Sprint 13 (commit bae9403) set OLLAMA_MODEL=kimi-k2.6:cloud.
kimi-k2.6 is a reasoning model that burns the entire max_tokens=800
budget on internal reasoning and never produces the JSON answer
for the Sprint 13 prompt. Every /api/llm/plan call has returned
picked_count=0 since 2026-06-05. The library fill (Sprint 6+)
silently took over, masking the bug. Every "Ask the LLM" click
paid Ollama costs for nothing.
Discovered while answering the user's "is there anything else to
refine?" question. Added a temp debug log to _ask_llm, saw
raw_response='' with finish_reason: length. Verified on Ollama
Cloud: gpt-oss:20b (OpenAI's open-source 20B non-reasoning
model) returns 21 valid picks in 2074 chars on the same prompt.
finish_reason: stop. Reasoning field is 239 chars vs kimi-k2.6's
8206+ chars.
Two-line fix:
- backend/app/config.py:38 — OLLAMA_MODEL: str = "gpt-oss:20b"
(was "kimi-k2.6:cloud")
- backend/app/api/llm_plan.py:117 — max_tokens: 4000 (was 800).
21 picks × ~100 chars + reasoning + boilerplate ≈ 2100+ chars;
4000 gives 2x headroom.
Plus the host's .env (or docker-compose env) was also set to
OLLAMA_MODEL=gpt-oss:20b — pydantic settings read env first, so
the .env change is what actually fixed the running container. The
config.py default is a backup for new deploys.
Plus frontend/src/api/llm.test.ts (NEW, 4 cases) — Vitest
contract test on the LLM response shape. Locks plan_id (UUID),
picked_count / filled_count / failed_count (non-negative integers
summing to ≤ 21), and reasoning (string|null). Catches
response-shape regressions so a future model swap that breaks
the JSON contract is caught at npm test time. The 4 cases: 8a
(POST to /llm/plan with payload), 8b (response.plan_id is a
valid UUID), 8c (counts are non-negative integers summing to
≤ 21), 8d (reasoning is string or null).
Verified: 11/11 vitest cases pass (4 new from S16 + 7 from S14).
npm run build green. Live API: 5/5 test weeks return picked_count
15-21 (was 0/5 before). Backend env verified:
docker exec mealplanner-backend-1 env | grep OLLAMA_MODEL →
gpt-oss:20b. No new runtime dependencies. No migration. No
schema change. No UI change.
Deploy: git pull + docker compose up -d --build backend frontend.
The .env change should already be in place; verify with
docker exec mealplanner-backend-1 env | grep OLLAMA_MODEL.
F9-lite reuses the pre-existing OLLAMA_* config (config.py:36-38:
OLLAMA_BASE_URL=https://ollama.com/v1, OLLAMA_API_KEY,
OLLAMA_MODEL=kimi-k2.6:cloud). Avoids the local model pull
(F9-full would be 4 GB on disk + a separate uvicorn process).
Cloud LLM — operator’s existing OLLAMA billing applies per call.
Sprint 13 splits the Sprint 11 "Generate Meal Plan" CTA into a
2-step modal: "Use the recipe library" (default, Sprint 11’s
existing flow) or "Ask the LLM" (new). The LLM path POSTs to
/api/llm/plan with a free-text prompt; the backend calls
kimi-k2.6:cloud on ollama.com, parses the LLM’s JSON picks,
creates a fresh plan, fills the LLM’s picks, and falls through
to the Sprint 6+ fillEmptySlots pattern for the slots the LLM
didn’t cover.
Backend:
- backend/app/api/llm_plan.py (NEW, ~280 lines). 1 endpoint
(POST /api/llm/plan body {prompt, week_start}) + 4 helpers:
- _ensure_ollama_configured — 503 on missing OLLAMA_API_KEY.
- _serialize_library — reads up to 200 recipes for the
family, sorted alphabetically. Cap prevents prompt-token
overflow on kimi-k2.
- _ask_llm — mirrors llm_matcher._ask_ollama (same URL,
same headers, max_tokens=800, temperature=0, strips think
blocks, 60s timeout).
- _parse_picks — tolerant JSON parser. Handles markdown code
fences, trailing commentary, and bare JSON. On failure
returns []; the library fill takes over.
