- Drag-reorder used verticalListSortingStrategy on a multi-column grid,
which computes wrong transforms for grid reflow — swapped to
rectSortingStrategy so cards actually animate live while dragging.
- Grip handle was rendered as a sibling of (not a descendant of) the
`group` element its `group-hover:opacity-100` depended on, so it was
permanently invisible. Fixed the DOM nesting and made it always
partially visible instead of hover-only.
- common.edit was missing from both locales (not just French) —
edit-collection-dialog.tsx was the first caller to hit it.
- Root cause of "generated in my language but Translate still shows":
generate-meal, meal-plan/generate, and adapt never set recipes.language
on the row they inserted, so the button's `!recipe.language || ...`
check always fell back to "show it". Fixed at all three insert sites.
- Translate dialog was entirely hardcoded English (title, description,
language names, buttons) despite i18n keys already existing for most of
it — now uses them, plus new translated language-name keys.
- Recipe tags now render on the recipe detail page (previously grid-card
only).
- Collection header actions converted to icon-only + tooltip, matching
the recipe page's pattern instead of icon+label buttons.
- Collections list search now also matches recipe titles inside each
collection, not just the collection's own name/description.
- Explore page links to /collections/explore next to its tabs.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
New "Generate meal" button on any owned collection — pick a theme (free
text) and 2-6 courses (starter/main/side/dessert/drink), one generateObject
call produces a coherent recipe per course (shared cuisine/style, matching
flavors across courses) and all of them get saved as real recipes and
added to that collection in one action.
Follows the meal-plan generation route's established pattern: single
withAiQuota charge for the whole generation, then a pre-flight loop
charging the tier's recipe limit once per generated recipe (rolled back
on a partial breach) before the insert transaction runs. Recipes are
tagged with their course name for later filtering; visibility defaults to
private like every other AI-generated recipe.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Admins can now disable specific AI features per tier from Admin > Tier
Limits — new feature_flags table (feature x tier -> enabled, defaulting
to true so adding a new gated feature never needs a backfill).
Covers recipe variations, drink pairing, and meal pairing to start.
When disabled for a user's tier, the button stays visible (with a small
lock badge) but opens an upgrade dialog instead of running; the API
route rejects the call server-side either way (requireFeatureEnabled,
re-reads tier from the DB rather than trusting the session's cache,
same rationale as checkAndIncrementTierLimit).
The upgrade dialog is informational only — no Stripe checkout exists
yet (STRIPE_PLAN.md is still just a plan) — its CTA links to /support
prefilled as an upgrade-interest suggestion.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
All literal "team" tier-value references renamed to "family" across
API routes, admin UI, OpenAPI schemas, and lib/tiers.ts. The DB enum
value itself is renamed in place via ALTER TYPE ... RENAME VALUE
(migration 0044) rather than drizzle-kit's auto-generated
drop-and-recreate-the-enum migration, which would have failed against
any existing row still holding 'team' — RENAME VALUE preserves
existing data with no cast/backfill needed.
Also adds STRIPE_PLAN.md — a full Stripe billing integration plan
(Checkout+Portal, tier→Price mapping, admin billing dashboard, and a
multi-user Family-group design since Family is meant to cover several
accounts under one subscription, not one payer). Planning only, no
Stripe code yet.
v0.47.0
The system prompt only said "you can propose creating a recipe with
the tool" as a soft option — models were defaulting to just writing
the recipe out as prose in the reply instead of invoking the tool,
since that's the easier default behavior for a "write me a recipe"
request. Rewrote the instruction as a hard rule: full recipes must
go through createRecipe, plain-text ingredient lists/numbered steps
are disallowed in the reply for that case.
v0.46.1
Second tool alongside createRecipe, same safety shape: addToShoppingList's
execute only validates/echoes input, no DB write. The proposal card lets
the user pick an existing list (fetched lazily) or name a new one, then
confirms through the exact two-call flow AddToShoppingListButton already
uses — POST /api/v1/shopping-lists (if new) then POST .../items — no new
persistence code path.
