Skip to content

ROADMAP.md

Latest commit

 

History

History
82 lines (69 loc) · 10.3 KB

ROADMAP.md

File metadata and controls

82 lines (69 loc) · 10.3 KB
Raw

Roadmap

Actionable work only. Historical and completed roadmap material is archived in CHANGELOG.md; blocked work is kept in Roadmap_Blocked.md.

Actionable Items

  • P1: Build inspected install plans and correct deployment exports Why: Separate installs can leave partial dependency state, multi-app queues produce invalid App Installer semantics, bundle diffs miss inner manifests, and rollback ignores architecture. Evidence: MSStoreHelper.py::read_appx_manifest_details, generate_appinstaller_manifest, install_package, _install_worker, rollback_candidates, diff_appx_manifests; Microsoft Add-AppxPackage, MSIX update, App Installer, and DISM documentation. Touches: AppX/bundle inspector, install planner, command runner, AppInstaller/DISM/Intune exporters, rollback/diff, tests. Acceptance: Outer and inner manifests produce a per-main-app plan with exact identity, publisher, architecture, min OS, dependencies, optional/resource packages, capabilities, and installed-state conflicts; ambiguous multi-app queues are split or rejected; installs use documented dependency/optional-package batching where supported; inventory failures never imply absence; rollback/diff are architecture-safe; and dry-run output plus generated artifacts round-trip through schema/XML/PowerShell validation. Complexity: L

  • P1: Publish the support and trust contract Why: The 2026-07-29 docs overstate Python compatibility, under-explain RG-Adguard/Store API risk, advertise LAN binding without a threat warning, and do not define repair recovery or supported Windows builds. Evidence: README.md, CHANGELOG.md, CLAUDE.md, pyproject.toml; Microsoft Store policy/troubleshooting and Python lifecycle documentation. Touches: README, CHANGELOG, CLAUDE, CLI/help copy, release checklist. Acceptance: Documentation states the tested Windows/LTSC/Python matrix, user-vs-machine and elevation semantics, source/fallback support status, trust/revocation modes, offline wheelhouse matrix, mirror exposure model, destructive repair plan/backup/restore flow, state/cache migrations, CLI schema/exit codes, and unsigned artifact verification; the malformed v3.2.0 changelog heading and stale repository counts are corrected; no local SHA is described as a Store-signed hash. Complexity: S

  • P1: Version the CLI contract and persist a local operation journal Why: RMM consumers need stable schemas and post-run evidence, while v3.35.0 JSON has no schema version and GUI logs disappear with the process. Evidence: MSStoreHelper.py::build_cli_parser, _cli_emit_summary, _cli_package_workflow, _log; WinGet JSON/configuration patterns; PDQ and Ninite audit/history features. Touches: typed result model, CLI JSON/exit codes, bounded journal, diagnostics, mirror/repair/download/install events, contract tests. Acceptance: JSON includes SchemaVersion, operation ID, absolute timestamps, source/query, trust outcome, per-package state, warnings/errors, reboot requirement, and recovery artifact paths; exit codes distinguish validation, privilege, partial, trust, network, and internal failure; a bounded local JSONL journal is atomic, redacted, exportable, and telemetry-free; backward-compatibility fixtures lock the v1 contract. Complexity: M

  • P1: Make every major UI surface responsive and accessible Why: 2026-07-29 live dark/light/system rendering found clipped or invisible actions, sub-4.5:1 small text, unnamed selection controls, and little keyboard/focus support. Evidence: MSStoreHelper.py::Theme, AppTile, SearchResultTile, PackageRow, _build_ui, _build_sidebar, _show_packages; tests/test_accessibility.py; WCAG 2.2 and Microsoft accessibility guidance. Touches: theme tokens, shared controls/layouts, welcome/search/packages/queue/log/help/dialogs, accessibility/render tests. Acceptance: Deterministic screenshots pass at 100%, 125%, 150%, and 200% in System/Dark/Light and Windows High Contrast at 1000×600 and 1280×800; normal text is at least 4.5:1; content reflows or scrolls without clipped controls; every control has a visible/automation name, role, state, focus cue, and logical tab order; Enter/Space/Escape and screen-reader workflows work; loading/conflicting actions are disabled; source errors differ from empty states. Complexity: L

  • P1: Broaden local release verification Why: The local release now produces a portable app, wheel, checksums, and an artifact manifest, but the full interpreter matrix and destructive Windows repair paths still need repeatable VM coverage. Evidence: scripts/build_release.ps1, the 2026-09-05 private-desktop launch, and the local 191-test run. Touches: local release scripts, test/tool configuration, release checklist, README. Acceptance: Local isolated environments test Python 3.11 through 3.14, install the wheel, exercise the CLI, and launch the packaged GUI; an isolated Windows VM validates PowerShell, AppX, DISM, repair dry-runs, and restore; releases add an SBOM and preserve the current checksums and artifact manifest. Complexity: L

