Customer and Company used to be two side-by-side DataTable columns,
which cost ~120 px of horizontal width and made rows feel sparse when
the operator is just eyeballing who's-who across many quotes.
Now the Customer column carries both names in one cell: customer in
the default font-medium row, company underneath in text-xs
text-muted-foreground so the visual hierarchy reads as
"Primary person > Their organisation".
Effect on row layout:
- Total table column count drops from 7 to 6.
- Row width shrinks on desktop, freeing space for the larger
Quote # column (which carries the follow-up / replied / aged
chips and was getting crowded).
- The mobile card render (used below the sm breakpoint) gets the
same stacking so the company line shows on every viewport.
Edge cases handled:
- Both names empty → '—' muted placeholder
- Customer present, no company → just the customer name
- Company present, no customer → '—' for the customer, company below
- Very long names → truncate via className='truncate'
The existing standalone Company column is removed; downstream
code (DataTable, mobileCardRender) reads from the same expanded
fields so no extra fetches.
quotes/page.tsx
- Reduce inter-tab gap from gap-1 (4px) to gap-0.5 (2px) so the
pill strip reads as one cohesive filter row instead of a row of
loose badges.
- Drop `flex flex-col sm:flex-row items-start sm:items-center` on
the page header in favour of `flex flex-row flex-wrap
items-center`. On mobile widths (375px tested), the Approvals
and "+" buttons now stay inline with the "Quotes" title instead
of dropping below it.
next.config.ts
- Add 192.168.1.42 to allowedDevOrigins. The dev server was
warning when the user hit the app from a LAN host (their iPad,
visible in ps as `192.168.1.42:newoak->ipad.home:55199`). Now
whitelisted explicitly alongside the existing 192.168.1.122 /
192.168.1.128 entries.
On mobile widths (verified at 375px viewport), the 9 status tabs
wrap to 4 rows but the TabsList DOM reported height = 32px (one
row), so space-y-6 placed the search bar at y=254 — directly
overlapping tab rows 3 and 4.
Root cause: the @base-ui/react/tabs Root has `flex gap-2
data-horizontal:flex-col` baked in. When the wrapped TabsList's
intrinsic height (4 rows of 28px = 112px+) didn't propagate to the
flex-column parent, the next sibling was positioned based on the
shorter (single-row) computed height.
Fix: add `min-h-fit` to the TabsList override so the flex item
forces its content height regardless of the column's intrinsic
shrinkage. Verified at 375px width: tabs now correctly occupy
y=198-338 (4 wrapped rows), search bar at y=362 — clear 24px gap
with no overlap.
Also tightened the override:
- `inline-flex` -> `flex` (so the container is a full-width block
flex item rather than inline fit-content)
- `items-center` -> `items-start` (so wrapped children don't
cluster in the middle of an under-sized box)
- added `w-full` and `min-h-fit` to lock the height floor
`pb-4` from the earlier 'some space between tabs and search' tweak
is kept — now correctly visible as 16px additional breathing room
above the search bar's 40px input.
Scoped to the Quotes page only (one-line override) so other pages
that use the default TabsList (login grid, pricelist, customers)
are unaffected.
Two small visual tweaks on the Quotes screen:
1. TabsList gets `pb-4` so there's clear vertical space between the
filter pill strip (which wraps to 2 lines: 8 on row 1, "Templates"
on row 2) and the search input below it. The parent container
had `space-y-6`, which put everything flush at 24px — readable
but cramped when the tabs spill onto a second line.
2. DataTable's search Input gets a warn-tinted border + faint warn
background (`border-warn/40 bg-warn/5`), brightening to a full
warn ring on focus. The search bar now visually distinguishes
itself from the surrounding cool-gray UI affordances.
Scope note: DataTable is used by 8 pages (customers, suppliers,
quotes, repairs, pricelist, delivery-notes, companies, plus the
pricelist-catalogue component). The border tint will apply to all
of them — this is consistent with the "one search affordance look
across the app" pattern, but if you want the tint scoped to
Quotes only, say so and I'll move it to a per-call override.
