancellation boundary remain immutable.
Updated 2026-10-24: a new account must choose a plan at its numbered workspace Plans page before new owner or agent work, except an owner listed in SHIPCONTENT_BILLING_BYPASS_USERS (see the 2026-11-10 note). Selecting Start Free starts its one seven-day Free period at that time; sign-in, verification and workspace creation do not start or restart it. The current Free catalogue gives 100 credits once after email verification; positive credits expire twelve calendar months after the grant. When Free ends, choose Pro monthly or annual. The page reads the verified Stripe price and monthly credit allowance from Billing and marks an unavailable choice. Existing accounts keep their saved Pro-level early access, paid or operator plans, end dates, credits and media.
Updated 2026-10-17: Billing shows current low-credit, expiry, plan-end and refill notices. Low-credit thresholds come from the account's saved plan version; zero is a separate notice. Purchased-credit expiry reminders group lots by UTC expiry date. An expired-credit notice uses the amount actually recorded as expired. Internal accounts are excluded. The owner-page banner refreshes while the page is visible. The Billing page reads its values and history again by itself; it has no Refresh button.
Email delivery is a separate operator gate. Leave SHIPCONTENT_BILLING_EMAIL_ENABLED=false until delivery is approved. To enable purchased-credit expiry emails alone, set SHIPCONTENT_PURCHASE_EXPIRY_EMAIL_ENABLED=true on the worker. The worker also needs SHIPCONTENT_AUTH_ISSUER, its email sender configuration, and the billing database connection. The broader billing email gate takes precedence and enables every customer billing email category. It does not enable billing incident email, which has its own gate. The worker can collect in-app notices without a Stripe key or an email sender. A provider receipt means the email was accepted, not that it reached the recipient. An uncertain sender is not replaced merely because its lease expired. Recovery retries keep their saved payload and key; unsafe or exhausted recovery requires operator attention. Updated 2026-10-30.
The expiry-day email waits until the purchased-credit lots for that UTC date are resolved. An unresolved reservation can delay the email. The expired amount comes from the ledger.
Updated 2026-10-17: eligible original Free accounts get a saved-content warning starting 23 days after their original Free end. The warning stays visible in Billing when credits are locked. A deletion date floor appears only after verified email delivery: it is no earlier than Free end plus 30 days or the verified delivery wait anchor plus seven days, whichever is later. This date is a minimum, not a scheduled deletion. Financial closure and storage checks are still required; cleanup remains disabled. Known paid history, including a zero-amount paid invoice or bundle, excludes the account. Internal access and suspension suppress warnings. An operator policy or suspension change cancels the old warning run; later expiry or restoration does not restart it automatically. Retention warnings are owner-only and are excluded from agent balance responses.
Updated 2026-10-17: /api/billing/email-events accepts signed Resend delivery events. Use the owner app host and configure SHIPCONTENT_BILLING_EMAIL_WEBHOOK_SECRET for this exact endpoint and retain the matching RESEND_API_KEY. Verification can continue while outbound billing email is disabled. Only email.delivered, matched to one saved provider email ID, sender namespace, recipient and subject, supplies delivery evidence. This means delivery to the recipient's mail server, not proof that the person read the message. Other events do not start a retention wait. A receipt that arrives late uses the later of its provider delivery time and first local observation; an exact duplicate keeps that time. A distinct delivery event can move the wait later, never earlier. A changed credential namespace or conflicting event requires reconciliation. This evidence path does not enable retention deletion. Cleanup remains paused pending its separate financial and storage checks.
- Select See plans, then Subscribe to open Stripe Checkout.
- Complete Checkout. In test mode, use Stripe test payment details.
- Stripe returns you to the Billing page and the page confirms the purchase. Credits appear after the worker verifies the payment.
- In Payment, select Invoices or Update payment method for invoice history and payment details. In Plan, select Cancel plan to cancel at period end. All three open the Stripe Portal.
Updated 2026-10-24: on the Plans page, Continue to payment opens Checkout with a saved request ID. A canceled or delayed Checkout keeps the account on Plans. Check plan request reads the saved payment status. If Checkout is still open, select Resume payment to reopen that same request. Updated 2026-10-28: to start again, for example in the other currency, select Cancel purchase. Billing asks Stripe to close the open Checkout, then reads it again. When Stripe shows it as expired, you can start a new Checkout immediately. A cancel never charges. If the payment completed first, the purchase stays and Pro access follows verification. Automatic return checks stay on Plans. Do not start a second Checkout while it is unresolved. Pro access begins only after the server confirms payment. A browser success URL is not proof. Billing, account settings, saved files, sign-out and payment recovery stay available while selection is required. Expiry does not delete saved media. Updated 2026-10-30: Cancel purchase never opens a Checkout. It only closes a Checkout that Stripe already has. A saved pack or Pro request that was never sent stays as it is. To pay for it, select Check purchase for a pack or Check plan request for Pro. That action continues the same request to payment. To drop it, leave it until it expires. The same action then closes it, and you can start a new purchase.
Updated 2026-10-24: Stripe Checkout and the Portal return to /billing/return. That page sends you back to the Billing page where you started, in the same browser tab, or to /workspaces. After a Checkout, Billing reads the purchase from Stripe once. A paid purchase shows a thank-you message when the worker has applied it. A Checkout that you left with Back shows that it was not completed and that you were not charged.
Updated 2026-10-16: Billing reads the current verified monthly and annual plan choices. Each choice shows its saved cash price and monthly credit allowance. A missing verified catalogue or Portal configuration is unavailable; the app does not invent a price. Sandbox automatic tax remains off.
Select See plans (Change plan when the account has a paid plan), choose Monthly or Yearly, then select Subscribe, or Switch to {name} when the account has an active subscription. The button for the price that the account pays now reads Current plan and is disabled. A plan Checkout has one subscription price and accepts promotion codes. The account can have only one nonterminal subscription. Billing keeps every Customer, Checkout, plan change, schedule step and Portal issuance request, including its exact parameters and first send authorization. An uncertain response stays pending. Check plan request reads the known object or causal request evidence; it does not submit a replacement payment. A completed listing with no match, an expired URL or an old request key does not prove that a prior request was never sent. Unresolved requests can require operator recovery.
Updated 2026-10-28: Billing and Plans show a USD and MXN switch when every offering has a recorded MXN amount. The switch selects USD first, or MXN when the browser time zone is in Mexico. American Express cards and Mexican cards pay in MXN. An MXN Checkout charges the fixed MXN amount of the same Price and grants the same credits. A subscription keeps the currency of its Checkout, so a plan change has no switch. Without recorded MXN amounts, Billing shows no switch and charges in USD.