generateMealPlan intentionally not built as a tool: the existing
/api/v1/ai/meal-plan/generate endpoint generates and writes in one step
with a week/preferences input, not a specific plan payload, so it has no
"save this exact draft" entry point to confirm against without a larger
refactor. Documented as a known gap rather than forcing a weaker pattern.
v0.46.0
Gives the general chat a createRecipe tool (Vercel AI SDK, stepCountIs(3))
scoped so it can only ever produce a draft — the tool's execute is a pure
echo, no DB write. The route surfaces the tool call as proposedRecipe
alongside the normal text answer; the chat UI renders it as a card with
explicit Create/Discard buttons. Create POSTs to the existing
/api/v1/recipes endpoint — the same code path and tier/recipe-count limit
the manual editor already goes through — so there's exactly one place
that actually creates a recipe row, and nothing happens without the user
clicking Create.
Scoped to the general assistant only (not per-recipe chat), and to one
tool for now — addToShoppingList/generateMealPlan are follow-ups.
v0.45.0
Widens the tier enum from free/pro to free/pro/team and every
"free" | "pro" cast that assumed exactly two tiers (~30 call sites:
every AI route's withAiQuota/checkAndIncrementTierLimit call, admin
user/invite management, upload quota checks, OpenAPI schemas). Team
sits above Pro with genuinely unlimited recipes/public-recipes (the
-1 sentinel, which Pro doesn't actually use — Pro uses large finite
numbers instead) and a higher AI-call/storage cap. Seeded via
db:seed, editable afterward from Admin > Tiers.
role (user/moderator/admin — permissions) and tier (free/pro/team —
billing limits) stay separate concepts, as they already were; this
does not touch role-based permissions.
Requires migration 0043 to run against a live DB — not applied in
this sandbox (no Docker here); run `pnpm db:migrate` then `pnpm db:seed`.
v0.44.0
The general assistant had exactly one conversation per user forever
(recipeId null on chat_messages) — no way to start fresh or organize
by topic. Adds an ai_conversations table (title, timestamps) and a
nullable conversationId FK on chat_messages; the assistant panel gets
a conversations menu to create/switch/rename/delete, auto-titling a
new conversation from its first question.
Per-recipe chat is untouched — each recipe already has one natural
thread, so multi-conversation support only applies to the homepage
assistant. Pre-existing general messages (no conversationId) aren't
migrated into the new model and won't appear in the conversation list.
Requires migration 0042 to run against a live DB — not applied in
this sandbox (no Docker here); run `pnpm db:migrate`.
v0.43.0
Every existing AI entry point (generate, generate-from-idea, adapt,
variations) either creates a new recipe or a saved variation — none
let you revise the draft you're currently editing in place. Adds a
stateless /api/v1/ai/regenerate endpoint that takes the editor's
current in-progress fields (not a recipeId, so unsaved edits are
included) plus a free-text instruction, and returns a full revised
draft the editor merges into its own state. No DB write happens;
the user still saves normally.
v0.42.0
AI adapt/variations flows had no catch on their fetch calls, so a
network failure surfaced as nothing happening rather than an error
toast. They also unconditionally persisted the AI's output even when
it was identical to the source recipe, and the adapt route never
recorded a recipeVariations row (only plain forks did) or enforced
the per-tier recipe-count limit other create paths already check.
v0.39.0
The photo-import flow used one vision-capable model to both read the
photo and structure the full recipe (quantities, steps, timing) in a
single call. Split it into two: a vision model recognizes what's in
the picture, then a text model reconstructs the recipe from that
description — same pattern the pantry photo-scan already uses for
recognition, and lets structuring use whichever model is actually
configured for text generation. Still counts as one AI-quota unit.
v0.37.0
URL import extracts content in the source page's own language (no
translation), but never recorded what that language was, so the
Translate button's "recipe.language !== my locale" check always saw
null and showed the button unconditionally. The extractor now returns
a detected ISO 639-1 language code alongside the recipe.