The Quotes page filter strip showed 9 tabs with 'Templates 0' pushed
far to the right with a big visible gap — caused by `flex-1` on every
TabsTrigger combined with `inline-flex w-fit` on the parent
TabsList. `flex-1` forced every trigger into an equal-width cell,
so the narrower 'Templates' tab left a wide empty right margin.
Tested: min/max width across the 9 tabs dropped from a 1:1 equal
distribution (every cell = container-width/9 ≈ 130px) to a content-
proportional 70–182px range; the strip wraps tightly on one row
with 'Templates' falling to line 2 because there isn't room. The
visible gap is gone.
Other consumers of TabsList (`customers/[id]`, `pricelist`,
`login`) are unaffected:
- pricelist has 2 tabs, content-sized is the same or better.
- customers/[id] has 3 tabs — equal-width was visually fine,
content-sized still reads cleanly.
- login uses `grid w-full grid-cols-3`, so flex-1 on the children
had no effect to begin with.
Also bumped horizontal padding from `px-1.5` to `px-3` for
breathing room now that pills aren't width-stretched.
The form's deprecated generateQuoteNumber() was swapped for
fetch('/api/v1/quotes/next-number'). Three tests' setups pointed
fetchMock only at the upload response — and one asserted
fetchMock was never called (it now always is, for next-number).
- 'uploads per-item files' + 'shows a partial failure': chain two
mockResolvedValueOnce responses (next-number, then upload).
- 'allows a company-only selection': one mockResolvedValueOnce for
next-number.
- 'stops later uploads when a quote item create fails': add the
next-number mock and replace 'not.toHaveBeenCalled()' with
'toHaveBeenCalledTimes(1)' plus an explicit assertion on the
URL — item-create branch throws before the upload fetch, so
the original negative assertion was always-true and is now wrong.
The generateQuoteNumberMock in the hoisted block is now dead (the
form no longer imports it). Left in place to keep this commit
narrow; safe to delete in a future PR alongside the deprecated
function in src/lib/pb-helpers.ts (which now has zero callers).
src/app/api/v1/quotes/next-number/route.ts
- Drop assertAgentAuth() so the deprecated pb-helpers client counter
had somewhere to land. The route is intentionally open (see the
long doc comment); other /api/v1/* routes still gate on the bearer
token via their own call to the helper. Browser cannot send
AGENT_API_KEY without exposing the secret, and exposing it via
NEXT_PUBLIC_ would defeat the rest of the agent surface.
src/components/quote-form.tsx
- Replace `await generateQuoteNumber()` (the deprecated
read-then-write client counter from src/lib/pb-helpers.ts) with
a `fetch('/api/v1/quotes/next-number')`. Falls through to the
existing Math.random fallback if the endpoint errors. Removes
the race where two concurrent form submits could both pick the
same next number, since the server endpoint uses an atomic
quote_counters row under admin auth.
- Drop `generateQuoteNumber` from the @/lib/pb-helpers import.
Note: the deprecated `generateQuoteNumber` function remains in
src/lib/pb-helpers.ts — it has zero callers now. Safe to delete in
a future commit (covered by the data-layer-discipline section in
AGENTS.md).
src/app/api/v1/sprint/attachment-counts/route.ts
- Chunk itemIds into batches of 80 before building the PocketBase filter;
the live cap was hit at ~115 IDs × 15 chars and surfaced as a
generic Next 500 ("Something went wrong while processing your
request.") that hid the real cause. 7 chunks for the sprint
page's max of 500 items.
src/app/layout.tsx
- Add metadataBase derived from NEXT_PUBLIC_SITE_URL /
NEXT_PUBLIC_POCKETBASE_URL. Silences the per-page
"metadataBase not set" warning during dev renders and gives
OG / Twitter cards a stable canonical base.
src/app/api/v1/{quotes,order-items,purchase-orders,stock,
stock-purchases,users,sprint,driver,ai-insights}/route.ts (NEW)
- Add the missing top-level collection routes the agent API
needs. seven are straight passthroughs to their PB collection;
/sprint proxies to quote_items with optional claimStatus and
category query params; /driver lists PB users (no role field
in PB so callers narrow with ?filter=); /ai-insights returns
a roll-up count + most-recent regenerate timestamp across the
four collections that carry ai_* JSON fields.
src/app/api/v1/collection-pool/{route.ts,route.test.ts},
src/app/(app)/collection-pool/{page.tsx,page.test.tsx,
view-model.ts,view-model.test.ts,components.tsx} (DELETED)
- Drop the collection_pool feature entirely: PB has no
collection_pool collection on the live instance
(pb:migrate:orders-v2 was never run), the codebase comment
in purchase-orders-service.ts says "now-removed", and the
sidebar entrypoint has been migrated to /sprint + /deliveries.
