# CLAUDE.md

This file is the primary context source for Claude Code in this repository. Read it fully before touching any code.

---

## Project Overview

**Personal Internet Cell (PIC)** is a self-hosted digital infrastructure platform for individuals who want full ownership of their core internet services without relying on cloud providers.

A PIC instance runs DNS, DHCP, NTP, WireGuard VPN, email (SMTP/IMAP), calendar/contacts (CalDAV/CardDAV), file storage (WebDAV), HTTPS reverse proxy (Caddy), an internal certificate authority, and optional third-party services — all managed from a single REST API and a React web UI. No manual config-file editing is required for normal operations.

**Primary users:** technically capable individuals, homelab operators, small families or teams.

**What the product optimizes for:**
- One-command install, browser-based first-run wizard, no manual `.env` editing for identity
- Everything managed through the API and UI — the user should never need to `ssh` for day-to-day operations
- Security by default: session auth, CSRF protection, WireGuard isolation, internal CA, no open API port
- Reliability and observability: structured logs, health monitoring, automated config backups

**Key constraints:**
- Runs on a single Linux host with Docker; no Kubernetes, no swarm
- Must work on Debian, Ubuntu, Fedora, RHEL, and Alpine
- The Flask API must never be exposed directly; Caddy always proxies it
- All secrets live in `data/` (git-ignored), never in the repo

---

## Tech Stack

### Backend
- **Python 3.11** — Flask REST API (`api/app.py`)
- **Flask** — routing, sessions, before-request hooks (enforce_setup, enforce_auth, check_csrf)
- **bcrypt** — password hashing in `AuthManager`
- **Docker SDK for Python** — container lifecycle in `ContainerManager`
- **PyNaCl / Age** — encryption in `VaultManager`
- **pyotp** — TOTP for DDNS registration

### Frontend
- **React 18** — SPA
- **Vite** — dev server and build (proxies `/api` → `:3000`)
- **Tailwind CSS** — all styling; no custom CSS files
- **Axios** — all API calls go through `src/services/api.js`

### Infrastructure
- **Docker Compose** — all 12+ service containers
- **Caddy** — reverse proxy, TLS termination (Let's Encrypt DNS-01 or HTTP-01 or internal CA)
- **CoreDNS** — `.cell` TLD authoritative DNS
- **dnsmasq** — DHCP
- **chrony** — NTP
- **WireGuard** — VPN (kernel module, not userspace)
- **Postfix + Dovecot** — email via `docker-mailserver`
- **Radicale** — CalDAV/CardDAV
- **PowerDNS** — authoritative DNS on the DDNS VPS (separate repo: `pic-ddns`)

### CI/CD
- **Gitea Actions** — unit tests on every push, image builds on tag
- **act_runner** — self-hosted runner on pic0 (192.168.31.51)
- **Gitea Container Registry** — images pushed to `git.pic.ngo`

Do not introduce: Redux, styled-components, SQLAlchemy, Celery, or any async framework (asyncio/FastAPI) into the main API unless explicitly requested.

---

## Architecture

```
Browser / WireGuard peer
  └── Caddy (:80/:443)           TLS termination, reverse proxy
        └── React SPA (:8081)    Vite + Tailwind (Nginx in container)
        └── Flask API (:3000)    REST API, bound to 127.0.0.1 only
              ├── NetworkManager       CoreDNS, dnsmasq, chrony
              ├── WireGuardManager     WireGuard peer lifecycle
              ├── PeerRegistry         peer registration and trust
              ├── EmailManager         Postfix + Dovecot
              ├── CalendarManager      Radicale CalDAV/CardDAV
              ├── FileManager          WebDAV + Filegator
              ├── RoutingManager       iptables NAT and routing
              ├── FirewallManager      iptables INPUT/FORWARD rules
              ├── VaultManager         internal CA, TLS certs, Age encryption
              ├── ContainerManager     Docker SDK
              ├── CellLinkManager      site-to-site WireGuard links
              ├── ConnectivityManager  per-peer exit routing (WG ext, OpenVPN, Tor)
              ├── DDNSManager          dynamic DNS heartbeat
              ├── ServiceStoreManager  optional service install/remove
              ├── CaddyManager         Caddyfile generation and reload
              ├── AuthManager          bcrypt passwords, session auth, RBAC
              └── SetupManager         first-run wizard state
```