Photo import always writes in the caller's locale already (see its
langInstruction) but never recorded that either — now sets it
directly to locale since it's known, not detected.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
generate, generate-from-idea, URL import, and photo import all now
have the AI set recipeType directly (defaulting to "dish" whenever
ambiguous, per instruction), with cookMins force-cleared for drinks
as a backstop in case the model doesn't follow the prompt guidance.
1-serving-by-default for unspecified drink quantities is left to the
model's own judgment of the request text, since there's no reliable
way to detect "was a serving count stated" without re-parsing intent.
Drink recipes get a distinct default icon (no cover photo case), and
Explore gains the same dish/drink Type filter My Recipes already had.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Full list of the audit's confirmed findings and their fixes:
- Stored XSS via unescaped JSON-LD on the public recipe page
(app/r/[id]/page.tsx) — escape < before injecting.
- CSP allowed unsafe-eval in production — now dev-only (Next prod
never eval()s; only its HMR does).
- avatarUrl accepted any URL with no ownership check — now takes an
avatarKey issued by avatar-presign, validated server-side, same
pattern as recipe/review photos.
- No session revocation on password change/reset — both now revoke
other sessions (revokeOtherSessions: true, revokeSessionsOnPasswordReset).
- Rate-limit bypass via spoofable X-Forwarded-For — take the last
(proxy-appended) hop instead of the first (client-supplied) one,
matching the single-Traefik-hop topology.
- Webhook signing secrets stored plaintext — now AES-256-GCM
encrypted like every other secret in this app, with a legacy-
plaintext fallback for pre-existing rows (bare hex has no ":", our
ciphertext format always does).
- Better Auth's own rate limiter defaulted to in-memory storage,
ineffective across replicas — now backed by the same Redis as
lib/rate-limit.ts (secondaryStorage), with storeSessionInDatabase
explicit so session storage itself doesn't move as a side effect.
- Presigned upload URLs didn't bind the declared file size to the
actual upload, letting a client under-declare size (and quota
charge) then PUT an arbitrarily large object — switched to S3
presigned POST with a signed content-length-range condition,
enforced by the storage server itself.
- generateMetadata() on the recipe page skipped the visibility
filter the page body uses, leaking a private recipe's title via
<title> to any signed-in user with the id.
- Block/unblock had no rate limit, unlike follow/unfollow.
- AI quota was charged even when a user's own BYOK key was used
(their own credentials/billing) — added an isByok flag through
the config-resolution chain and skip the charge when set. Also
wired BYOK into generate/generate-from-idea/translate/import-url,
which never looked it up at all before.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Add DEFAULT_{TEXT,VISION,MEAL_PLAN}_{PROVIDER,MODEL} site settings,
editable from Settings -> Admin (new AdminDefaultModelForm, mirroring
the per-user model-prefs UI). getDefaultProviderWithKey now accepts
the use case and checks the admin default (after the user's own BYOK
key, before the old "first configured site key" heuristic) so admins
can pin a specific provider/model per feature instead of it being
whichever key happens to exist.
Wired into scale/substitute/batch-cook/meal-plan generation routes,
which previously called getDefaultProviderWithKey() without a use
case and so never had visibility into per-feature routing at all.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
importFromPhoto() accepted a locale param but silently discarded it
(named _locale, never read) — now builds a language instruction the
same way generate-recipe.ts does for text generation.
The AI output schema had no way to signal "couldn't find a recipe in
this image" — generateObject would always fabricate a title/fields.
Added a `found` boolean the model sets to false in that case; the
route now returns 422 instead of creating a garbage recipe and
sending the user into the editor with it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Convert requireSession -> requireSessionOrApiKey across recipes,
collections, meal-plans, shopping-lists, pantry, feed, and ai/*
(52 routes) so API keys work end-to-end, not just for the handful of
endpoints that supported them before. Scope was explicitly confirmed
per-resource-family with the user before any file was touched.