The replacement operational status is the `item_status =
"collection_pool"` enum value on order_items, which stays.
src/lib/server/orders-service.ts (+ test), src/lib/pb-helpers.ts,
src/components/sidebar.tsx, src/components/ui/assign-driver-dialog.tsx,
src/app/(app)/collection-map/sidebar.tsx,
src/app/api/mcp/route.ts (+ test) (MODIFIED)
- Remove listCollectionPool + getCollectionPool +
updateCollectionPoolStatus, the /collection-pool sidebar entry,
the deep-link in assign-driver-dialog and collection-map
sidebar, and the collection_pool.list MCP tool (the tool,
handler, and test mock).
scripts/migrations/2026-07-26-fix-quote-counter-and-reassign-numbers.ts (NEW)
- One-shot migration script that:
1. Reads every quote, sorts by created ASC.
2. Reassigns clean Q202607001 … Q202607{N} over the whole
table (including the e+22 garbage and the legacy
Qnnnnnn rows).
3. Sets quote_counters.(year=2026, month=7).last_seq = N
so the next /api/v1/quotes/next-number returns a clean,
atomic, unique number.
Default mode = dry-run; pass --apply to write.
Already executed against the live cloud PB during this session
(verified: 200/200 rows updated, counter at 200, next-number
returns Q202607201, Q202607202, Q202607203 on consecutive calls).
AGENTS.md (MODIFIED)
- Add data-layer-discipline block matching the style of the
three existing blocks. Documents which layer is legacy (pb-helpers)
vs current (lib/server/*-service.ts via /api/v1/*), gives the
migration order (customers → orders → quotes → suppliers →
companies → rest), and includes the grep one-liner future AI
agents should run before extending pb-helpers.
Not in this commit (your WIP, untouched):
src/app/(app)/{dashboard,delivery-notes,layout,orders/[id],
quotes/[id],quotes/approvals,quotes,sprint}/page.tsx,
src/app/(auth)/login/page.tsx, src/components/charts/*,
src/components/{company-form,customer-form,quote-email-card,
sprint-item-pricing-modal,supplier-form,theme-bootstrap,
theme-picker}*.{tsx,test.tsx}, src/app/api/invoicing/delivery-note/*
src/app/api/v1/quotes/[id]/pdf/route.ts, src/app/globals.css,
src/lib/{item-status-ui,pdf/quote-document,quote-status*,themes,
types}.ts, src/lib/server/{order-item-status*,quotes-service}.ts,
src/styles/themes.css, APP_CAPABILITIES.md, README.md,
docs/SHIPPED.md, scripts/smoke/screenshot-pages.mjs,
and the new untracked files under docs/, scripts/migrations/
2026-07-19…25, src/app/{(app)/deliveries,(app)/delivery-notes,
api/v1/ai-insights/rerun,api/v1/deliveries,api/v1/delivery-notes,
api/v1/quotes/[id]/partial-approve,api/v1/quotes/[id]/split},
src/components/{ai-insights-panel,ai-insights-renderer,
ai-item-history-trigger,brand-stripe,delivery-notes,ui/stock-item-search,
ui/duplicate-warning,ui/duplicate-warning.test}.tsx, supplier-form.test.tsx,
src/lib/{pb-helpers-duplicate.test,server/deliveries-service*,
server/delivery-notes-service*}, public/wfdash-logo*.png,
opencode.jsonc.
