Christian Hood
5c5c777837
Uses Anthropic claude-sonnet-4-6 server-side to explain the semantic meaning of code diffs in the budget app domain (paychecks, bills, financing, actuals). Input validation rejects empty or oversized (>50KB) diffs. Tests mock the Anthropic client via direct method replacement (same pattern as db.pool.query). Nightshift-Task: semantic-diff Nightshift-Ref: https://github.com/marcus/nightshift
99 lines
6.9 KiB
Markdown
99 lines
6.9 KiB
Markdown
# CLAUDE.md
|
|
|
|
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
|
|
|
|
## Agent Workflow Rules
|
|
|
|
- **Commit after every task**: When a task is complete, stage all changed files and create a git commit before marking the task done.
|
|
- **Write tests with features**: New features and bug fixes must include unit tests. Run `npm test` in both `server/` and `client/` before committing.
|
|
- **Keep documentation current**: Update `CLAUDE.md` after every task that adds, changes, or removes a feature, API endpoint, or architectural pattern. This is mandatory, not optional. Update `PRD.md` only if scope/design decisions changed.
|
|
- **Mark tasks in td**: `td start <id>` when beginning, `td close <id>` when done.
|
|
- **Run td commands one at a time**: Never chain `td` commands with `&&` or `;`. Each `td` call must be its own separate shell invocation.
|
|
- **Only the orchestrator touches td**: Sub-agents must never call `td` commands directly. Concurrent `td` writes from parallel agents corrupt the SQLite database. The orchestrator handles all `td start`/`td close` calls before and after delegating to sub-agents.
|
|
- **td approve requires a separate session**: `td approve` cannot be run by the implementing agent or any sub-agent spawned from the same session — `td` tracks session involvement. Approval must come from a new conversation/session. Leave implemented tasks in review state (`td review <id>`) at the end of the session; they can be approved at the start of a fresh session.
|
|
|
|
## Task Management
|
|
|
|
This project uses `td` (a local CLI) for task tracking. At the start of each session:
|
|
|
|
```bash
|
|
td usage --new-session # required at conversation start or after /clear
|
|
td usage -q # quick check for subsequent reads
|
|
```
|
|
|
|
Optional session labeling:
|
|
```bash
|
|
td session "name" # label the current session
|
|
td session --new # force a new session in the same terminal context
|
|
```
|
|
|
|
Task state is stored in `.todos/issues.db` (SQLite).
|
|
|
|
## Development
|
|
|
|
**Run production stack (Docker):**
|
|
```bash
|
|
docker compose up
|
|
```
|
|
|
|
**Run development stack with live reload (Docker):**
|
|
```bash
|
|
docker compose -f docker-compose.yml -f docker-compose.dev.yml up
|
|
```
|
|
|
|
**Frontend only (Vite dev server):**
|
|
```bash
|
|
cd client && npm install && npm run dev
|
|
```
|
|
|
|
**Backend only (nodemon):**
|
|
```bash
|
|
cd server && npm install && npm run dev
|
|
```
|
|
|
|
## Testing
|
|
|
|
Unit tests are required when adding or modifying features. Both server and client use [Vitest](https://vitest.dev/).
|
|
|
|
**Run all tests:**
|
|
```bash
|
|
cd server && npm test # server unit tests
|
|
cd client && npm test # client unit tests
|
|
```
|
|
|
|
**Watch mode (re-runs on file change):**
|
|
```bash
|
|
cd server && npm run test:watch
|
|
cd client && npm run test:watch
|
|
```
|
|
|
|
**Server tests** (`server/src/__tests__/`): Use Vitest + [Supertest](https://github.com/ladakh/supertest) for route testing. The CJS server code requires mocking `db.pool.query` directly (replace the method on the shared pool object) rather than using `vi.mock` for CJS modules. Validation and pure logic functions are exported and tested directly. See `bills.validation.test.js` and `bills.routes.test.js` for patterns.