- _validate_picks — drops invalid entries: missing fields,
out-of-range day_of_week, unknown meal_type, unknown
recipe_id. Returns a list of LLMPickedItem.
Flow: rejects duplicate week (400) and empty library (400),
builds the prompt, calls the LLM, validates picks, creates
the plan, inserts the LLM-picked items, fills the rest from
the library (Sprint 6+ pattern, re-implemented inline to
avoid a self-HTTP-call), returns {plan_id, picked_count,
filled_count, failed_count, reasoning}.
- backend/app/schemas/__init__.py — added LLMPlanRequest +
LLMPlanResponse.
- backend/app/main.py:65-66 — registered llm_plan_api.router
at the /api/llm prefix. No collision with the pre-existing
WIP recipes.py.
Frontend:
- frontend/src/api/index.ts — added llm.plan(data) method.
- frontend/src/pages/Dashboard.tsx — added the prompt modal
(radio for library vs. LLM + textarea for the LLM path with
500-char counter) + new state (showPromptModal, promptMode,
promptText, promptBusy) + extracted Sprint 11’s body into
generateFromLibrary + added generateFromLLM. The modal is
inline (not a separate component) because it depends on 4
local states + 3 handlers. Click-outside-to-dismiss is
disabled while promptBusy is true. The textarea autoFocuses
when LLM mode is selected. Added the Button import.
LLM tolerance: a 60s timeout, parse-failure (markdown code
fences, trailing commentary), or empty response all return 0
picks; the library fill takes over. The user never sees a
crash — at worst, picked_count: 0 and the toast reads "Planned
N meals (LLM picked 0, library filled the rest)".
Verified: npm run build green (tsc 0 errors, vite 0 errors).
Bundle: 500.28 → 503.82 kB (+3.5 kB). Backend AST clean on
all 3 changed files. No new dependencies, no migration, no
pre-existing WIP files touched.
Deploy: git pull + docker compose up -d --build backend
frontend (no migration, no new dependencies).
Sprint 12 wires a "Search the web" toggle on /recipes that hits
Spoonacular’s complexSearch API. Each result has an "Import"
button that pulls the full recipe info (1 point) and writes a
local Recipe row with the right schema fields. Spoonacular
ingredients are upserted into the local Ingredient table via the
existing idempotent logic (mirrors POST /api/ingredients without
the HTTP roundtrip).
No pre-existing WIP files touched. Sprint 12 creates a new
backend/app/api/recipe_search.py router (separate from the WIP
recipes.py) and adds 2 Pydantic models to backend/app/schemas/
__init__.py (the canonical location). The WIP recipes.py is
registered in main.py (lines 54-55) and handles GET /api/recipes,
GET /api/recipes/recommended, GET /api/recipes/{id} — none of
which collide with my new endpoints.
Backend:
- backend/app/api/recipe_search.py (NEW, ~270 lines). 2 endpoints:
- GET /api/recipes/search?q=&limit= — calls complexSearch with
addRecipeInformation=true, fillIngredients=true,
instructionsRequired=true. Returns normalized
RecipeSearchHit[]. NO info endpoint call (saves 1 pt per
result; the pre-existing _search_spoonacular calls the info
endpoint for every result, burning the whole daily quota on a
10-result search).
- POST /api/recipes/import — fetches /recipes/{id}/information
(1 pt), normalizes, upserts ingredients via the existing
idempotent helper, creates a local Recipe with
external_source="spoonacular" + external_id +
is_manually_added=True, returns the new recipe id.
- Process-wide _points_used counter (module-level singleton +
threading.Lock). 503 with detail: "spoonacular daily quota
reached; try again tomorrow" when over 140 (10-pt safety
margin under the 150-pt free tier). Resets on process restart.
- 503 with clear "SPOONACULAR_API_KEY not configured" when env
var unset.
- Idempotent import: 409 on duplicate (external_source,
external_id).
- backend/app/config.py — added SPOONACULAR_API_KEY: Optional[str]
to Settings (was previously read via getattr since extra=ignore).
- backend/app/schemas/__init__.py — added RecipeSearchHit +
RecipeImportRequest.
- backend/app/main.py:62-63 — registered recipe_search_api.router
at the /api/recipes prefix. No collision with the WIP.