Add complete stock purchase order workflow including:
- Status tracking with sending/mail_failed states and UI labels
- Server-side PDF rendering and document storage
- File proxy API support for stock PO attachments
- Send outcome webhook callback for email notifications
Add driver app support:
- Unified stop list API for both customer deliveries and stock pickups
- Status update endpoint for collected quantities and problem reporting
- Collection map page support for stock pickup locations
Include matching PocketBase migrations, type definitions, and stock inventory fulfillment logic.
Add full integration for stock supplier pickup lines into the collection pool workflow:
- Add two idempotent database migrations for stock line drv_id and state tracking
- Update stock purchase number format to POyyMMddNNSTK convention
- Extend collection pool to load both customer order items and stock purchase items
- Add stock-specific UI, driver assignment flow, and mark collected action
- Update type definitions and service logic for stock purchases and line items
- Add comprehensive test coverage for all new workflows
This commit adds the complete v2/v3 stock management functionality:
- add new stock-specific item statuses and update the status visual map
- implement full stock purchase API suite (create, list, retrieve, send, receive)
- reorganize sidebar navigation to group stock-related pages together
- add stock item CRUD APIs, status utilities, low stock filtering, and surplus booking logic
- add database migration for the `stock_items.target_qty` field
- update type definitions and add comprehensive tests for all new stock components
This commit adds the full end-to-end delivery note workflow for the invoicing system:
- Added `letterheadForCompany` utility to infer the correct letterhead from a company name
- Created the DeliveryNoteDocument PDF component and server-side render function
- Added API endpoints for delivery note creation, PDF retrieval, printing, and print tracking
- Updated the PickingSlip component to support both invoice and delivery note headers
- Added view models for grouping, filtering, and displaying delivery note items
- Updated AGENTS.md with PocketBase schema and attachment best practices for new workflows
- Added comprehensive test coverage for all new components and API routes
- Lift customer query state to parent page to maintain consistent cache keys for QB customer lookups
- Restructure SingleOrderQbModal UI into grouped sections for better organization
- Update modal footer padding and add mobile-responsive layout
- Revise line item input labels and add accessible aria labels matching test selectors
- Refactor QB customer lookup effects and callbacks to support edited search queries
- Fix test selectors for invoice quantity and price fields to match updated labels
- add `costPrice` field to invoicing row types, plus ZAR currency formatting and profit margin calculations
- update invoice row displays to show ordered/received quantity, unit cost, selling price and margin metrics
- add per-row selection in QuickBooks send modals, letting users filter which order lines to include on invoices
- update ModalShell component to support responsive breakpoint-based sizing instead of fixed sm-only widths
- refine UI styling across the invoicing page to use consistent theme tokens (bg-card, muted-foreground)
- update tests to use dynamic yesterday dates instead of hardcoded values for stable, time-agnostic assertions
- improve button states, labels and text for QuickBooks invoice actions, including showing selected item counts
Add full Tangerine brand theme with light and dark mode variants to the app theme system.
Add four new design and integration documents for collection pool, map and driver app workflows.
Add initial scaffolding for three new plugin source directories for the collection pool bulk assignment feature.
Both the per-row Assign Driver button and the bulk-assign button in each
supplier section header now disable when the target row(s) have not been
marked ready for collection. Tooltips explain the reason.
This enforces the rule that a stop must be ready before it can be
assigned to a driver — same gate for single-row and bulk flows.
The supplier name is already shown in the SupplierGroup section header,
so repeating it on each row inside the section adds visual weight
without information. Keep it on the checkbox aria-label for screen
readers only.