Updated 2026-10-31: when Spanish is available for your account, Billing, Plans, the access notices and the credit count show your account language, English or Spanish. The language never changes the currency, the offered prices, the selected plan or card, or the amount that a payment sends. Each language marks the other dollar. In English, MXN shows as MX$. In Spanish, USD shows as USD and MXN shows as $.
Dates show in your time zone. A notice deadline keeps its exact time. A saved billing failure that the app knows shows in your language. Other billing failures show a general sentence in Spanish. When Spanish is available, a billing email uses your account language at the time it is first sent. A retry of that email always sends the same text. Stripe pages, receipts and invoices do not change with this setting.
A supported allowance upgrade uses a pending payment update. Declined or incomplete authentication keeps the previous paid allowance within its original period. Only a complete verified paid invoice history changes entitlement and credits. An interval change alone does not add another base grant or change the monthly credit anchor.
Updated 2026-10-18: a Stripe subscription canceled at the exact start of its paid term is recorded as ended, with both original timestamps unchanged. It provides no paid access interval. Repeated or older active snapshots cannot restore that canceled subscription. Existing credit and payment history remains intact.
Updated 2026-10-17: a retention payment fence stops new app requests and dispatch permissions for Customer creation, plan or bundle Checkout, plan changes, schedules, Portal sessions and automatic refill. Existing request status and late payment evidence remain available. A blocked refill does not undo completed AI usage. This local fence does not close an existing Stripe payment opportunity or prove that an earlier sender finished. Financial closure and safe fence release remain separate checks; the cleanup worker cannot activate this fence yet, and cleanup stays disabled.
Updated 2026-10-17: Stripe submissions use one transport attempt. The server records completion only for the exact original dispatch. A finished call can still have an unknown payment outcome that requires reconciliation. A pending native call interrupted by a timeout supplies no completion proof. Completion receipts may retry their database write, but they do not repeat the Stripe request or prevent a successful payment response from being reconciled. Cleanup still requires the full financial and storage closure checks.
Updated 2026-10-17: retention review inventories the saved app payment requests in bounded pages through the existing Stripe worker queue. Each request keeps separate evidence for its original dispatches. An unfinished or conflicting sender blocks review, even when a Stripe object ID is known. Late paid history, changed customer binding or changed configuration invalidates the review. Finishing this inventory does not prove financial closure: provider history, external payment sources and configuration still need verification. Starting a review remains private, and cleanup stays disabled.
Updated 2026-10-17: a provider read that shows paid history protects saved content immediately, even if a later graph page fails. This includes paid zero invoices, completed zero-payment Checkout, partial captured funds and later refunded payments. The server retains the first-class customer binding and the positive evidence; changing a customer binding cannot erase that history. These records stop retention only. Credits and paid-plan access still require normal financial reconciliation, and incomplete graphs remain unresolved.
Updated 2026-10-17: retention review reads one provider page or exact object per Stripe worker claim. It retains cursors, request IDs, object references and child links for current and former customers and their payment roots. Known native IDs are read even if a customer list omitted them. Invalid ownership, live objects, stalled pages or missing evidence stop progress. An object's terminal state does not close its children: reverse payment lists, customer-balance history and external configuration still need verification. Recovery-enabled Checkout, manual capture and unresolved external records require review. A paid discovery protects the account and queues normal financial reconciliation once, including earlier paid timestamps or recorded off-Stripe payment. Relationship pages retain allocations, reverse links and balance history. Newly discovered objects return to exact review. PaymentRecord scans use the merchant-wide API and first-class customer or invoice allocation links; email and metadata do not establish ownership. Observed native Customer results stay associated for preservation before they become current. Funded cash protects content, while a refund of uncaptured authorization alone does not prove captured payment. Updated 2026-10-17: a separate graph check verifies complete page chains, reverse allocations, dependent objects and known native child IDs. Late native evidence invalidates its result. Customer associations must use recorded owner requests; the former direct-binding action is retired in every billing mode. Updated 2026-10-17: accounts with no current or historical provider identity can receive a payment-closure certificate only after complete native history proves every creation request unsent. Accounts with any provider identity still need verified external controls and full closure evidence. Certificates are checked again against current policy and evidence before use. Tracked storage preparation waits for original writer outcomes and captures exact immutable object identities, including unmaterialized writes. Conflicting receipts remain blocked. Physical inspection checks bucket versioning, pending multipart uploads and bounded object pages against the sealed set. Multipart inspection uses bucket-wide pages with both continuation markers, then filters by workspace. Oversized pages and incomplete chains stop the scan; earlier prefix-only evidence cannot certify it. Missing, untracked or contradictory objects block it. A successful scan still requires runtime writer proof and deletion controls; cleanup remains disabled.
Updated 2026-10-18: Portal handles supported changes and cancellation at period end. When Portal cannot defer a same-Pro monthly-to-annual change, the app creates a two-phase schedule at the verified current item-period end. Portal compares the item amounts, not their annualized cost: a $20/month to $200/year change uses the app schedule even though the annual rate is cheaper. It preserves the current price, discounts and recurring settings. If Stripe describes a copied phase by coupon or promotion code, the app resolves it to the one matching discount already applied at that level. It reuses that discount instead of redeeming it again or resetting its duration. Missing or ambiguous matches require recovery. If an applied coupon is no longer redeemable, for example after its redemption limit is reached, the app refuses a period-end change before opening Portal or creating a schedule. Stripe can drop that discount while copying it. Exact removal of an existing native schedule remains available; removal does not restore a lost discount. A pending invoice, unrelated schedule or unsupported setting requires recovery. Remove scheduled change releases only the identified change and preserves independent cancellation. If the provider crosses the boundary during release, Billing reports the actual resulting price; it does not restore the old price automatically. After a verified app-owned target-phase transition, bounded worker recovery releases the schedule so Portal management can resume. Updated 2026-10-19: release verifies the exact configured phases, current item and target invoice lines. It does not depend on invoice payment or collection status. Missing or changed evidence stops release. This management action does not grant credits or paid plan limits. If the same target invoice later pays, the worker verifies its payment and grants its credits once. A release that crosses the period boundary must have exact provider event evidence. Missing or ambiguous evidence keeps recovery pending; it does not send another release. A change conclusively removed while the original price remained active cannot claim a later boundary payment. Replaying the later paid change keeps its own schedule identity.
The success page does not grant credits. A signed webhook first saves an event. The worker retrieves its current Stripe objects and applies each paid invoice once. An unpaid or failed invoice grants no credits. A failed renewal has a three-day access grace, measured from the last verified paid period end. Repeated events do not restart that grace. The existing access rules stay in effect until billing enforcement is enabled separately.