Frontend:
- frontend/src/api/index.ts — added 5 new methods to
mealPlannerApi.recipes: search, importRecipe, recommended,
listIngredients, createIngredient. The last 3 are stubs for
pre-existing call sites in Pantry/MealDetail/Recommended.tsx
that were previously hidden by a smaller API surface.
- frontend/src/pages/Recipes.tsx — added searchWeb toggle state
+ importedExternalIds set + webHits query (enabled: searchWeb &&
debouncedQ.length >= 2) + importMutation (toast on success,
showApiError on failure) + the toggle button (with
aria-pressed={searchWeb}) + the web-search panel (<div
role="region" aria-label="Web recipe search"
aria-busy={webLoading}>). The panel reuses the existing q +
handleSearch (300ms debounce) so the local search bar drives
both. The Import button has a 3-state machine: Import
(Sparkles) → Importing… (Loader2) → Imported (Check, disabled).
- frontend/src/types/index.ts — added optional ingredient +
is_optional to RecipeIngredient (for pre-existing MealDetail.tsx
call sites).
Verified: npm run build green (tsc 0 errors, vite 0 errors).
Bundle: 496.48 → 500.28 kB (+3.8 kB). Backend AST clean on all 4
changed files. Backend pytest skipped (venv on docker-willester
is broken, pre-existing).
Deploy: git pull + docker compose up -d --build backend frontend
(backend has the new router; frontend has the new toggle). No
migration, no new dependencies.
User-driven follow-up to Sprint 8: surface the Sprint 1-3 NeverSuggest
infrastructure on the Recipes surface so a family can pre-emptively
mark a recipe as never-suggest before it appears in a plan.
Backend (3 changes):
- POST /api/never-suggest (public, webui-facing). Idempotent on
(family, recipe, reason). Returns the row joined with recipe_name.
- DELETE /api/never-suggest/{ns_id} (public, webui-facing). Row-level
ownership check (403 if cross-family), 404 if absent.
- NeverSuggestRead.recipe_name + .ingredient_name server-side joins
via _attach_names() helper (one LEFT OUTER JOIN per kind).
- Admin path (POST/DELETE /api/admin/never-suggest) unchanged.
Frontend (4 changes):
- New NeverSuggestButton component (~290 lines). Two variants: card
(overlay on RecipeCard) and detail (text buttons in RecipeDetail
top bar). Popover with Allergy (red, window.confirm) + Dislike
(neutral, no confirm). Undo toast via showToast.undo() (Sprint 3
B12 pattern, 6s window). Pre-existing block detection shows a
Blocked state with an Unblock path.
- mealPlannerApi.neverSuggest.list/add/remove in api/index.ts.
- Recipes.tsx overlay: RecipeCard has position: relative; button is
opacity-0 group-hover:opacity-100 focus:opacity-100. e.preventDefault
+ e.stopPropagation prevents accidental navigation.
- RecipeDetail.tsx top bar: new Deny forever button group to the left
of Add to Plan.
Build: npm run build green (tsc 0 errors, vite 0 errors) on
docker-willester. Bundle 487 -> 495 kB. No new dependencies. No
migration (NeverSuggest table exists from prior sprints).
Tracking: Review/sprint10-verification.md (9-step browser smoke +
5 API curls + undo test + a11y check).
User policy decision (2026-06-05, exact): 'Hard filter. If it is denied
this week twice, it should be considered denied for good.'
The planner had no cross-week memory of denials: a denial on
meal_plan_item.approval_status was never consulted by the planner,
and NeverSuggest (the per-family permanent blocklist) was empty for
the user. The 'Roasted Sweet Potato and Chickpea Bowl' the user
denied on 2026-05-15 was still in the planner's pool 3 weeks
later.
Implements C + Z (explicit two-button model + soft-decay +
hard-filter escalation):
- Approve: untouched.
- Deny this week (1st in 90d): denial_expires_at = now() + 90d.
- Deny this week (2nd in 90d, server-side auto-escalation):
denial_expires_at = NULL + a NeverSuggest row written.
- Never again (explicit): same as the 2nd-time auto-escalation.