- selectedIds Set state plus toggleRowSelected / toggleSectionSelected callbacks
- PoolRow signature extended with selected + onToggleSelected
- New flex wrapper around supplier/description block with leading checkbox input
- bg-primary/5 row tint applied when selected
- aria-label composes supplier + description for screen readers
- Standardize quote number format from Q-YYYY-MM-NNN to QYYYYMMNNN across all client, server, test and API code
- Add atomic backfill logic for quote counters to avoid sequence collisions with existing records
- Rearrange Delivery Notes and Repairs to the Operations sidebar section
- Add supplier count badge to pricelist catalogue and improve filter tab overflow styling
- Optimize font loading in root layout to skip preloading unused font weights
- Improve table styling for customer and company pages with better text wrapping and empty states
- Fix rep display logic in quotes page to show "Unassigned" when no rep is assigned
- Add supporting documentation, PocketBase counter schemas and setup migration scripts
- Remove embedded SupplierMap from suppliers page for cleaner layout
- Improve invoice card layout for company and customer info spacing
- Improve filter label styling and accessibility in collection map page
- Add Supplier Map component to suppliers page and reorganize sidebar navigation
- Implement picking slip PDF generation with supplier per line item and picked-by signature footer
- Add invoice bulk print tracking API and view models to track printed invoice state
- Extend company search and form to support email/telephone fields and improve searchability
- Fix data table debounce race condition by using useRef for onSearch handler
- Fix supplier page fetch race condition with debounced timeout
- Reduce list page item counts to 10 per page for customers, companies, and suppliers
- Update company search filter to include email and telephone in search queries
- Add comprehensive test suite for new and existing components and APIs
- Add pdf-lib dependency for PDF generation and update related tooling
- Clean up unused code and fix minor linting issues
Add complete end-to-end QuickBooks invoicing workflow for the invoicing page:
- Add `jspdf` dependency for client-side PDF generation
- Implement updated item status handling with fallback for new invoice statuses
- Add server-side proxy API for secure QuickBooks customer lookup
- Build utilities to combine QuickBooks invoices and picking slips into a single PDF
- Add invoicing view models for row grouping, payload construction, and status checks
- Add full test coverage for all new components, APIs, and utilities
- Include feature design documentation and sample n8n workflow config
Implement the complete collection confirm feature set:
- Add new `/collection-confirm` operational page for reviewing and confirming collected order items
- Add sidebar navigation entries for Collection Confirm and Invoicing
- Add `order_items` to the file proxy API whitelist for attachment access
- Install required testing dependencies and configure Vitest test setup
- Add full design documentation, test suite, and page overview screenshot
- Update package.json and build configuration files
Add optional nearby-supplier cluster sort mode to the collection pool table, keeping operational sort as the default. Also includes:
- Update UserSearch component to support fallback selected label for pre-selected values
- Refactor AssignDriverDialog into a separate inner component with intent and pre-filled driver label support
- Add assignment badge helper types and functions
- Add full test coverage for new sort logic and component behaviors
- Update implementation plans and documentation files
- Fix minor code consistency and lint issues across touched files
Remove redundant Assign Drivers top-level navigation link since driver assignment is now embedded directly in the collection pool page per-row, and add an Other nav section for leftover navigation items
Extract collection pool view logic into a dedicated view-model module with full Vitest test coverage for assignment state classification, row sorting, pool age labeling, and proximity marker grouping
Flatten the collection pool page into a single unified list replacing the old grouped assigned/unassigned layout, with updated row UI showing all required operational details and context-aware actions
Add database migration to add optional `collection_photo` field to order_items for driver pickup proof photos
Add complete design documentation, planning artifacts, and brainstorming session files for the collection pool redesign project
overhaul the collection pool page's ready status UI: split the clickable chip into a static status indicator and dedicated toggle button with loading state, improving user clarity between status and actionable elements
update all related copy (toast messages, filter labels, button text) to use consistent "awaiting supplier confirmation" terminology across the interface
remove invalid `transparent` prop from traffic tile layers in both the collection map and traffic tiles components, update comments to reflect Leaflet does not support this option
add wildcard to allowedDevOrigins in next.config.ts to simplify local development testing across different network interfaces
add a comprehensive database backfill script to fix assigned order items missing the `ready_for_collection` flag, ensuring drivers see their correct assigned stops in the driver app
Overhaul the app's color system by replacing all hardcoded Tailwind color classes with semantic theme variables to ensure consistent theming across all components, pages, and tests.
