# Windows Analyzer Design Brief (Draft)

> Owners: Scanner Guild, Policy Guild, Offline Kit Guild, Security Guild  
> Related backlog (proposed): SCANNER-ENG-0024..0027, DOCS-SCANNER-BENCH-62-002  
> Status: Draft — contingent on Windows demand threshold (see `docs/benchmarks/scanner/windows-macos-demand.md`)

## 1. Objectives & boundaries
- Provide deterministic inventory for Windows Server/container images covering MSI/WinSxS assemblies, Chocolatey packages, and registry-derived installers.
- Preserve replayability (layer fragments, provenance metadata) and align outputs with existing SBOM/policy pipelines.
- Respect sovereignty constraints: offline-friendly, signed rule bundles, no reliance on Windows APIs unavailable in containerized scans.

Out of scope (Phase 1):
- Live registry queries on running Windows hosts (requires runtime agent; defer to Zastava/Runtime roadmap).
- Windows Update patch baseline comparison (tracked separately under Runtime/Posture).
- UWP/MSIX packages (flagged for follow-up once MSI parity is complete).

## 2. Architecture overview
```
Scanner.Worker (Windows profile)
 ├─ Surface.Validation (enforce layer size, path allowlists)
 ├─ Surface.FS (materialized NTFS image via 7z/guestmount)
 ├─ MsiCollector              -> LayerComponentFragment (windows-msi)
 ├─ WinSxSCollector           -> LayerComponentFragment (windows-winsxs)
 ├─ ChocolateyCollector       -> LayerComponentFragment (windows-choco)
 ├─ RegistryCollector         -> Evidence overlays (uninstall/services)
 ├─ DriverCapabilityMapper    -> Capability overlays (kernel/user drivers)
 └─ WindowsComponentMapper    -> ComponentGraph + capability metadata
```

- Collectors operate on extracted filesystem snapshots; registry access performed on exported hive files produced during image extraction (document in ops runbooks).
- `WindowsComponentMapper` normalizes component identities (ProductCode, AssemblyIdentity, Chocolatey package ID) and merges overlapping evidence into deterministic fragments.

## 3. Collectors
### 3.1 MSI collector
- Input: `Windows/Installer/*.msi` database files (Jet OLE DB), registry hive exports for product mapping.
- Implementation approach:
  - Use open-source MSI parser (custom or MIT-compatible) to avoid COM dependencies.
  - Extract Product, Component, File, Feature, Media tables.
  - Compute SHA256 for installed files via Component table, linking to WinSxS manifests.
- Output metadata: `productCode`, `upgradeCode`, `productVersion`, `manufacturer`, `language`, `installContext`, `packageCode`, `sourceList`.
- Evidence: file paths with digests, component IDs, CAB/patch references.

### 3.2 WinSxS collector
- Input: `Windows/WinSxS/Manifests/*.manifest`, `Windows/WinSxS/` payload directories, catalog (.cat) files.
- Parse XML assembly identities (name, version, processor architecture, public key token, language).
- Map to MSI components when file hashes match.
- Capture catalog signature thumbprint and optional patch KB references for policy gating.

### 3.3 Chocolatey collector
- Input: `ProgramData/Chocolatey/lib/**`, `ProgramData/Chocolatey/package.backup`, `chocolateyinstall.ps1`, `.nuspec`.
- Extract package ID, version, checksum, source feed, installed files and scripts.
- Note whether install used cache or remote feed; record script hash for determinism.

### 3.4 Registry collector
- Input: Exported `SOFTWARE` hive covering:
  - `Microsoft\Windows\CurrentVersion\Uninstall`
  - `Microsoft\Windows\CurrentVersion\Installer\UserData`
  - `Microsoft\Windows\CurrentVersion\Run` (startup apps)
  - Service/driver configuration from `SYSTEM` hive under `Services`.
- Emit fallback evidence for installers not captured by MSI/Chocolatey (legacy EXE installers).
- Record uninstall strings, install dates, publisher, estimated size, install location.

### 3.5 Driver & service mapper
- Parse `SYSTEM` hive `Services` entries to detect drivers (type=1 or 2) and critical services (start mode auto/boot).
- Output capability overlays (e.g., `windows.driver.kernelMode(true)`, `windows.service.autoStart("Spooler")`) for Policy Engine.