Both soft and permanent denials are hard filters in the planner
(per user). A denied recipe never reappears until either the 90d
window expires or the user un-blocks via the NeverSuggest API.
Changes:
- Migration 0016: meal_plan_item.denial_expires_at (partial index)
and meal_plan_vote.denial_scope.
- 3 backend helpers (_apply_denial, _ensure_never_suggest_recipe,
_has_prior_active_soft_denial) — single source of truth for the
deny path.
- POST /api/meals/items/{id}/deny?scope=this_week|never_again
(default this_week). Returns promoted_to_permanent.
- POST /api/meals/vote/{id} extended: vote=approve|deny|never_again.
Returns denial_scope + promoted_to_permanent.
- GET /api/meals/vote/{id} HTML page renders 3 buttons; supports
one-click ?scope=... for email direct-action links.
- Email template (step_email): 3 direct-action links per recipe
plus a secondary 'open vote page' link.
- Planner: _load_blocklists returns 3 sets; soft_denied_recipes
is hard-filtered (union with blocked_recipes at the call site).
- Frontend: MealCard renders 3 buttons (Approve / Deny this week
/ Never again) for pending items. handleDeny is scope-aware;
toast reflects promoted_to_permanent. window.confirm on
'Never again' prevents accidental permanent blocks.
Verification:
- npm run build green.
- 21/21 planner tests pass (1 pre-existing test_filter_blocks_by_cost
failure is NOT introduced by Sprint 8 — verified via git stash).
- Review/sprint8-verification.md: 11-step browser smoke + 4 API
curls + email-render procedure + rollback.
Files:
- backend/alembic/versions/0016_denial_decay_and_scope.py (new)
- backend/app/models/__init__.py:221-242, 250-269
- backend/app/schemas/__init__.py:204-219, 248-269
- backend/app/api/meals.py:30-138 (helpers), 240-330 (HTML page),
380-455 (submit_vote), 486-552 (deny_meal_item)
- backend/app/services/orchestrator/steps.py:283-300
- backend/app/services/planner/generate.py:59-99, 150-194
- frontend/src/api/index.ts:48-58
- frontend/src/pages/Dashboard.tsx:38-50, 385-410
- Review/{sprint8-verification,ui-nielsen-audit,handoff-ui-audit}.md
- fix-ui-audit.md
- docs/HANDOFF.md
- .agent/{plan,context}.md
Deploy (user runs on deployment host):
cd ~/MealPlanner && git pull
docker compose exec backend alembic upgrade head
docker compose -f docker-compose.yml up -d --build backend frontend
F3 — Bulk 'add checked to pantry' on ShoppingList (the audit's F3 /
H7 finding). ShoppingList already had a 'checked' Set keyed on
ingredient_id and persisted to localStorage — that selection state
is the natural substrate for a bulk action.
Backend (POST /api/pantry/bulk):
- New endpoint that accepts {items: HomePantryCreate[]} and returns
HomePantryBulkResult with per-item status (added / updated /
skipped) and totals. Each item follows the same upsert semantics
as POST /api/pantry (insert or overwrite qty/unit/expires_at).
- Items with an unknown ingredient id are reported as 'skipped'
with reason='Unknown ingredient' rather than aborting the batch.
Per-item failure is the chosen model (partial-success) so the
user gets a precise count of what actually went in.
- New Pydantic schemas: HomePantryBulkCreate, HomePantryBulkResult,
HomePantryBulkResultItem.
Frontend:
- mealPlannerApi.pantry.addBulk(items) is the API binding.
- ShoppingList gets a new 'Add N to pantry' primary button (next
to the existing Reset button) that appears when checked.size > 0.
Click → POST /api/pantry/bulk → toast shows 'added X, updated Y,
skipped Z' counts. On success, only the items that actually
landed in the pantry are removed from the checked set; skipped
items stay checked so the user can see what failed.
- Disabled state with 'Adding…' label while the request is in
flight; button text shows the count dynamically (matches the
F4 design language: tell the user what they're about to do).
F4 — Plan the whole week (the audit's F4 / H7 finding).
Backend (POST /api/meals/{id}/fill-empty-slots):
- New endpoint that takes {meal_types: [str, ...]} and fills every
empty slot in the plan whose meal_type is in the request. Per-day
iteration (1-7) per meal_type, skipping already-occupied slots.