Additional changes:
- Add assigned date support for order item status updates: add `assignedDate` field to the order item status PATCH API, implement server-side normalization/validation, and add corresponding tests
- Refactor purchase order PDF workflows to use the consolidated `purchase_orders.documents` field instead of the legacy separate `purchase_order_pdfs` collection
- Fix quote number parsing and formatting to support both dashed (Q-YYYY-MM-NNN) and dashless (QYYYYMMNNN) formats
- Rename `driver_pool` item status to `collection_pool` across the codebase
- Add null handling for order item quantities and pricing to display "—" for missing values
- Improve filter tab styling: make mobile tabs scrollable, prevent text wrapping, hide scrollbar
- Add new utility scripts, API routes, migrations, and polish small UI/UX fixes including tsconfig updates and theme hook improvements
capture the thrown error, extract its message if available, and append it to the base "PB admin unavailable" error response. fall back to a generic unknown error message for non-Error failures, revealing actionable debugging details like missing env vars or invalid PB credentials.
Add full sprint pricing workflow and supporting infrastructure:
- New sidebar navigation entry for the sprint dashboard
- PocketBase category collection schema and migration script
- Server API routes for sprint item claims, attachments, counts, and file proxying
- Dev identity utilities and UI switcher for multi-user testing
- Reusable FilterTabs component replacing ItemStatusTabs
- Updated TypeScript types for sprint-related entities
Also include:
- Fix PocketBase file URL method casing (getUrl → getURL) across codebase
- Add development utility, smoke test, and helper scripts
- Update package.json with new sprint migration command
This commit adds the full end-to-end SKU pricing match flow for quote items:
- Add new QuoteItem and PriceListRow fields for SKU and pricing metadata
- Add sku_normalised to price_list for robust format-agnostic SKU matching
- Refactor price list search to use modular search builder and filter helpers
- Add database migrations, backfill scripts, and new package commands
- Update catalogue search UI and fix quote list accidental click behavior
- Add comprehensive unit, smoke, and end-to-end test suites
- Update documentation for the new price search and extraction logic
The Catalogue tab's search box used to OR across just sku,
description, and brand. Now it hits:
- sku, description, brand, supplier, category, subcategory
(PB's case-insensitive `~` operator, each indexed)
- search_text (a denormalised text blob)
The denormalised column is the key part: PB can't substring-
match a JSON field, so without this, synonyms from
`specifications.synonyms` ("slugger cutter", "turbo tough",
etc.) would never match in the search box. The n8n workflow's
Code node now flattens every searchable bit — including the
synonyms array — into search_text on every POST. The 46 pre-
existing rows were backfilled once with the same formula.
The `price_list.supplier` field also stored the PB id on the
46 pre-existing rows (n8n was setting the resolved name, but
the original LLM tests that seeded the table didn't). Fixed
that in the same backfill — supplier now reads 'JEISA -
WESTSCOTT GROUP' instead of 'q6hpwbxsicyoiuu', which makes
the supplier filter dropdown work AND makes
'supplier ~ "jeisa"' match.
Migration `2026-06-30-add-price-list-search-text`:
- Adds `search_text` (text) column
- Adds `idx_pl_search_text` non-unique index
- Idempotent — re-runs are safe and self-heal the index
Updates:
- `getPriceLists` helper ORs across all 6 columns + search_text
- `PriceList` TS type gains `search_text: string`
- `pricelist-extraction.md` Code node snippet shows how the
n8n workflow should populate `search_text` on every POST
- Search input placeholder reads 'Search anything — SKU,
description, brand, supplier, synonyms…'
Tests: 229/230 (same pre-existing failure, unrelated).
TypeScript clean.
Adds a second tab on /pricelist that shows the structured
`price_list` collection (AI-extracted line items) instead of the
raw PDF uploads. Operators can now:
- Search across SKU, description, and brand
- Filter by category (twist_drill, hss_cutter, countersink, ...)
- Filter by brand (JEI, TurboSteel, TurboTough, ...)
- See cost vs selling price side-by-side in ZAR
- Click a row's 'Source' link to jump back to the raw PDF
The distinct filter values are pulled from a one-shot getList(1, 500)
on mount, with hardcoded fallbacks (FALLBACK_CATEGORIES /
FALLBACK_BRANDS) so the dropdowns are useful even before any
extractions have run.
Layout: Tabs (Raw Uploads | Catalogue) wrap both views; the
header + page chrome stay shared. The DeleteDialog stays outside
the Tabs since it's a modal.