Left session-cookie-only, deliberately: users/me*, ai-keys/*,
webhooks/*, conversations/*, notifications/*, push/subscribe, admin/*
— account/credential-adjacent surface that shouldn't widen without a
separate, explicit decision.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Manual: DELETE /api/v1/ai/chat-history (scoped by recipeId or scope=general,
matching GET's scoping) plus a trash-icon button + confirm dialog in both
chat panels' headers.
Automatic: new internal cron endpoint (chat-cleanup), same shared-secret
pattern as the existing leftover-reminders/weekly-digest crons, deletes
any chat_messages older than 90 days. Wired into cron/crontab (daily,
03:00 UTC) and the Dockerfile's cron stage.
Verified locally: cleared a real conversation through the actual UI and
confirmed it didn't come back on reopen (not just cleared client-side);
inserted a 100-day-old and a 5-day-old message directly, called the cron
endpoint with the real shared-secret check, confirmed only the old one
was deleted and the recent one survived; confirmed the endpoint 401s with
no/wrong secret.
Both chat panels (per-recipe and the general cooking assistant) previously
kept messages in component state only — closing the sheet or reloading
the page lost everything. Both POST routes now persist question+answer
pairs to a new chat_messages table (recipeId null for the general
assistant), and both panels load their own history back on open.
New GET /api/v1/ai/chat-history (scoped by recipeId, or scope=general for
the homepage assistant, or neither to search across everything) backs a
new search control in both panels' header — debounced, shows matched
messages with date and (when searching across everything) which recipe
they belonged to.
Verified locally: asked a real question, closed and reopened the panel
and confirmed the conversation reloaded, then searched a keyword from
that conversation and confirmed both the question and answer surfaced
in the results dropdown.
Moved the homepage cooking-assistant button from bottom-24 to bottom-16 —
still clears the recipe list's floating bulk-action bar, just not as far.
Both chat system prompts (cooking-chat and the per-recipe chat) now
explicitly identify as "Epicure" and are told never to name the underlying
model/provider if asked — previously they'd just answer honestly with
whatever the model calls itself.
Verified locally: screenshotted the new button position, and asked "what
model or AI are you" through the real UI — response identified as Epicure
with no model name.
The existing recipe chat (RecipeChatPanel) is tightly scoped to one
recipe — recipeId is required and its system prompt embeds that recipe's
full ingredient/step context. Added a sibling, not a modification: a new
floating assistant on the recipes list page for questions that aren't
about any specific recipe (technique, substitutions, timing, food safety).
New /api/v1/ai/cooking-chat route (same quota/rate-limit/model-resolution
pattern as recipe-chat, minus the recipe lookup) and CookingAssistantPanel
component (same floating-button + Sheet UI). Positioned at bottom-24
rather than bottom-6 so it doesn't sit in the same band as the recipe
list's floating bulk-action bar during select mode.
Verified locally: asked a real cooking question through the actual UI,
got a correctly-formatted markdown answer with no recipe context leaking
in (confirmed the request payload only carries the question, no recipeId).
- AI recipe generation always used the app's own locale rather than a
separate language picker — removed the picker, generation and the
saved recipe's language now both follow useLocale() directly.
- Bulk-select action bar had an ungated max-w-md that capped it at 448px
even past the sm: breakpoint where sm:w-auto was supposed to let it
grow to content — added sm:max-w-none so desktop never scrolls.
- Batch-cooking generation had no free-text "describe" field like the
standard recipe generator does — added one (BatchCookFields, both the
standalone dialog and the AI dialog's batch tab), threaded through the
API route and generateBatchCook's prompt as additional guidance.
- Recipes imported from a URL never actually got their sourceUrl
persisted — CreateRecipeSchema didn't declare the field, so Zod
silently stripped it before it ever reached the insert. The detail
page's "Source: <hostname>" link already existed but was dead code
for every real import until now.
