# Dyolink — Agent guide

This file orients Cursor agents at the start of a **new chat**. Project conventions live in **`.cursor/rules/`** (auto-loaded). Workflow playbooks live in **`.cursor/skills/`**.

## What Dyolink is

Dental clinic ↔ lab platform (monorepo):

| Path | Stack |
|------|--------|
| `backend/` | NestJS, Prisma, PostgreSQL |
| `frontend/` | Next.js 16, React 19, next-intl, Tailwind |
| `infrastructure/` | Docker, nginx, deploy scripts |

**Organization types:** `CLINIC` (patients, appointments, treatment) and `LAB` (cases, tasks). Many features are org-type-specific. Permissions use `TAB_*_READ` / `TAB_*_EDIT` codes — see `backend/src/common/permissions.ts`.

## Before you code

1. **Read applicable rules** in `.cursor/rules/` (especially `dyolink-overview` and the file-scoped rule for the area you touch).
2. **Match existing patterns** in the nearest feature folder — do not invent parallel structures.
3. **Keep diffs small** — one concern per change unless the user asks for a refactor.
4. **Verify:** `npm run build` (backend) and `npx tsc --noEmit` (frontend) when you change types or cross-cutting code.

## Frontend layout (critical)

```
frontend/src/
  app/                          → thin page.tsx only; compose from ui/
  components/
    ui/shared/                  → cross-feature UI (Button, Sidebar, …)
    ui/{feature}/               → feature UI (+ {Feature}Page.tsx for route logic)
    shared/                     → cross-feature non-UI (formatApiError, permissions, …)
    {feature}/                  → feature non-UI (helpers, config, pure functions)
  lib/                          → api clients, hooks
  types/                        → shared TS types
  messages/{en,fa,nl}.json      → all user-facing strings
```

**Example thin page:** `app/.../treatment/page.tsx` → imports `TreatmentWorkspace` from `components/ui/treatment/`.

**Treatment tab:** Preview and editable form are **separate** until the user clicks **Load into workspace** on a history item. See `.cursor/skills/treatment-workspace/SKILL.md` before changing that flow.

**Treatment lab rules (quick ref):**
- Lab-dependent details (e.g. prosthesis) **without teeth** can save but **cannot ship** — show `LabShipmentBlockedNotice` + inline banner; toast on dispatch add.
- **History filters** are client-side only (`treatmentHistoryFilters.ts`): “Not shipped to lab” + single date on already-fetched patient history; includes live current draft when filtering.
- **Lab shipments rail**: unified list with scope toggle **This patient** vs **All updates** (unread across org, includes patient name).
- **Unread semantics**: Treatment tab badge = count of unread cases (per-case read cursor) and clears when a case is opened/marked read (not on tab visit).
- **Lab shipment progress + comments**: shown in **Lab dispatch panel** for the active shipment; expanding activity / opening comments marks that case read.

**Lab Tasks tab:** Newest case first; steps ordered 1→N; case grouping when sorted by date; `stepCompleted` filter; prosthesis colors from `PROSTHESIS_TYPE_COLORS` via catalog; task assignment in **Cases** (compact row: status + assignee + last update); on **Tasks**, all staff see every task but only assignee (or unassigned pool) can change status — others see “Assigned to {name}” instead of the status dropdown; **case due dates** set/edited in clinic Treatment lab dispatch, shown on lab Cases/Tasks with overdue filter + sort; **mobile:** larger task status controls, sticky case header when grouped; **tab badges:** `LabCaseActivity` + `GET /notifications/tab-counts` (lab Cases/Tasks split, clinic Treatment) — see `.cursor/skills/lab-tasks/SKILL.md` and `.cursor/skills/lab-notifications/SKILL.md`.

## Backend layout

```
backend/src/
  modules/{feature}/            → controller, service, dto, module
  common/                       → guards, permissions, errors, utils
  prisma/                       → schema, migrations, seed
```

Errors: `AppException` + `ErrorCode` → frontend `getUserFacingError()`. Never throw raw strings for user-facing failures.

## Git & commits

- **Do not commit or push** unless the user explicitly asks.
- **Do not** amend commits, force-push, or skip hooks unless explicitly requested.

## Skills (workflows)

| Skill | When to use |
|-------|-------------|
| `.cursor/skills/add-feature/` | New tab, API module, or end-to-end feature |
| `.cursor/skills/treatment-workspace/` | Treatment tab: preview vs form, history, load flow, drafts |
| `.cursor/skills/lab-tasks/` | Lab Tasks tab: sort, case grouping, step-completed filter, prosthesis colors |
| `.cursor/skills/lab-notifications/` | Tab badges: LabCaseActivity, tab-counts API, read cursors |
| `.cursor/skills/frontend-structure/` | Moving components, auditing folder layout |
| `.cursor/skills/api-errors/` | New backend errors + frontend translations |

## Subagents (Task tool)

Use subagents to **save context**, not to avoid work:

| Type | Use for |
|------|---------|
| `explore` | Broad codebase search, unfamiliar areas |
| `shell` | Git, npm, long command sequences |
| `generalPurpose` | Multi-step research when parent context is large |

Do **not** delegate the user's main task to a subagent and return its summary — implement in the parent unless the user asked for exploration only.

## Improving this setup

When you and the user agree on a new convention, **add or update a rule** in `.cursor/rules/` (keep each rule under ~50 lines, one topic). For multi-step workflows, extend `.cursor/skills/`.

**To save a convention mid-task**, say: *"Remember this"* or *"Add to project rules"* — the agent uses the `capture-convention` skill and updates the repo (commit with your code).

| You say | Agent does |
|---------|------------|
| "Remember this: …" | Updates the right `.mdc` rule or skill |
| "Add a skill for …" | Creates `.cursor/skills/{name}/SKILL.md` |
| "This rule is wrong" | Edits the rule file; you commit |

Rules/skills **load automatically** in new chats; they do **not** update themselves unless you ask.