Also ships an idempotent migration to add `source_file` (text)
to `price_list` — used by the n8n workflow to track which raw
PDF each extracted row came from, and by the verify-after-POST
step the workflow uses to decide whether to flip
`raw_price_list.status` to Completed vs Failed.
Price type is in PriceListRow (shared subset) and PriceList
(full record). Supplier names are resolved client-side via
getAllSuppliers() since the live schema has supplier as a text
field rather than a relation.
Tests: 229/230 (same pre-existing failure, unrelated). TypeScript
clean.
Operators can now type per-upload instructions for the AI extraction
step in the Pricelist upload form. The notes go into a new
`raw_price_list.notes` field that the n8n workflow reads and
injects into the LLM user message as per-upload context.
Common uses:
- "Brand is JEI" → AI stamps every item with brand="JEI"
- "Apply 30% markup" → Code node regexes this out for selling_price
- "All prices in ZAR excl. VAT" → AI uses ZAR
- "Page 4 has discount codes" → AI skips those rows
Also auto-populates `raw_price_list.file_name` (the original
browser-side filename) on upload — PB slugifies the upload to a
token stored inside supplier_file, so without this field the
operator saw the slugged version. Existing rows backfilled by the
admin script.
The `log_summary` field is now reserved for the AI's write-back
summary (set by the n8n workflow after extraction). It was
previously operator-editable on the form, which was wrong —
overwriting it clobbered the AI's output.
New migration: scripts/migrations/2026-06-30-add-pricelist-notes.ts
(idempotent, only adds `notes` if missing — `file_name` was
added manually via the PB admin UI before this commit).
doc/pricelist-extraction.md updated with the n8n LLM user message
shape, the markup regex, and the notes field's role in the pipeline.
Adds /pricelist under the Management section. Operators drag a supplier
price-list file (PDF / CSV / text / image) into the upload zone — or
use the explicit Choose file button — and it lands in PocketBase as an
attachment on the raw_price_list collection. The table below shows the
queue: file, supplier (relation to suppliers), status badge, log
summary preview, upload date, and a delete action per row.
Layout matches the rest of the app: Card for the upload zone, DataTable
for the queue, DeleteDialog for the destructive action. Live updates
via the PB subscribe channel so other tabs / the n8n ingest workflow
show new rows without a manual refresh.
PB schema is sourced from docs/pocketbase/raw_price_list.collection.json.
The migration script is idempotent: it creates the collection from the
schema file on a fresh deploy, and on a pre-existing one (the live PB
already had it, with PB's default null/superuser-only rules) it relaxes
the auth rules to @request.auth.id != '' so any logged-in user can
CRUD — matching every other collection in the app.
Field names match the live PB collection (discovered via admin API):
- supplier_file (the file)
- supplier_name (relation to suppliers)
- status (select: Pending / Processing / Completed / Failed)
- log_summary (free text)
The ApproveAndSendDialog had a useEffect whose deps array included
`openNonce` — a state counter the effect itself incremented:
useEffect(() => {
if (!open) return;
setOpenNonce((n) => n + 1); // updates openNonce
// ... seed defaults ...
}, [open, openNonce, ...]); // openNonce is in deps
This produced Maximum update depth exceeded: the effect fires → state
updates → deps change → effect re-fires → loop, until React pulls the
plug.
The intent was 'seed once per open cycle, not on every claimerName /
repEmail update' (so async lookups arriving after the modal opened
didn't clobber operator edits). The state-counter approach was the
wrong mechanism for that — mutating a dep from inside the effect
*is* the loop.
Fix: replace openNonce (state) with seededFor (ref). A ref mutation
doesn't trigger re-render, so the loop is impossible. We still depend
on [open, quote.id] in the effect, and the ref's purpose is purely
'has this (open, quote.id) cycle been seeded yet?'. Reset to null on
open=false so the next open seeds fresh.
Dropped claimerName / repEmail from the deps too — they were the
reason openNonce existed in the first place (don't re-seed when they
update), and the ref now enforces that without the side-effect trap.
Tests: 229/230 vitest (the 1 failure is the pre-existing attachments
off-by-one, unrelated to this change).