Updated 2026-10-19: a confirmed failed-renewal cancellation keeps the prior paid plan until that same grace deadline. This includes verified period-end schedule changes; the unpaid target plan grants no limits or credits. Voluntary cancellation, disputed payment, a missing cancellation reason and a failed first payment do not start this grace. A provider cancellation stays terminal; later events do not restart access.
Updated 2026-10-16: an active verified paid plan can buy a credit bundle when its plan enables bundles. Billing shows the applied bundle price, credits and expiry. Internal accounts do not get bundle purchase prompts. Updated 2026-10-28: a Free account and an early access account can also buy credit bundles. The first purchase creates the Stripe customer, through the same saved request as a Pro Checkout. An account with no plan cannot buy. If no bundle Price has been verified and mapped, Billing has no Buy credits button.
Select Buy credits, choose a bundle and select Buy. One payment Checkout accepts promotion codes in USD; sandbox tax stays off. The optional Save my card so I can turn on auto-refill checkbox records consent for future off-session use. It is off by default. Saving a method does not enable auto-refill, and a card saved with a subscription is not consent. Updated 2026-10-16: a completed zero-dollar promotional bundle can report paid or no_payment_required. Both require verified request, bundle mapping, discount and zero total, with no PaymentIntent. They grant credits without refundable cash. An explicit reconciliation maintenance item can revisit a source parked by the earlier zero-cash status restriction; it verifies the original purchase again and does not create another Checkout.
A pending or uncertain purchase stays visible in Credits, with its status and Check purchase. Use that action to recover its saved Checkout or check payment progress. An uncertain response does not start another payment request. The browser retains its purchase operation ID; reopening Billing checks that exact operation, even if another device completed a later purchase. An unrelated latest purchase never closes an unresolved operation. Reopening a pending purchase recovers the saved request. A completed operation resolves its original purchase without reopening Checkout. Select Buy credits and Buy again after completion to start a separate purchase. Updated 2026-10-28: an open purchase also shows Cancel purchase. It closes the open Checkout in Stripe and reads it again. When Stripe shows it as expired, the purchase closes and you can buy again immediately, for example in the other currency. If the payment completed first, the purchase stays and its credits follow verification. An expired request with any saved send authorization stays unresolved until Stripe recovery proves its outcome. A request that never had send authorization can close as unsent-expired; its history stays saved and the owner can start a new purchase. Billing keeps this local status readable if Stripe configuration is unavailable. Credits appear only after complete payment verification, never from a redirect or Checkout creation response.
Each purchase keeps its original offering and payment ownership. Its expiry starts at the first verified fulfillment decision, even when existing financial loss prevents an initial grant. Updated 2026-11-06: you can ask for a refund of a credit bundle within seven days of the purchase. A person reviews each request by hand, under the Terms of Service. The app has no refund button, and there is no owner refund endpoint. The operator refund tool has its own limit. It stops at the refund deadline saved with the purchase, or at the credit expiry if that comes first. The refund deadline is the paid account period end saved before the first Checkout request, with a 30-day fallback if no period applies. A delayed payment keeps that deadline. This limit can end before day 7. The operator then refunds in the Stripe Dashboard, and the worker reconciles that refund like any other. Later issuance and restored credits keep the original expiry. A Checkout and its PaymentIntent produce one purchase source. Unknown direct PaymentIntents cannot create purchases from metadata.
Mismatched purchase requests or line items park as PURCHASE_SOURCE_MISMATCH; unbound requests park as PURCHASE_SOURCE_UNBOUND; conflicting payments park as PURCHASE_PAYMENT_MISMATCH. A changed pending selection returns CHECKOUT_INPUT_CHANGED. A request that cannot safely submit again returns CHECKOUT_RECOVERY_REQUIRED. Current account restrictions stop new Checkout submissions while verified accepted payments still reconcile.
Updated 2026-10-16: reconciliation supports complete paid renewal history and paid plan changes with a verified predecessor. Requested changes with an unpaid, declined, action-required or expired payment keep the prior paid limits. A pending-update event alone is not proof of payment. The portal does not offer plan changes yet; provider acceptance of that flow is still required.
Within one monthly credit period, paid increases use the original allowance plus the exact sum of prorated increases. The calculation rounds that sum once and subtracts credits already granted for the period. Spent or expired grants still count. Equal-allowance paid changes can update limits without new credits. Upgrade lots keep the original expiry. Monthly/annual changes in the same subscription retain its monthly credit anchor; a change of payment interval alone gives no second base grant. Verified annual payments store the actual monthly periods from that retained UTC anchor. Each boundary is calculated from the original day and time, including month-end clamps.
Verified downgrades and cancellation use their period boundary. Late paid invoices keep their original credit dates; expired periods are recorded without new credits. An older canceled subscription cannot remove a newer paid source. A deleted bound customer creates a durable review block; it does not replace the customer or remove internal operator policy. Grace and plan-end notices are saved once for later delivery; this change sends no email.
Updated 2026-10-17: internal access and unlimited credits are separate operator choices. An unlimited-credit account records authorized usage without reserving or deducting credit lots. Billing shows Unlimited and the all-time credits used without charges. Removing unlimited authority affects new work; work already authorized keeps its recorded exemption. Reassigning internal access does not reactivate an earlier unlimited-credit choice. Finite internal accounts still use their granted credits. The billing enforcement bypass is not this exemption. See the 2026-11-10 note.
An active internal plan keeps its operator policy when a payment succeeds, fails or is canceled. Payments still settle and appear in the ledger. Internal accounts cannot start a new subscription Checkout. After internal access ends, the account uses its valid paid term or grace, then its original unexpired Free term, then no plan. This does not start another Free term. An explicit no-plan policy or account suspension keeps its force.
Updated 2026-10-16: payment verification reads every required page of invoice lines, invoice payments, historical charges, refunds, disputes and Checkout lines. It retains each payment's amount allocated to each invoice. It verifies the bound sandbox account, customer, currency and immutable Price mapping, then reads the governing objects and lists again. No partial graph grants credits.
A verified zero-dollar promotional monthly invoice can grant without a PaymentIntent. Payment records, manual payments and customer credit-balance adjustments are saved for review as PAYMENT_EVIDENCE_UNSUPPORTED. Missing or ambiguous change history parks as STRIPE_HISTORY_AMBIGUOUS; changed immutable paid history parks as STRIPE_HISTORY_CHANGED. Conflicting subscriptions park as STRIPE_SUBSCRIPTION_CONFLICT. Manual bundles use verified purchase sources. Auto-refill remains separate work. Changed allocations after a grant park as STRIPE_ALLOCATION_CHANGED, including when a payment no longer includes its previously funded invoice. They do not duplicate credits. Missing or contradictory refund identities park as STRIPE_REFUND_EVIDENCE_INCOMPLETE. Succeeded refunds adjust only their fixed payment-owned credits. Pending, failed and canceled refunds do not assess credit loss. The database also checks object ownership, matching successful cash charges and total paid allocations before a grant.
