Files
BoltClam/docs/ARCHITECTURE.md
TheMajesticMagicianandClaude Code addd00364d BoltClam 1.0.0: ClamAV desktop protection with Bolt sign-in
Adds the PyQt6 Windows client (Homepage, Scan & results, About & updates, Settings), ClamAV scanning and quarantine, browser-download watch with an exclusive scan gate, Bolt loopback/PKCE sign-in with DPAPI key storage, Gitea release feed with checksum verification, opt-in diagnostics, and Windows/portable build tooling.

Co-Authored-By: Claude Code <noreply@anthropic.com>
2026-10-25 16:21:14 -06:00

108 lines
5.0 KiB
Markdown

# BoltClam architecture
## Process layout
BoltClam is one PyQt6 process on the user's Windows desktop. It starts no
Windows service and installs no driver. ClamAV remains an external program.
```
BoltClam.exe (PyInstaller or portable launcher)
└─ boltclam.app
├─ MainWindow ── HomePage / ScanPage / AboutPage / SettingsPage
├─ RealtimeProtection ── DownloadWatcher + QThreadPool scan tasks
├─ ScanWorker / DefinitionWorker (QThread)
├─ Storage (SQLite, WAL)
├─ QuarantineService
├─ ClamAVService → clamscan.exe / freshclam.exe
└─ ApiClient → https://api.themajesticnetwork.com
```
Settings are JSON under `%APPDATA%\BoltClam`; history and quarantine live
under `%LOCALAPPDATA%\BoltClam`. `BOLTCLAM_CONFIG_DIR` and
`BOLTCLAM_DATA_DIR` override both for tests and portable/development use.
## Page responsibilities
- **HomePage** calls `system_info.collect()` on a worker. PowerShell supplies
Defender/firewall/network facts on Windows; missing data is "not reported".
The shell gauge and concern list are derived from ClamAV state, definition
age, real-time state, firewall, last scan, active detections and quarantine
count. It never labels an unknown state as protected.
- **ScanPage** persists scope/auto-response preferences immediately, runs
`ScanWorker`, records each detection in SQLite before remediation, and
exposes manual quarantine/delete/ignore after a scan. Quarantine management
restores to the original path or a chosen path and refuses overwrites.
- **AboutPage** calls `/boltclam/account` for user/plan data and
`/boltclam/updates` for the release feed. Downloads use the API asset proxy
and optionally verify `SHA256SUMS.txt`.
- **SettingsPage** validates ClamAV paths and extra options, applies logging
changes, updates the real-time watcher, and creates/uploads diagnostics
bundles only when the user has opted in.
## Sign-in sequence
```
BoltClam Browser Bolt desk API
|-- PKCE verifier/challenge --| |
|-- loopback listener :port/cb| |
|-- open /desk/auth/boltclam/start?... ------------------->|
| |<-- branded login page -----|
| |-- email/password or OAuth->|
|<-- 302 /cb?code=...&state=... ----------------------------|
|-- POST /desk/auth/oauth/exchange (code+verifier) -------->|
|<-- {api_key, user} --------------------------------------|
|-- DPAPI-store key, GET /boltclam/account ---------------->|
```
The one-time code is bound to the PKCE challenge, the exact loopback URI and
a short TTL. The app rejects a callback whose state does not match. The API
burns an exchange code on the first attempt, successful or not.
## Scan sequence
1. `ClamAVService.info()` locates `clamscan.exe`, runs `--version`, and
chooses the managed database if it contains signatures.
2. `scan_targets("homes")` reads the Windows `ProfileList` registry and
falls back to `C:\Users`; `"full"` enumerates fixed local drives.
3. `ScanWorker` starts clamscan with an argv list, no shell, and parses
detections/progress/summary. Each `path: Signature FOUND` becomes a
`Detection` row first, then quarantine/delete/notify is applied.
4. The scan row is finished even on cancellation or process failure. The UI
reloads history and the Homepage aggregates.
## Real-time sequence
`DownloadWatcher` polls each watched download folder at 1.5 s intervals (and
refreshes browser preferences every 30 s), ignores in-progress suffixes, and
waits for a stable size/mtime. `RealtimeScanTask` runs in a QThreadPool:
```
stable file → exclusive handle (share=0) → copy → ClamAV scans copy
→ clean: release → detection: release + quarantine/delete
```
The exclusive handle is held for the entire scan. A fallback copy is used
when the file is already open, with that fact recorded in the event detail.
## Update flow
- The Bolt API reads Gitea's `/api/v1/repos/TheMajesticNetwork/BoltClam/releases`,
caches for five minutes, filters draft/prerelease by channel, and returns
normalized release metadata.
- Asset downloads go through the API, so a future private repository needs
no desktop token. The API streams the Gitea asset with its own token.
- The client compares semantic versions, selects a Windows `.exe`/`.zip`
asset, downloads it to `%LOCALAPPDATA%\BoltClam\updates`, and verifies the
release checksum when published.
## Storage schema
- `scans` — scope, status, paths JSON, counts, engine/DB versions, duration.
- `detections` — scan id, path, signature, action, quarantine id, timestamp.
- `quarantine` — id, original/quarantine path, signature, SHA-256, status.
- `realtime_events` — download path, signature, action, detail, timestamp.
All writes open their own SQLite connection; WAL mode lets the UI read while
a scan thread writes. Quarantine sidecars are JSON, so the record survives a
database rebuild.