Recipe selection: prefer un-used, fall back to any (same as the
existing generate-item).
- Per-slot failure model: never aborts mid-batch. Returns
FillEmptySlotsResult { filled: [{day, meal_type, item}],
failed: [{day, meal_type, reason}] }. Invalid meal_types
(e.g. 'brunch') return immediately with a single FailedSlot
explaining why.
- Same approval_status=pending semantics as generate-item.
Frontend:
- mealPlannerApi.meals.fillEmptySlots(planId, mealTypes) is the
API binding.
- New 'Plan the week' button on the Dashboard header (next to the
week-nav control from Sprint 5). Primary color, Sparkles icon,
ChevronDown caret indicates a dropdown. Disabled + spinner
('Planning…') while the request runs.
- Dropdown has two options: 'Dinners only' (sends
meal_types=['dinner']) and 'All meals' (sends
meal_types=['breakfast','lunch','dinner']). Each option has a
one-line secondary label explaining the action.
- Toast on success: 'Planned N meal slots' (full) or 'Planned N
of M meal slots — X failed (e.g. <reason>)' (partial). The
query is then invalidated so the new slots show up.
Files: backend/app/api/meals.py, backend/app/api/pantry.py,
backend/app/schemas/__init__.py, frontend/src/api/index.ts,
frontend/src/pages/Dashboard.tsx, frontend/src/pages/ShoppingList.tsx.
Build: tsc 0 errors, vite 0 errors. Bundle +3.6KB (the new code
fits in the existing chunk).
Curl smoke on local dev DB confirms both new endpoints behave as
designed: /api/pantry/bulk returns proper skipped count for
unknown ingredients, /api/meals/{id}/fill-empty-slots returns
the partial-success result for the dinners-only call.
F5 — Persistent week selector in URL (the audit's F5 / H7 finding).
Backend:
- GET /api/meals and GET /api/shopping-list now accept an optional
?week_start=YYYY-MM-DD query param. When set, the response is the
MealPlan for that week (any status). When omitted, behaviour is
unchanged: meals returns the latest plan; shopping-list returns
the latest approved/locked plan with fallback to latest.
- No new dependencies; uses FastAPI's Optional[date] Query type
which auto-validates the YYYY-MM-DD format.
- Files: backend/app/api/meals.py:30-57, shopping_list.py:27-60.
Frontend:
- New week helpers in lib/utils.ts: isoMonday(), parseIsoDate(),
shiftIsoDate(), formatIsoDate(). All UTC-based to match the
backend's date column. isoMonday returns the ISO date of the
Monday of a given date's week.
- api/index.ts: meals.getPlanned(weekStart?) and
shoppingList.get(weekStart?) take an optional ISO date string.
Axios drops undefined params, so callers can omit them.
- Dashboard: useSearchParams('week') reads the URL; if absent or
invalid, falls back to this week's Monday (so the default URL is
empty). The queryKey now includes weekStart, so navigating weeks
fetches the right plan. A new segmented control in the header
(chevron-left | 'This week' / 'Current' jump button | chevron-
right) lets the user step weeks; the jump button highlights
primary-50 when the displayed week IS the current week. 'This
week' clears the ?week param. Mutations (move/approve/deny/
delete/generate) now invalidate ['mealPlan', weekStart] so the
right week refetches.
- ShoppingList: same URL sync, same segmented control, same
weekStart in queryKey. The 'no plan' empty state branches on
isCurrentWeek: 'No shopping list yet' (current) vs 'No plan for
that week' (any other week). The local-storage check-state key
naturally isolates per week (it uses shoppingList.week_start_date
which is the server's view of the current plan's week).
Migration 0015 cast fix:
- Discovered while smoke-testing on the local dev DB: the
CASE expression in 0015_normalize_pantry_aisles.py failed
with 'operator does not exist: text = boolean' on the
varchar(100) aisle column. Root cause: the CASE branches were
inferred as different types (string vs NULL) so the SET
target type couldn't be unified.
- Fix: explicit ::varchar(100) cast on the CASE expression.
Also simplified the WHEN '' branch (was NULLIF(...) IS NULL
with implicit bool comparison). Tested on local dev DB:
alembic upgrade head now succeeds; the 21196 rows that the
Sprint 2 dry-run predicted actually normalize correctly.