  • P2: Extract backend services from the GUI monolith Why: The 5,303-line module couples trust-critical backend behavior to CustomTkinter state, making concurrency, CLI, testing, and migration changes harder to isolate. Evidence: MSStoreHelper.py, existing seams in msstore_package_resolution.py and store_sources.py. Touches: new store client, downloader/cache, state, trust/inspection, command runner, repair, exporter, operation, and GUI view-model modules; imports/tests/package metadata. Acceptance: CLI startup does not import CustomTkinter; backend services accept typed immutable inputs and have no Tk globals; GUI binds only through a view model/operation coordinator; PowerShell construction is centralized; persistence is behind repositories/migrations; existing public CLI behavior and all tests remain green after each incremental extraction. Complexity: L

  • P2: Consume verified offline repository indexes Why: MSStoreHelper can write shared caches and mirror indexes but cannot browse, import, or install from them as a first-class offline source. Evidence: MSStoreHelper.py::build_mirror_index, cache_downloaded_artifact, _rollback_cache_folders; Raven, StoreLib, PDQ local repository/cache patterns. Touches: source adapter interface, mirror/index client, trust gate, cache browser, GUI/CLI import, queue URL refresh, tests. Acceptance: A versioned index can be opened from disk or an authenticated mirror; schema, size, hash, signature/identity, architecture, and query metadata are verified before queueing; packages group by main app/dependency; missing or stale entries can be refreshed only when online; a fully disconnected machine can browse, plan, install, and export from verified cached artifacts without contacting Store services. Complexity: L

  • P2: Add official Store and WinGet execution backends Why: Microsoft’s 2026-02-11 Store release exposes store install/update, while WinGet already provides msstore policy/error semantics; both are useful when present but cannot be assumed. Evidence: store_sources.py fallback hints, MSStoreHelper.py::detect_source_health; Microsoft Store developer-tools announcement and WinGet documentation. Touches: source/execution adapters, capability discovery, CLI/GUI source selection, typed results, diagnostics, tests. Acceptance: Capability detection distinguishes missing, policy-blocked, unauthenticated, and usable Store/WinGet backends; users can explicitly choose or order backends; commands use exact product IDs/arguments and return typed results; no backend silently changes ring/market/architecture intent; direct/offline workflows remain available and no Store/App Installer dependency becomes mandatory. Complexity: M

  • P2: Add package and queue workflow controls Why: Competitors expose filtering, grouping, per-item retry/remove/cancel, version pins, and operation history; the v3.35.0 package list and queue are all-or-nothing. Evidence: MSStoreHelper.py::_show_packages, QueueItem, _update_queue_ui; Raven and UniGetUI; WinGet pinning. Touches: package/queue view models, responsive UI, persisted state schema, operation coordinator, CLI. Acceptance: Packages filter/sort/group by app, dependency role, architecture, type, encryption, version, and size; queue items support remove, retry, cancel, and safe app-group reorder; users can choose output/cache folders in GUI; estimated bytes/free space and pin/skip reasons are visible; all controls persist through the versioned state model and remain keyboard accessible. Complexity: M

  • P2: Externalize strings and preserve Unicode end to end Why: Locale/market controls exist, but UI/log/help strings are embedded in code and release-note decoding corrupts non-ASCII text. Evidence: MSStoreHelper.py::parse_release_notes_html and embedded UI strings; Raven and UniGetUI localization; Microsoft accessibility/globalization guidance. Touches: resource loader/catalog, GUI/CLI/help/log copy, locale selection, release-note parser, layout/render tests. Acceptance: User-facing strings use locale resources with stable keys and en-US fallback; no unicode_escape round-trip is used for already-decoded text; Japanese and accented regression fixtures preserve exact Unicode; pseudo-localization catches clipping; locale/market are separate concepts; untranslated keys and formatting mismatches fail tests. Complexity: L

  • P2: Add deterministic local deployment policy profiles Why: Commercial tools derive value from approvals, pins, rings, maintenance windows, detection, and retry evidence without requiring every local utility to become a fleet server. Evidence: Patch My PC rings/alerts, PDQ retry/deployment history, Ninite update policies, Intune detection/supersedence, WinGet pin/configuration. Touches: versioned profile schema, planner, keep-updated mode, CLI, WinGet/Intune/AppInstaller exports, journal, docs/tests. Acceptance: Importable/exportable profiles declare approved product IDs, architecture/market/ring, version pin/ignore rules, maintenance window, retry/reboot policy, trust mode, and dry-run behavior; Keep Updated and CLI consume the same plan; Intune exports include deterministic detection and supersedence metadata; policy changes produce a human-readable diff and never enable paid/restricted acquisition. Complexity: L