Verified all 4 live: generate dialog has no language selector, batch
tab has a working describe textarea, bulk-select bar renders at full
content width with zero horizontal overflow on desktop, and a recipe
created with sourceUrl now round-trips and renders its source link.
Shopping list add already worked generically for batch-cook recipes —
no code needed there.
New: mark a specific batch-cook dish as "cooked today", track its
fridge expiry (cookingHistory.batchDishId), surface a "Leftovers
expiring soon" widget on the pantry page, and send a daily push+email
reminder via a new /api/internal/cron/leftover-reminders endpoint
(mirrors the weekly-digest cron pattern; doesn't use the social
notifications table, which requires a non-null actor and isn't built
for self-reminders).
Also fixes, from user-reported bugs:
- Recipe cards showed no batch-cook badge/dish-count/prep-time in some
views — added dishCount + prepMins/cookMins (now generated by the AI
and persisted) to the card component and /recipes query.
- Batch-cook descriptions occasionally contained raw markdown
(**bold**) — added explicit "plain prose only" prompt instructions
and a stripMarkdown() defensive fallback at render time.
- Truncated/cut-off descriptions — the generateObject call had no
maxOutputTokens set, so long structured responses could get cut off
mid-field; now capped explicitly at 8000.
- Generate dialogs (batch-cook + the main AI dialog) could show
buttons unreachable once the progress bar appeared mid-generation —
restructured so the action row is pinned outside the scrollable
content area, not affected by content height changes.
- /api/internal/* routes were being redirected to /login by middleware
before their own CRON_SECRET check ever ran (pre-existing bug,
affected the weekly-digest cron too) — added to PUBLIC_PATHS.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
New AI-generated batch-cooking flow: pick N dinners/lunches + servings,
get one unified recipe covering a full prep session — merged shopping
list, steps tagged per-dish (a step can advance several dishes at once,
e.g. a shared oven bake), and per-dish storage (fridge/freezer) + exact
day-of reheat/finishing instructions.
Schema: recipes.isBatchCook, recipeSteps.appliesTo (dish names), new
recipeBatchDishes table. Batch recipes are excluded from the normal
/recipes list and live under their own /batch-cooking section.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Six M-sized items from HANDOFF.md's new-features backlog:
- Profile tabs: cooking-history stats (total cooked, last-cooked, streak)
and a "cooked it" photo gallery, both owner-only
- Display-time unit conversion (metric<->imperial) for recipe ingredients,
respecting each user's unitPref; original value always shown alongside
the conversion
- Nutrition daily diary: per-day macro totals computed from cooking history
x recipe nutritionData, compared against user goals
- Pantry scan: real barcode lookup (zxing + Open Food Facts, no API key)
with an AI-vision fallback for unbarcoded items, always confirm-before-
insert, both paths tier/rate-limited like other AI features
- Weekly digest email: new followers/comments/ratings + trending recipes,
sent via a new `cron` Docker stage (alpine+crond+curl) and `digest-cron`
compose service hitting a bearer-token-protected internal route
- Meal-plan generation can now target a user's nutrition goals as a
prompt-level nudge (recipes are AI-invented, not DB-sourced, so this
can't be a hard macro constraint)
Caught a real deploy-breaking issue while adding the cron stage: appending
it after `runner` silently changed the Dockerfile's default build target,
and `web`'s compose config didn't pin one — fixed by pinning `target:
runner` explicitly. Verified with typecheck, lint, and three separate
`docker build --target` runs (runner/cron/migrator) plus `docker compose
config` validation.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Full audit (bugs/UI-UX/backend/feature-gap) turned up a money-leak AI quota
bypass, webhook SSRF, and a long tail of missing pagination/auth/a11y work.
Fixes land together since HANDOFF.md tracked them as one backlog.