The worker claims one shared lease before it reads Stripe. Each claim handles one graph and lasts at most 60 seconds, including renewals. The graph has a limit of 128 requests, 4,000 returned rows and 20 pages per list, including verification reads. Each request has at most 10 seconds and must finish before the current lease ends. Large graphs park as STRIPE_GRAPH_BUDGET. Expired or changed snapshots are discarded and retried from fresh reads. Event and maintenance work alternate when both are due. Annual refills use this same queue and final lease check. Each claim issues or skips one account credit period after a fresh provider read. An unchanged provider graph can have a new due period without creating another provider revision. Auto-refill remains separate lifecycle work.
Updated 2026-10-16: operator purchase refunds now use a read-only cash preview and a saved refund request. Only the database operator can prepare a request. Owners, agents and runtime workers cannot choose arbitrary refunds. There is no owner refund endpoint. All captured cash must belong to eligible purchases within their original refund and expiry dates. Mixed plan cash, incomplete ownership, unresolved disputes and unknown cash amounts stop approval. Shared credits use only the amount that is provably unused for that payment. Failed, uncaptured charge attempts remain in the payment history and do not block its single successful captured charge. Multiple successful captured charges for one PaymentIntent, or unexplained positive captured cash, are inconsistent evidence and stop submission.
Preparation saves the exact cents, operation key, actor, reason and eligibility evidence. It reserves the cash and freezes each supporting lot, including a lot whose rounded credit loss is zero. Existing reserved work can settle. Before each possible Stripe POST, the worker reads and reconciles complete current evidence, checks authority and the cash limit again, and commits a single-use send permit. It keeps one provider key, disables SDK retries and stops sends after the original windows or the saved 23-hour cutoff, which starts at the first durable submission authorization. Queue time before that authorization does not consume the retry window. The cutoff never resets after a possible send.
Updated 2026-10-18: a refund with no submission authorization closes as canceled if its authority is revoked, its original refund or expiry deadline passes, or verified cash or dispute evidence makes it ineligible. This releases only that request's cash hold and freeze. Other loss causes remain. Local closure stops repeated refund reads and does not require Stripe to be available. Bounded checks rotate across pending requests, so parked provider work cannot starve later expired or revoked requests. Once any submission was authorized, the hold remains until provider readback resolves it.
Timeouts and uncertain database commits keep the request, reserved cash and freezes. REFUND_REFRESH_REQUIRED requires a new complete provider read; REFUND_PENDING keeps the existing operation; REFUND_CASH_LIMIT refuses cents above the exact cumulative limit. Verified provider identity and matching financial fields resolve the request. Provider failure or reversal clears only that request's freeze after credit reconciliation. A late success still reconciles, and verified excess creates an operator review flag. An independent Stripe Dashboard action can race a local send; the local permit cannot prevent that remote action. The operator CLI and automatic refill remain separate integrations.
Server setup#
Updated 2026-10-16: annual schedules keep the original payment basis and expiry for every owned component. Ordinary new annual terms have twelve monthly bases. Retained anchors can have eleven or thirteen due bases and up to fourteen intersections. A funded current period gets no duplicate base. An unfunded current period gets its exact partial allowance. A paid annual-to-monthly change supersedes only boundaries after the credit period containing payment activation; already-due ownership stays fixed. Superseded components cannot issue after a later eligibility release.
Refills require an active subscription and current verified paid authority. Inactive work records a deferred decision and retries. Operator policy remains independent of paid accounting. Expired catch-up records a skip and continues without extending expiry. Zero-owned or fully withheld components create decisions without zero-amount lots. Supported terms are positive and at most 31 days for monthly prices or 366 days for annual prices; longer terms park for review. Provider/test-clock acceptance remains a separate release gate.
Keep the Stripe key in server secrets as STRIPE_SECRET_KEY; use a restricted key where possible: rk_test_ in the sandbox and rk_live_ in production. A secret key (sk_) also works. Plan Prices and the Portal configuration come from the verified mappings and Portal snapshot in the database, not from environment values. Updated 2026-10-30. Set STRIPE_WEBHOOK_SECRET to the signing secret for this exact endpoint or local Stripe CLI listener.
Configure the worker with SHIPCONTENT_BILLING_DATABASE_URL, using the separate shipcontent_billing_runtime login. Bootstrap needs SHIPCONTENT_BOOTSTRAP_BILLING_PASSWORD. Apply migrations without resetting saved data. Apply the existing billing catalogue. Prepare an explicit JSON array with priceId, kind (plan or bundle), key, version, interval (month, year, or null for a bundle), productId, currency (usd), and decimal-string unitAmount. Then run pnpm billing:stripe:sandbox --account acct_YOUR_SANDBOX --mapping /absolute/path/prices.json for a dry run. Add --apply to save the mappings and sandbox account binding. The command only reads Stripe; it does not create products, Prices or portal settings. All plan Prices must use one product.
Updated 2026-10-23: a billing fault cannot stop the worker. If a billing service cannot start, the worker logs its name and error code and runs without it. A STRIPE_SECRET_KEY that is neither a test key nor a live key logs one warning and keeps Stripe work off. A key whose mode is not the mode of the database keeps Stripe work off too: each step stops with BILLING_NOT_CONFIGURED before it reads or writes anything. Retention work runs on the billing database login alone, in its own loop. Retention steps that read Stripe wait until a key with the mode of the database is set. A Stripe outage does not stop the other retention steps.
Updated 2026-10-16: the operator can preview remote catalogue changes with pnpm billing:catalog --remote sandbox --input /absolute/path/catalogue.json. Updated 2026-10-23: --remote live does the same with a live key. The key, the --remote value and the mode of the database must be the same, or the command stops with CATALOG_ENVIRONMENT_MISMATCH. This preview makes no remote or durable database writes. Explicit apply requires --apply, an operator name, a reason and a stable operation UUID. The input supplies exact local offering versions, USD amounts, product IDs or a new product name, and an existing Portal configuration or an explicit template for a new copy. No annual cash amount is assumed.