## 4. Component mapping & output
- `WindowsComponentMapper`:
  - Generate `LayerComponentFragment`s with synthetic layer digests (e.g., `sha256:stellaops-windows-msi`).
  - Build `ComponentIdentity` with PURL-like scheme: `pkg:msi/<productCode>` or `pkg:winsxs/<assemblyIdentity>`.
  - Include metadata: signature thumbprint, catalog hash, KB references, install context, manufacturer.
- Capability overlays stored under `ScanAnalysisKeys.capability.windows` for policy consumption.
- Export Center bundling:
  - Include MSI manifest extracts, WinSxS assembly manifests, Chocolatey nuspec snapshots, and service/driver capability CSV.

## 5. Policy integration
- Predicates to introduce:
  - `windows.package.signed(expectedThumbprint?)`
  - `windows.package.unsupportedInstallerType`
  - `windows.driver.kernelMode`, `windows.driver.unsigned`
  - `windows.service.autoStart(name)`
  - `windows.choco.sourceAllowed(feed)`
- Lattice approach:
  - Unsigned kernel drivers → default `fail`.
  - Unknown installer sources → `warn` with escalation on critical services.
  - Chocolatey packages from non-whitelisted feeds → configurable severity.
- Waiver semantics bind to product code + signature thumbprint; waivers expire when package version changes.

## 6. Offline kit & distribution
- Package:
  - MSI schema definitions and parser binaries (signed).
  - Chocolatey feed snapshot (nupkg archives + index) for allow-listed feeds.
  - Windows catalog certificate chains + optional CRL/OCSP caches.
- Documentation:
  - Provide instructions for exporting registry hives during image extraction (PowerShell script included).
  - Note disk space expectations (Chocolatey snapshot size, WinSxS manifest volume).

## 7. Testing strategy
- Fixtures:
  - Sample MSI packages (with/without transforms), WinSxS manifests, Chocolatey packages.
  - Registry hive exports representing mixed installer types.
- Tests:
  - Unit tests for each collector parsing edge cases (language-specific manifests, transforms, script hashing).
  - Integration tests using synthetic Windows container image layers (generated via CI on Windows worker).
  - Determinism checks ensuring repeated runs produce identical fragments.
- Security review:
  - Validate script execution paths (collectors must never execute Chocolatey scripts; inspect only).

## 8. Dependencies & open questions
| Item | Description | Owner | Status |
| --- | --- | --- | --- |
| MSI parser choice | Select MIT/Apache-compatible parser or build internal reader | Scanner Guild | TBD |
| Registry export tooling | Determine standard script/utility for hive exports in container context | Ops Guild | TBD |
| Authenticodes verification locus | Decide scanner vs policy responsibility for signature verification | Security Guild | TBD |
| Feed mirroring policy | Which Chocolatey feeds to mirror by default | Product + Security Guilds | TBD |

## 9. Proposed backlog entries
| ID (proposed) | Title | Summary |
| --- | --- | --- |
| SCANNER-ENG-0024 | Implement Windows MSI collector | Parse MSI databases, emit component fragments with provenance metadata. |
| SCANNER-ENG-0025 | Implement WinSxS manifest collector | Correlate assemblies with MSI components and catalog signatures. |
| SCANNER-ENG-0026 | Implement Chocolatey & registry collectors | Harvest nuspec metadata and uninstall/service registry data. |
| SCANNER-ENG-0027 | Policy & Offline integration for Windows | Define predicates, CLI toggles, Offline Kit packaging, documentation. |

## 10. References
- `docs/benchmarks/scanner/deep-dives/windows.md`
- `docs/benchmarks/scanner/windows-macos-demand.md`
- `docs/modules/scanner/design/macos-analyzer.md` (structure/composition parallels)
- Surface design docs (`surface-fs.md`, `surface-validation.md`, `surface-secrets.md`) for interfacing expectations.

Further reading: `../../api/scanner/windows-coverage.md` (summary) and `../../api/scanner/windows-macos-summary.md` (metrics dashboard).

Policy readiness alignment: see `../policy/windows-package-readiness.md` (POLICY-READINESS-0002).

Upcoming milestone: FinSecure Corp PCI review requires Authenticode/feed decision by 2025-11-07 before Windows analyzer spike kickoff.