|
|
|
|
**Client tests** (`client/src/__tests__/`): Use Vitest + [React Testing Library](https://testing-library.com/docs/react-testing-library/intro/). jsdom environment is configured via `client/vitest.config.js`. The test setup file (`client/src/test/setup.js`) provides `@testing-library/jest-dom` matchers and polyfills like `window.matchMedia`. See `ThemeContext.test.jsx` and `App.test.jsx` for patterns.
|
|
|
|
**When adding features:**
|
|
- Add unit tests for new validation logic, utility functions, and API routes
|
|
- Add component tests for new React components or significant UI changes
|
|
- Export pure functions (validators, formatters, etc.) for direct testing
|
|
- Run `npm test` in both `server/` and `client/` before committing
|
|
|
|
## Application Structure
|
|
|
|
The default route `/` renders the paycheck-centric main view (`client/src/pages/PaycheckView.jsx`). It shows the current month's two paychecks side-by-side with bills, paid status, one-time expenses, and remaining balance. Month navigation (prev/next) fetches data via `GET /api/paychecks?year=&month=`.
|
|
|
|
**Theming:** `client/src/ThemeContext.jsx` provides light/dark mode via CSS custom properties on `[data-theme]`. Preference persists in `localStorage` and defaults to `prefers-color-scheme`. All design tokens live in `client/src/index.css`.
|
|
|
|
**Charts:** Monthly Summary and Annual Overview use [Recharts](https://recharts.org) (SVG-based). Monthly Summary shows a spending breakdown donut and variable-by-category bar. Annual Overview shows income vs. spending, surplus/deficit trend, and stacked variable spending by category — all driven by the single `GET /api/summary/annual?year=` endpoint.
|
|
|
|
**Bill amount locking:** When a `paycheck_bill` is marked paid, `amount_override` is set to the bill's current amount, locking in the historical value. Unmarking clears the override.
|
|
|
|
**Variable amount bills:** Bills with `variable_amount = true` require the amount to be entered each month in the paycheck view (stored as `amount_override` on `paycheck_bills`). The bill's `amount` field serves as an optional typical/estimated value.
|
|
|
|
**Lazy paycheck generation:** `GET /api/paychecks` returns virtual (unsaved) data with `id: null` when no DB record exists for the month. Paychecks are only persisted when the first interaction occurs (bill toggle, expense add, etc.). The "↺ Refresh amounts" button on the paycheck view re-runs `POST /api/paychecks/generate` to sync gross/net from current Settings. Individual paycheck gross/net can also be edited inline via the pencil icon.
|
|
|
|
**Financing:** `GET/POST /api/financing`, `PUT/DELETE /api/financing/:id`, `PATCH /api/financing-payments/:id/paid`. Plans track a total amount, payoff due date, and `start_date`. Payment per period is auto-calculated as `(remaining balance) / (remaining periods)`. Split plans (`assigned_paycheck = null`) divide each period's payment across both paychecks. Plans auto-close when fully paid. Financing payments are included in the paycheck remaining balance. `start_date` prevents a plan from appearing on paycheck months before it was created — both virtual previews and `generate` respect this guard.
|
|
|
|
**Migrations:** SQL files in `db/migrations/` are applied in filename order on server startup. Add new migrations as `00N_description.sql` — they run once and are tracked in the `migrations` table.
|
|
|
|
**Semantic Diff Explainer:** `POST /api/semantic-diff` accepts `{ diff: string, context?: string }` and returns `{ explanation: string }`. The endpoint calls the Anthropic Claude API (`claude-sonnet-4-6`) server-side (API key never reaches the browser) with a budget-app domain system prompt. Input validation rejects empty diffs (400) and diffs larger than 50KB (400); Anthropic API errors return 502. Requires `ANTHROPIC_API_KEY` in the server environment. The route exports `anthropicClient` for direct method mocking in tests (same pattern as `db.pool.query`).
|