Updated 2026-10-28: pnpm billing:currency --price price_123=20000 adds a fixed MXN amount, in centavos, to a mapped USD Price. Repeat --price for each Price. Without --apply the command only reads Stripe and changes nothing. With --apply it adds the amount to the Price in Stripe, reads the Price again and records the amount. The Stripe key and the database must be in the same mode. A recorded amount does not change: a new amount needs a new Price.
Apply preserves immutable price mappings and saves a verified configuration snapshot. Before authorization it saves the adopted product and Portal source snapshots; later edits require review instead of becoming a new baseline. Completed readback rechecks current and historical plan tax compatibility without a provider write. It checks invoice history, cancellation at period end, current monthly and annual destinations, both supported scheduling conditions and existing payment-method settings. Resume uses the same operation and input. A lost provider response or uncertain commit requires readback; a new request key is not a recovery method. Unresolved operations block competing catalogue writes. This operator path does not grant owners or agents financial authority. Native Portal behavior remains a separate acceptance check.
Updated 2026-10-18: native Portal transitions use the complete, repeatedly read customer schedule history. A released schedule must have reached its target phase and match the prior paid term, new invoice, customer, subscription, price, item and billing period. An earlier undone schedule cannot authorize the transition. Existing credit anchors and current-time issuance rules remain in effect; a test clock reaching a future paid term does not make future credits spendable now.
Mappings retain old Price IDs when lookup keys move. A saved mapping cannot be reassigned. Existing mappings need their product and amount verified before reconciliation can use them. A dry run rolls back database changes. If an apply reports commit-unknown, inspect the database before retrying.
Send events to POST /api/stripe/webhook. The endpoint accepts events of its mode only (livemode false in test mode, true in live mode): refund.created, refund.updated, refund.failed, charge.refund.updated, charge.refunded, charge.dispute.created, .updated, .closed, .funds_withdrawn, .funds_reinstated, checkout.session.completed, checkout.session.async_payment_succeeded, checkout.session.async_payment_failed, payment_intent.succeeded, payment_intent.payment_failed, invoice.paid, invoice.payment_failed, invoice.payment_action_required and customer.subscription.created, .updated, .deleted, .paused, .resumed, .pending_update_applied, .pending_update_expired, and customer.deleted. Use Stripe CLI forwarding for local development. Requests need the original JSON body and Stripe-Signature header. Invalid signatures and events of the other mode are rejected; a database failure returns an error so Stripe can retry. Parked events retry with backoff; inspect dead events before retrying them manually.
Tax#
Stripe Tax is off. Tax collection needs a confirmed business address, product tax category, applicable registrations and a decision about tax-inclusive or tax-exclusive prices. A Stripe invoice is not a Mexican CFDI factura.
Updated 2026-10-16: agent balance reads can include the account plan, balances, expiry and limits. Transaction history, adjustment reasons and pagination are owner-only. Agent responses retain the common shape with ledger: [] and nextCursor: null, including when a cursor is supplied. A revoked agent grant cannot read the account balance.
Updated 2026-10-16: the credit ledger tracks each loss cause separately. Released credits first pay for confirmed work, then repay pending loss. A decrease cancels only that cause's pending debt before restoring its recovered credits. Restoration keeps the original lot and expiry; expired credits are restored and expired in the same transaction. Clearing one freeze or flag does not clear another cause. Owner history can show positive restore entries. Agent reads still omit transaction history. Verified refund reconciliation uses this path. Verified dispute reconciliation also uses this path. The operator CLI and actual provider acceptance remain separate work.
Updated 2026-10-16: each payment keeps its original captured cash principal and fixed share of every credit component. Uncaptured authorization is excluded from the principal; captured cash that was not allocated to an invoice remains included. Refund loss uses succeeded principal only, with one exact cumulative calculation. Payment shares and original lot contributions do not change after spending or refunds. A reversal restores recovered credits in the original lot and releases only never-issued eligible credits. It cannot cancel another payment's debt. Future, expired and superseded entitlement does not create debt or change plan limits.
Updated 2026-10-18: if an earlier Stripe read already issued an annual credit period before a later annual-to-monthly payment is applied, that issued allocation stays in its original lot. The monthly successor deducts the existing allocation. Later unissued annual periods are superseded. Replaying the change adds no credits.
Annual refund reversal creates durable period work for newly eligible credits. Each claim still handles one period, with the original expiry and current paid authority. The worker can drain several due periods without another webhook. Identical accepted snapshots do not repeat financial revisions when their payment, refund and dispute evidence also matches the current shared state. A fresh reversal through another event root creates a new decision, even when that root saw the same facts earlier. Pending and failed statuses remain distinct evidence even when both assess zero credit loss.
Operators can enqueue a fresh provider read for old parked refund graphs with shipcontent.enqueue_stripe_refund_adoption() using the billing role. This also enrolls existing subscription bases. STRIPE_REFUND_POLICY_ADOPTION_REQUIRED means an unchanged old snapshot needs that maintenance path. The first accepted adoption records one new revision; later identical work adds no financial revision. No stored provider timestamp is changed. Sandbox/provider acceptance remains a separate release check.
Updated 2026-10-16: open disputes freeze the lots funded by that payment. A verified lost dispute keeps its freeze and adds principal loss; fees do not add credit loss. A verified won or warning_closed result removes only that dispute's cause. Funds movement events cause a fresh read and do not prove the result. The reader includes closed dispute history even when the charge no longer has its disputed flag. Missing or changed identities park as STRIPE_DISPUTE_EVIDENCE_INCOMPLETE. An unsupported status, including prevented, parks as STRIPE_DISPUTE_STATUS_UNSUPPORTED and preserves the last accepted financial state.
Refund and dispute loss share one ceiling: the original payment-owned credits. Lost disputes receive priority in order of original creation time, then dispute ID. Each receives the difference between successive capped credit prefixes, including a recorded zero share. A change of cause transfers existing recovery attribution without returning and debiting the same credits. Only a net loss change moves credits. Open or lost causes also freeze newly issued annual credits from that payment; clearing one cause preserves other disputes, pending refund requests and legacy freezes. Principal outside the captured payment bounds is retained and flagged for review without increasing the credit ceiling.
The billing role can call shipcontent.enqueue_stripe_dispute_adoption(limit) to queue fresh reads of pre-policy parked dispute graphs and saved charge.dispute.* roots that failed as STRIPE_WORK_UNSUPPORTED, including exhausted dead events. Other unsupported event types are excluded, and the original event history is retained. The limit is 1–100, with a default of 100. Each call selects only roots not already queued; repeat calls can process a larger backlog. Accepted replay stays inert. This local implementation still needs separate sandbox acceptance. It does not submit refunds, contest disputes or enable purchases or auto-refill.