- AI routes charge tier quota before generating; nutrition POST is author-only
- Webhook dispatch re-validates URL per delivery (SSRF/DNS-rebinding), treats
redirects as failures; recipe.published now actually dispatches
- New indexes/unique constraints on recipes, meal-planning, comments FK cascade
- Recipe PUT/restore snapshot only inside the transaction, after validation
- Recipe DELETE cleans up S3 objects (recipe + review photos)
- Optimistic UI (favorite/star/follow/shopping-list) rolls back on failure
- Upload presign enforces file size cap + per-tier storage quota
- Route-level loading/error/not-found states across (app), admin, and root
- middleware.ts guards (app)/admin; requireAdmin checks DB role, not cached
session; rate limiting applied to both session and API-key branches,
bucketed per key; Stripe webhook dedupes by event id
- Pagination added to recipes, feed, profile, comments, pantry, admin tables
- Nav shows real avatar + profile link + dark-mode toggle; destructive actions
standardized on AlertDialog
- Unsaved-changes guard + real ingredient/step validation on recipe form;
canonical /recipes/[id] used in-app; next/image migration; aria-labels and
alt text across icon buttons, avatars, recipe photos
- packages/api-types removed (zero callers, too drifted to safely rewire);
openapi.ts and ai-keys error shape drift fixed; BYOK decrypt failures now
surface instead of silently falling back to the platform key
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
AI routes had no consistent error handling — a provider failure (e.g.
OpenRouter returning a degraded-model error) crashed the route as an
uncaught 500 and, worse, still charged the user's monthly aiCall
quota for a request that never succeeded.
Adds withAiQuota()/aiErrorResponse() (lib/ai/ai-error.ts) and applies
them across all 12 AI generation routes: charges the quota, runs the
AI call, refunds the credit and returns a clean user-facing message
if it throws. Frontend needs no changes — existing dialogs already
toast whatever `error` string comes back.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Fixes the 13-item codebase health scan backlog: wraps meal-plan
generation in a transaction, adds missing userId/GIN indexes, fixes
an IPv6-parsing gap in the webhook SSRF guard (and an identical
duplicated bug in the AI URL-import path, now consolidated onto one
implementation), paginates the collections list, dedupes the AI
recipe Zod schemas, wires up Stripe tier sync, rate-limits AI key
rotation, gets `pnpm typecheck` actually working, and adds test
coverage for the previously-untested admin/webhooks routes.
Two flagged issues (collection removeRecipeId IDOR, tier-limit race)
turned out to already be fixed/non-issues on inspection — noted in
TODO.md rather than silently dropped.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Recipe cards, the recipe detail page, and meal planner still rendered raw
"{n} servings"/"{n}m cook"/"{n} srv" text and untranslated difficulty labels
outside the earlier i18n pass. Also wires the user's locale into AI features
that previously always answered in English regardless of app language:
recipe chat, ingredient substitution, recipe ideas, and generate-from-idea.
The recipe-language picker in the AI generate dialog now defaults to the
user's locale instead of always defaulting to English.
- Substitute finder never applied the user's BYOK/admin AI key (only fell
back to raw process.env), so it silently failed whenever the key was
stored via settings rather than a literal env var. Now resolves via
getDefaultProviderWithKey like every other AI route. Popover also
surfaces real error messages instead of swallowing failures.
- Ingredients with quantity 0 (salt, pepper, "to taste") rendered the
literal digit "0" in cooking mode, print view, public recipe page,
shopping lists, and serving scaler — several sites relied on generic
truthiness/filter(Boolean), which doesn't catch a stored "0" string.
Added a shared hasQuantity() helper and applied it everywhere quantity
is rendered, plus in the AI recipe-chat context sent to the model.
- Recipe chat panel rendered a duplicate close button on top of shadcn
Sheet's own built-in close X, producing a garbled overlapping glyph.
Removed the duplicate.
- Recipe chat assistant replies are markdown from the model but were
rendered as raw text; added react-markdown so formatting actually
renders.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Arnaud