### Key files

| File | Role |
|---|---|
| `api/app.py` | Flask app, all REST endpoints, before-request hooks, health monitor thread |
| `api/managers.py` | Singleton instantiation of all service managers |
| `api/base_service_manager.py` | Abstract base class: `get_status`, `get_config`, `update_config`, `validate_config`, `test_connectivity`, `get_logs`, `restart_service` |
| `api/config_manager.py` | Single source of truth for `cell_config.json` — all read/write goes through here |
| `api/service_bus.py` | Pub/sub event system between managers |
| `webui/src/services/api.js` | Axios API client — all UI→API calls |
| `docker-compose.yml` | Container definitions and network topology |
| `Makefile` | All operational commands |
| `install.sh` | Bash installer served via `https://install.pic.ngo` |

### Directory layout

```
api/           Flask API and all service managers
webui/         React SPA (Vite + Tailwind)
tests/         pytest unit tests (no running services required)
tests/integration/  require a running PIC stack
tests/e2e/     Playwright UI and WireGuard e2e tests
config/        Runtime config per service (mostly git-ignored)
data/          Runtime secrets and state (fully git-ignored)
scripts/       Setup and maintenance scripts
install.sh     One-line installer entry point
Makefile       All make targets
docker-compose.yml
```

### Config and secrets

- Runtime config: `config/api/cell_config.json` — managed by `ConfigManager`, never edit directly
- Secrets and user data: `data/` — git-ignored, contains `auth_users.json`, WireGuard keys, DDNS token, CA key
- DDNS config lives under the top-level `ddns` key in `cell_config.json`, accessed via `config_manager.configs.get('ddns', {})`
- Do not read `_identity.domain` expecting a dict — it is a plain string (the domain mode, e.g. `"pic_ngo"`)

### Before-request hooks (app.py)

Three hooks run on every request in this order:
1. `enforce_setup` — returns 428 for all `/api/*` except `/api/setup/*` and `/health` until setup is complete. Skipped when `app.config['TESTING']` is True.
2. `enforce_auth` — returns 401 if no session; returns 503 if users file exists but is empty (misconfiguration). Skipped when `app.config['TESTING']` is True.
3. `check_csrf` — requires `X-CSRF-Token` header on all mutating requests except `/api/auth/*` and `/api/setup/*`.

---

## Coding Conventions

### Python (API)

- All managers inherit `BaseServiceManager` — always implement all abstract methods
- Use `self.logger` (from `BaseServiceManager`) — never `print()` or raw `logging`
- Config reads go through `self.config_manager` — never open `cell_config.json` directly
- Use `threading.RLock` for shared state; managers run in a multi-threaded Flask app
- Do not use `any` typing; be explicit
- Keep Flask route handlers thin — business logic belongs in the manager, not in `app.py`
- Error responses must be JSON: `jsonify({'error': '...'}), <status_code>`
- Do not catch bare `Exception` and silently swallow it — log at minimum

### JavaScript (webui)

- All API calls go through `src/services/api.js` — never use `fetch` or a new Axios instance directly
- Use functional components; no class components
- Tailwind utilities only — no inline styles, no custom CSS files
- Keep page components in `src/pages/`, reusable UI in `src/components/`
- State: local `useState`/`useEffect` is fine; no Redux or global state library

### General

- No comments that describe *what* the code does — only *why* if non-obvious
- No dead code, no commented-out blocks
- No backwards-compat shims for things being removed
- Prefer editing existing files over creating new ones
- Tests that write to disk: mock `builtins.open` with `OSError` rather than relying on `/nonexistent/path` (CI runs as root and can create any path)

---

## Testing and Quality

Before considering any task complete:
1. Run `make test` — all 1500+ unit tests must pass
2. Fix failures before committing — the pre-commit hook will block the commit anyway

### Rules

- Use `unittest.mock` / `pytest-mock` for all Docker, filesystem, and subprocess calls
- Tests must pass in CI (rootless environment where filesystem assumptions don't hold)
- When testing write-failure paths, mock `builtins.open` with `side_effect=OSError` — do not rely on unwritable paths
- Integration tests (`tests/integration/`) require a running stack — exclude from CI with `--ignore=tests/integration`
- E2e tests (`tests/e2e/`) require Playwright — exclude from CI with `--ignore=tests/e2e`
- Add tests for any new API endpoint, manager method, or utility function
- Do not add tests for Flask routing boilerplate or trivial getters — test behaviour, not structure