Updated 2026-10-16: Auto-refill is a separate owner opt-in for an eligible plan (paid, Free or early access) and a previously consented saved card or Link payment method. Updated 2026-10-24: it is at the bottom of the Credits card. Select Set up auto-refill. Choose a bundle, enter a credit threshold in Refill when credits drop below and a Monthly limit (USD) in dollars, then explicitly authorize automatic charges and select Turn on auto-refill. Enter the limit as dollars, for example 10 or 12.50. The page saves whole cents and refuses any other value. Without a saved payment method, the card says to buy credits and choose Save my card. When refill is on, the card shows Turn off auto-refill. An unlimited account has no Auto-refill. A fresh successful owner-paid action below the threshold can queue one refill. Agent spending budget charges never queue a refill. Viewing Billing, failed credit reservations, releases and old off-mode actions do not trigger a charge. Billing off mode never authorizes an automatic payment. Updated 2026-10-28: a Free account can enable it, like an early access account, after a bundle purchase that saved its card. Internal accounts cannot enable it. Each possible payment submission checks current eligibility, opt-in, saved-method ownership and type, and the monthly cap again. Updated 2026-10-28: a refill charges the saved card in the currency of the Checkout that saved it. A card saved in an MXN Checkout pays the MXN amount of the bundle, and only a bundle with an MXN amount can be chosen. The monthly limit stays in USD and counts the USD amount of each bundle. Updated 2026-10-28: changing settings while refill is on keeps its current saved card and charge currency. To use the newest saved card, turn refill off and set it up again.
The monthly limit uses UTC calendar months and gross charges. A refund does not restore this budget. An unresolved payment retains its reservation and blocks another refill even after a month change. Turn off stops new payment authority; an already accepted payment can still complete and grant its verified credits. A declined payment or required authentication turns refill off and creates one failure notice intent. The worker closes the same payment where Stripe permits cancellation. Until complete readback proves no successful cash remains, Billing keeps payment review pending and blocks re-enabling. No automatic authentication or payment retry follows a decline. Unknown outcomes recover the same payment; a conservative 23-hour limit starts at the first durable submission authorization. The worker uses readback after any uncertain submission; missing or ambiguous results retain the reservation and do not cause another automatic POST. After that limit, only provider readback can resolve the attempt. Actual sandbox/browser acceptance and notification delivery are separate release checks.
Updated 2026-10-16: Billing keeps the saved bundle, credit threshold and monthly limit visible while refill is enabled or a payment remains unresolved. These are the saved server terms, including historical bundle prices that are no longer offered. A temporary payment-method read failure parks the attempt with no charge and keeps opt-in unchanged. Only verified missing, detached or incompatible method evidence disables it. Both verified saved card and link methods retain their original type; the worker never substitutes another method. Provider authentication and funding-source restrictions can still reject a payment, which follows the existing no-retry and closure rules.
Updated 2026-10-16: catalogue checks and owner Portal issuance request expanded product selections and reject incomplete responses. New configurations explicitly disable adjustable quantity. For a known, unpublished app-created clone with the wrong quantity default, an operator can save a read-only --preview-portal-quantity selection and explicitly apply --correct-portal-quantity --selection /absolute/path/selection.json. The separate corrective child preserves the original operation and receipts, changes only quantity, and needs its own authorization. Ordinary resume never starts this correction; it publishes only after verified correction and retains the original input and reason. Agents cannot perform these actions.
Updated 2026-10-16. Billing enforcement also checks X publication places and generic automatic schedules. An X approval reserves one place; final submission checks the current usage period. An unknown provider outcome keeps its submitted place until reconciliation. A known unsent failure or cancellation releases it. Scheduled places are shared across owned workspaces. Existing accepted schedules remain after a scheduled-cap decrease; publishing permission is checked again at dispatch.
Updated 2026-10-16: In enforce mode, new workspaces and connected channels share one owner account limit across active owner workspaces. Storage counts saved bytes plus live upload reservations in the selected workspace. Queue usage includes jobs awaiting approval and all other nonterminal jobs. Scheduled usage counts undispatched generic schedules and explicit uninitiated TikTok schedules once across the account. Owner-only account reads leave workspace storage and queue usage unknown. X period usage combines committed slots and qualifying holds. Enforce mode excludes unsubmitted shadow-only holds; submitted shadow work still counts. Shadow and off show all saved nonreleased slots in the current period. Its reset is the current plan period end. A null capacity means unbounded capacity for that limit. It does not mean unlimited credits.
Read balances and current prices#
Updated 2026-10-24. Open the numbered Billing page, such as /3/billing, or Billing in the Account menu for account recovery. Account reads work without Stripe configuration and after workspace access ends: if the read with the workspace fails, the page reads the account without it. The page shows the available credits, the credits held for running jobs, where the credits come from, the next expiry, the plan limits in Usage and the owner Credit history. Account-wide counts and scheduled posts cover all active owned workspaces. X usage is one aggregate in its current period, with the period end as its reset. A meter whose usage cannot be read, for example workspace storage when no workspace scope is available, is left out. The page does not show Unknown.
Call get_credit_balance with a workspace UUID through a current grant. The agent receives plan, balance, expiry and usage, with no owner ledger or cursor. Call get_credit_prices for current action units, versions, block costs and validUntil. Missing prices are unavailable. Both reads remain free with no plan or at zero credits. Without a plan, saved credits can remain locked for spending. Drafts and saved files stay readable.
Updated 2026-11-07: a price estimate grants no payment authority. Eligible new media jobs can use the calling grant's spending budget at current server prices. Otherwise the owner must approve the current quote, including after PRICE_CHANGED. Refresh prices after that error; never automatically retry a paid submission. Keep the exact request and accepted quote when a response is uncertain. Only owner cookie sessions can create payment sessions; publishing permission does not permit payments or owner paid approval.
Credit amounts are decimal strings. The Plan credits and Added credits amounts on the page count the credits that remain in each source. They can differ from spendable planBalance and purchasedBalance after reservations, expiry or freezes. Never infer unlimited credits from a plan name or an unbounded storage cap. get_credit_balance.unlimited is true for an explicit credit exemption or for an owner on the billing enforcement bypass. See the 2026-11-10 note. Durable low-credit notices remain a separate rollout gate.
Updated 2026-11-07: a spending budget belongs to one connected agent grant across workspaces. It is a total from the last time the spending budget was turned on. Changing a positive budget keeps its start time and usage. Turning it off and on starts a new period. Credit amounts are decimal strings. get_credit_balance.spendingBudget exposes only the calling grant's enabled, budgetCredits, usedCredits, leftCredits and since. set_spending_budget accepts "0", "1" through "1000000" or "unlimited". With No limit, budgetCredits is "unlimited" and leftCredits is null; changing between an amount and No limit keeps usage. Held and unknown reservations count at their reserved amount; committed reservations count at their final amount. Released reservations count as zero. Every budget setting has an audit record, even an unchanged amount. An owner whose access ended can still turn a budget off.