This means Sprint 2's deploy was blocked on the same bug
(the deployment host would have hit the same error).
- Verified via curl: /api/shopping-list?week_start=2026-05-15
returns 25 items with aisles 'Meat & Seafood', 'Pantry',
'Produce', 'Dairy & Eggs' (the canonical labels the migration
produces). Pre-migration aisles like 'meat_seafood' are gone.
Build: tsc 0 errors, vite 0 errors. 7 files, +196/-22.
- Migration 0012 adds score (float) and components (jsonb) to meal_plan_item
- generate.py: populates score and components at create time
- schemas/MealPlanItemResponse: include score + components fields
- GET /api/meal-plans/{id}: returns persisted values instead of zeros
- api/meal_plans.py: /regenerate now passes exclude_recipe_ids into generate_meal_plan
- planner/generate.py: filter recipe_dicts by exclude_recipe_ids set
- image_generation.py: OpenAI gpt-image-1 client with prompt building, b64_json handling
- main.py: StaticFiles mount at /static for generated images
- admin.py: POST /api/admin/trigger-images endpoint for batch generation
- scripts/generate_images.py: CLI for batch image generation
- docker-compose.yml + nginx: volume mounts for static/images persistence
- Verify MealPlanItem.votes ↔ MealPlanVote relationship is correct; no model bug exists
Backend:
- POST /api/ingredients now checks name_lower and aliases before inserting
- Returns existing ingredient on 409 instead of throwing error
Frontend:
- Removed fragile 409-recovery logic from Pantry.tsx handleAdd
- Added aliases field to Ingredient type for case-insensitive matching
Fixes pantry add for ingredients like 'Carrots' whose canonical name is 'Carrot'
- backend: expose POST /api/ingredients on public router so frontend can create ingredients without admin token
- frontend/api: point listIngredients and createIngredient to /api/ingredients
- frontend/pantry: replace ingredient dropdown with searchable text input + fuzzy matching + auto-create
Same root cause as meal detail: recipe JSONB stores ingredient_id but
not name. Shopping list now looks up names from the Ingredient table
before aggregating quantities, so items show "3 cups onion" instead
of "Unknown".
Recipe JSONB stores ingredient_id but not name. GET /api/meals/items/{id}
now queries the Ingredient table and injects names into the response so
the frontend displays "3 cups onion" instead of just "3 cups".
Replace two-step ORM update with single UPDATE ... CASE statement.
Eliminates IntegrityError from SQLAlchemy flush order violating the
unique constraint (meal_plan_id, day_of_week, meal_type).
- Create backend/app/api/orchestrate.py — new router for workflow steps
(scrape, generate, email, reminder, deadline, finalize) without admin auth.
- Remove orchestrate endpoints from backend/app/api/admin.py.
- Register orchestrate router in main.py under /api/orchestrate.
- Update frontend api/index.ts to call /orchestrate/{step} instead of
/admin/orchestrate/{step}.
This lets family members trigger vote emails without an admin bearer token.
- /api/admin/test-email now calls get_email_backend().send() instead of only logging.
- /api/meals/vote/{id} GET now queries MealPlanVote and renders 'already voted' confirmation if found.
- api/meals.py: fix remaining uppercase MealPlanItemStatus enum ref (DENIED, APPROVED, PENDING).
- Fixes the 'all meals show as pending' status regression and the 'Error: Already voted' bug.
POST/PATCH validate every ingredient_id against the ingredient table
and return 422 with the missing list when refs don't resolve. Replaces
the prior recipes.py stub. Public read routes + admin write routes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Backend (FastAPI):
- docker-compose with all 4 services
- FastAPI app with health endpoints
- SQLAlchemy models for all tables
- Placeholder API endpoints for all routes
- Config and database modules
- requirements.txt with all dependencies
Frontend (React):
- package.json with React, Tailwind, React Query, React Router
- Vite config with API proxy
- Tailwind and TypeScript configs
- Basic App with routing skeleton
- Placeholder pages (Dashboard, MealDetail, Pantry)
Infrastructure:
- nginx config for reverse proxy
- Dockerfile for backend and frontend
admin
MealPlanner