---

## File Placement Rules

| New thing | Where it goes |
|---|---|
| New service manager | `api/<name>_manager.py`, registered in `api/managers.py` and wired into `app.py` |
| New API endpoints | `app.py` — grouped with the relevant manager's existing endpoints |
| New React page | `webui/src/pages/` |
| Reusable UI component | `webui/src/components/` |
| New pytest test file | `tests/test_<module>.py` |
| Operational script | `scripts/` |
| Documentation | Update `README.md`, `QUICKSTART.md`, or `Personal Internet Cell – Project Wiki.md` as appropriate |

Do not create a new abstraction for a single use case. Do not create near-duplicate files — edit the existing one.

---

## Safety Rules

- **Never expose the Flask API port (3000) directly** — it must always be behind Caddy
- **Never commit secrets** — `data/`, `.env`, `*.key`, `*.pem` are all git-ignored; keep it that way
- **Do not modify `enforce_setup` or `enforce_auth` hooks** without understanding the full auth flow — these are the security boundary
- **Do not change the `cell_config.json` schema** without updating `ConfigManager` validation and all manager reads
- **Do not rename API route paths** without checking the webui `api.js` client and any external callers
- **Do not modify WireGuard key generation** — losing the server private key means all peers must be re-provisioned
- Flag any change to auth flow, CSRF logic, or session management as security-sensitive before implementing

---

## Commands

```bash
# Stack lifecycle (always use make — never call docker/docker-compose directly)
make start           # build and start all containers
make stop            # stop all containers
make restart         # restart containers
make status          # container status + API health check
make logs            # follow all container logs
make logs-api        # follow API logs only
make logs-caddy      # follow Caddy logs
make shell-api       # shell inside the API container
make build-api       # rebuild API image after code change
make build-webui     # rebuild webui image after code change

# Tests
make test            # pytest tests/ --ignore=tests/e2e --ignore=tests/integration
make test-coverage   # coverage report in htmlcov/
pytest tests/test_<module>.py -v   # single test file

# Local dev (no Docker)
pip install -r api/requirements.txt
python3 api/app.py              # Flask API on :3000

cd webui && npm install && npm run dev   # React UI on :5173 (proxies /api → :3000)

# Peer / WireGuard
make list-peers
make show-routes

# Admin password
make show-admin-password
make reset-admin-password

# Backup / restore
make backup
make restore

# Maintenance
make update          # git pull + rebuild + restart
make uninstall       # stop containers; prompt to delete config/ and data/
```

---

## Infrastructure Topology

| Machine | IP | Role |
|---|---|---|
| pic0 | 192.168.31.51 | Dev machine — you are here. Run all commands directly. |
| pic1 | 192.168.31.52 | Test/staging PIC instance |
| Gitea | 192.168.31.50 | Self-hosted git server (`gitea@192.168.31.50:roof/pic.git`) |
| DDNS VPS | 192.168.31.101 (LAN) / 178.168.15.65 (public) | PowerDNS + FastAPI for `*.pic.ngo` DDNS |

The `roof` user on pic0 has passwordless sudo and is in the `docker` group — use both freely.

---

## AI Collaboration Rules

These rules apply to every Claude Code session in this repo:

- **Read memory first** — load `/home/roof/.claude/projects/-home-roof/memory/MEMORY.md` at session start; follow referenced memory files for relevant context.
- **You are on pic0** — execute commands directly here; do not ask the user to run them.
- **`make` is the only container interface** — never call `docker` or `docker-compose` directly. All container lifecycle goes through `make start`, `make stop`, `make build`, `make logs`, etc.
- **Use specialized agents** — spawn `pic-remote` for VPS/pic1 SSH tasks, `pic-qa` for test writing, `pic-architect` for design decisions, `pic-designer` for UI review, `pic-devops` for docker-compose/Makefile changes, `pic-writer` for documentation.
- **Test before commit** — run `make test` and fix all failures before staging. The pre-commit hook enforces this, but run it manually first.
- **No skipping hooks** — never use `--no-verify` unless the only change is documentation or a workflow file with no Python/JS.
- **Commits need context** — write commit messages that explain *why*, not just *what*. Always add the Co-Authored-By trailer.