For propose_generation, propose_render, export_project and prepare_seedance_reference, all new jobs in a batch must fit together or wait for approval without new holds. Spending budget reservations identify the agent grant as their actor. Owner approval uses the owner actor and does not consume the agent budget. creditsCharged is the sum of new holds reserved by the call, not its final settlement; a replay or approval fallback returns "0". budgetLeft is the remaining allowance, or null when the budget is off or set to No limit. Read media decisions for requiresOwner and reason.
The spending budget never triggers auto-refill. Ordinary enforce accounts use only available account credits. Shadow mode and internal exempt accounts keep their existing authorization policies; their holds still count against the agent budget. With No limit the budget check is skipped. The ten-job workspace limit and replay rules still apply. Ordinary enforce accounts still use available credits (NO_CREDITS falls back to approval). Where the balance is not checked (a credit exemption, a billing enforcement bypass, or shadow mode), No limit has no credit ceiling. Billing mode off falls back to approval with BILLING_OFF, because it creates no holds. askOwner: true requests approval for new jobs; replay always preserves saved jobs. An expired unstarted spending budget hold requires owner approval, including in shadow mode. The worker cannot reserve those credits again. Never automatically retry paid submission or generate again to fix a failed download.
Turning the spending budget off keeps queued jobs. Disconnecting cancels eligible unsent agent jobs across workspaces and releases their holds. Generation submission or native render execution that already started cannot be stopped this way. Owner-approved jobs keep their state. Reconnecting creates a new grant with the spending budget off. The owner sees the agent name in job details and credit history, and sees budget changes in Agents or /agent-access. See spending budget before enabling it.
Billing analytics#
Updated 2026-10-17. Billing uses checkout_started, checkout_completed, credits_low, credits_depleted and plan_limit_reached. Checkout events cover explicit owner plan Checkout and manual bundle Checkout, including a verified zero-cash Checkout with a positive original credit grant. Upgrades, renewals and automatic refill are not Checkout events. Started follows durable send authorization; completed follows the exact original Checkout's canonical fulfillment and grant evidence. Later spending, refunds or restoration do not rewrite completion.
A private delivery queue reads committed payment evidence and uses one stable delivery identity per event and owner request. Delivery retries do not run payments. No Stripe configuration is needed to read the owner balance. If analytics configuration is absent, no analytics delivery is claimed. Low/depleted producers await the durable notice rollout. Analytics properties exclude Stripe identities, payment URLs, email, tokens, balances and raw provider errors.
Delivery keeps the queue's first observation timestamp across retries. Backfilled events use the time the collector first observed their evidence, not an inferred historical payment time. A collected Checkout start precedes its completion.
Updated 2026-10-18: new retention storage preparation refuses unresolved original senders before installing a storage fence. An operator can cancel a preparation run with a saved operation ID. For an existing fenced run with only unresolved PUTs, cancellation releases the payment fence while storage drain remains pending. Unknown or conflicting DELETE outcomes keep both fences. Unresolved byte reservations and storage fences remain until original sender outcomes are proved. Completion permanently closes the old run; replay cannot clear a newer fence. Starting another review requires explicit reevaluation, current eligibility and a new warning with its own delivery wait. App and agent credentials cannot perform these operations. Cleanup remains disabled.
Updated 2026-10-17: operator internal-access requests remain pending until earlier retention deletion outcomes are conclusive. Pending does not grant internal access or unlimited credits. The saved operation supports status, retry and withdrawal; its absolute expiry cannot be extended by replay. Activation reports prior successful DELETE receipts without claiming restoration. Suspension and paid evidence keep their existing authority. Cleanup remains disabled pending runtime and deletion-transport acceptance.
Updated 2026-10-17: each retention DELETE uses one fresh durable authorization and the exact captured key. The worker retries only receipt writes. A lost response stays unresolved and cannot cause a replacement storage request. Object receipts do not complete creative-content cleanup. Runtime and provider acceptance remain required; cleanup stays disabled.
Updated 2026-10-17: retention content review now compares an original Free-end cancellation with the full post state captured when it was scheduled. Later edits, new creative work, fresh approvals and later schedules preserve the post. Restoring old text does not restore its earlier eligibility. These checks do not enable content deletion; cleanup remains disabled.
Updated 2026-10-17: queued X, YouTube and Instagram schedules stopped at original Free end can qualify for content review only when saved history proves that no publication step was authorized. A cleared current intent does not prove this. Older jobs without that history remain protected. Instagram checks the plan before its first staging step; an existing step keeps its original recovery rules. Cleanup remains disabled.
Updated 2026-10-17: the private content-removal transaction clears creative text and copied publication payloads while retaining reference IDs and financial evidence. Permanent removal records prevent stale writers from restoring the content. Runtime checks, worker integration and complete UI acceptance are still required; cleanup remains disabled.
Updated 2026-10-17: after content removal, Instagram approval and readiness reads have no current payload. Minimal request history remains available. TikTok reviews report that the post content is unavailable and keep the canceled operation. Old request IDs cannot recreate removed work; erased approval or readiness inputs can return IDEMPOTENCY_CONFLICT. Create a separate draft for new work. Cleanup remains disabled.
Updated 2026-10-17: content cleanup processes one captured post per maintenance claim and records whether it was removed or preserved. Later schedules cannot expand the captured scope. The content phase requires complete post outcomes and object-deletion receipts before it can finish. Runtime/provider acceptance is still required, and cleanup remains disabled.
Updated 2026-10-17: owner and agent library reads omit files with an original successful retention DELETE receipt. Owner website-file reads use the same rule. Reference and financial history remain. An unknown outcome does not prove removal. Restoring access does not restore removed bytes or make that file available again.
Updated 2026-10-17: confirmed file removal queues quota accounting for the exact original asset charge. A durable receipt prevents a second deduction on retry. Unknown or conflicting removal evidence does not release quota. Stored reference rows remain; restoring access does not restore removed bytes.
Updated 2026-10-17: Media and Motion reject new work that uses a file whose retention deletion is confirmed or unresolved. STORAGE_KEY_RETIRED requires another source file. The refused attempt creates no new job or credit hold. Existing submitted requests keep their original result and billing reconciliation path.
Updated 2026-10-17: new schedules and channel publication work reject files whose retention deletion is confirmed or unresolved. Upload and staging steps check again before starting. Existing provider receipts and completion of already-uploaded remote media keep their original recovery path.
Updated 2026-10-17: website file reads and saved-file reuse reject retired storage keys. Analysis checks its selected current file inputs before submission. Archived captures do not block new work, and URL refresh can capture fresh data. A refused new analysis saves no new operation or credit reservation.
Updated 2026-10-17: retention completion now requires a saved check for each captured file, including files removed in an earlier run. The existing worker checks one file per claim before post-content work. It reconciles the original asset charge once and records retained file references. A later run uses the original deletion receipt; it does not send another DELETE for the same removed object. File checks do not release fences or establish runtime safety. Cleanup remains disabled.
Updated 2026-10-17: saved owner and agent upload reads withhold unavailable files from their asset field. Media jobs omit unavailable output IDs. Original completed statuses and request identities remain as history. Replaying one of these operations cannot restore its bytes or start another upload or paid generation.
Updated 2026-10-17: confirmed retention accounting also removes the exact original reservation for files that never became saved assets. Other reservations remain. An unresolved or conflicting deletion does not release reserved bytes. Quota release does not mark a PUT as rolled back or reopen its upload. A retired MP4 receiver keeps canUpload: false after access returns.
Updated 2026-10-17: the worker records daily local accounting checks in bounded account steps without changing balances or releasing unknown holds. Unknown age comes from the first recorded state transition; older records with no reliable timestamp remain age unknown. Local scan completion does not prove Stripe agreement. Provider checks and incident delivery remain pending.
Updated 2026-10-17: daily local checks compare recorded plan and purchase decisions with original credit grants. They distinguish positive credits from zero cash and preserve valid withheld, expired and future-period outcomes. Older purchase evidence stays labeled as derived. A discrepancy creates a private operator finding; the check does not issue replacement credits or prove Stripe agreement.
Updated 2026-10-17: daily checks queue known Stripe objects and due plan periods for fresh provider readback through the existing billing worker. Failed reads remain pending under the same work ID. Successful readback records a receipt; it does not prove that the account has no other incidents. Missing provider configuration or an unknown object identity cannot be reported as verified.
Updated 2026-10-17: reconciliation reports unresolved payment and retention operations under their original identities. Retrying failed work does not clear its incident. Events without a trusted account binding stay in a separate private operator queue. A later account binding can resolve that attribution issue while the payment issue remains open. These checks do not retry payments, release holds or enable cleanup.
Updated 2026-10-17: authorized operators can use pnpm billing:ops to preview a saved operation file and apply it with --apply. Each change has a stable operation UUID, actor, reason and atomic audit. The CLI requires the dedicated shipcontent_billing_operator role through the server-only operator database connection. App and agent credentials cannot use it. Reuse the same file after an uncertain COMMIT and read operation status before another action. Reports are read-only; preparing a refund does not mean Stripe completed it.
Updated 2026-10-17: operator incident alerts use a separate configured email or Buzz destination. Missing configuration leaves them pending. An accepted send is separate from incident resolution and does not prove recipient delivery. Uncertain sends keep their original event and delivery key; lease expiry does not authorize another send. Operator deliveries reports show pending, accepted and unknown outcomes. App and agent credentials cannot configure destinations or replay alerts.
Updated 2026-10-27: server operators select SHIPCONTENT_BILLING_INCIDENT_DESTINATION as email or buzz. Email needs SHIPCONTENT_BILLING_INCIDENT_EMAIL_TO, SHIPCONTENT_BILLING_INCIDENT_EMAIL_ENABLED=true and a worker Resend key. This gate does not enable customer billing emails, and the customer gate does not enable it: when it is unset or empty, incident email stays off (updated 2026-10-30). Buzz needs SHIPCONTENT_BILLING_INCIDENT_BUZZ_ENABLED, SHIPCONTENT_BILLING_INCIDENT_BUZZ_URL and the private SHIPCONTENT_BILLING_INCIDENT_BUZZ_SECRET. Empty destination configuration sends nothing. The billing workflow is separate from Feedback. Never include these credentials in an agent request.
Updated 2026-10-17: the low- or zero-credit recovery link on a numbered Billing page keeps the selected workspace. The page reads again after an action, on window focus and on a billing change elsewhere in the app; it updates the credit values, account banner and payment actions together. Account recovery remains available when no paid plan is active. Workspace usage that cannot be verified is not shown.
Updated 2026-10-17: an unfinished owner Library upload keeps its request ID in browser storage, bound to the owner, workspace, file manifest and file contents. After an uncertain response, reload and select the same file to recover that operation. A confirmed completed upload reuses its saved asset without another transfer. Browser storage failure stops submission. A completed file removed by retention reports that result; a later explicit file selection can start a new upload.
Updated 2026-10-17: Studio billing errors open Billing for the selected workspace. A rejected generation preserves the prompt and saved files. A changed credit price refreshes the estimate and waits for another explicit Generate action.
Updated 2026-10-17: a retention generation completes only after current payment, runtime and physical-storage proof, all captured file and post receipts, and unchanged scope pass the final worker check. One transaction records completion and releases the fences into a fresh storage generation. Replaying that result does not reopen the old generation. Old payment requests remain closed; a later purchase needs a new request. Missing proof, late conflicts or an expired worker claim leave both fences in place. Cleanup remains disabled.
Updated 2026-10-19: retention admission can use current evidence from a separately authorized runtime verifier. Application, owner, agent and billing-worker roles cannot create this evidence. It binds the original account, payment certificate, storage generation, configured namespace, artifact and physical scan. Missing, changed, expired or invalidated evidence blocks new deletion and completion. The original DELETE window stays ten seconds; expiry never proves that an earlier writer stopped. The supported local integration control plane uses the ordinary storage factories and real private S3 calls. It does not certify Railway or Customers with Stripe identities. Cleanup remains disabled.
Updated 2026-10-18: caption requests keep an immutable database link to the source post. An admitted retention scrub removes their saved instructions, previous suggestions and output, while keeping financial receipts and request identity. New caption work after schedule capture invalidates that old content proof. Historical captions without a trusted post link block content admission and completion; a JSON post ID is not deletion authority. Runtime and provider closure checks remain required, and cleanup stays disabled.
Updated 2026-10-18: retention-storage-cancel-pending keeps a canceled storage fence visible in operator findings after payment access is restored. It remains visible after sender drain until the original cancellation is repeated and the fence is released. Upload and website senders record not_sent only when the original invocation proves that the SDK call never began. An abort, timeout or rejected response after dispatch remains unknown; it does not release byte reservations or permit deletion.