Skip to content

Latest commit

 

History

History
100 lines (84 loc) · 6.28 KB

ROADMAP.md

File metadata and controls

100 lines (84 loc) · 6.28 KB

Roadmap

Last reconciled: 2026-07-31 against origin/master at 705f89d.

Weaver is a capable cross-platform engine in pre-alpha, not yet a complete end-user product. The authoring/runtime/artifact spine is real and the noro-player fidelity gate is complete. The missing center is the trust and management shell that turns those pieces into the full Conjure → Share → Remix loop.

What exists now

Surface State Receipt / boundary
TSX + Tailwind-like authoring Working on Windows and macOS Retained reconciler, hooks, styling packs, icons/images/stack/interaction, Canvas, and agent-readable weaver check; public contract in sdk/CONTRACT.md
Native runtime and desktop windowing Working developer builds Zig + QuickJS, no browser/webview, per-Widget crash isolation, transparent desktop surfaces, Windows per-monitor DPI, AppKit parity
Renderer Working; performance hardening active Shared GPU renderer on Windows; macOS shared-renderer cutover landed in PR #47 after the per-process Metal submission wall was measured
Providers Working, bounded surface Time, CPU, memory, audio, and media; collection is host-owned and subscriber-driven; network access is per-Widget through declared origins
Media parity gate Complete Metadata, artwork, transport, seek, and noro-player attended on both OSes; macOS exact 15.4 floor and shipping notarization remain unverified
Dev lifecycle Working init, check, bundle, state-preserving dev, logs, host status, crash/backoff recovery
Portable artifact lifecycle Working Deterministic source-carrying .weave, inspect/install/replace/uninstall, install-owned source boundary
End-user product Missing No manager, loud grant UI, remix command, signed installer/updater, or public gallery
Beta breadth 1 target complete noro-player is done; the remaining in-scope skin parity targets are the demand signal

“Working” here means the developer-build path exists with tests and recorded evidence. It does not mean stable API, public distribution, or beta support.

The gaps, in dependency order

  1. Close the current performance lineage. The macOS shared renderer is on master, while the Visualizer/Canvas CPU reductions were developed on the older feat/macos-memory-work lineage. Reconcile only the still-relevant patches onto the shared-renderer head, then remeasure the complete workload. Publish 1/2/4/8-Widget physical-footprint, CPU, wakeup, frame, recovery, and 30-minute drift receipts. ADR 0018 records the shipped architecture and supersedes the older in-process decision.
  2. Build the trust surface and thin manager together. Add the tray/menu-bar manager with Widget list, start/stop, health, measured cost, logs, declared origins/capabilities, and permission state. Implement ADR 0002's loud grant flow here, beginning with the first skin-demanded rung (launch-app or open-url). Every denial and unavailable provider must remain actionable from both the UI and weaver status --json.
  3. Complete the Loop with weaver remix. Export the installed source to an editable directory, preserve provenance, append lineage, and hand the result to the agent. Upgrade the conjure skill to cover the full current SDK and add a remix skill that can act from the artifact and its diagnostics alone.
  4. Earn beta breadth one named skin at a time. Use skin-parity-targets.md, not a speculative API wishlist. The derived backlog currently includes animation, gradients, blur/glow, canvas paths/arcs, rotated text, text input, multi-Widget suites, launch/open capabilities, and weather/network/GPU/RSS/mail providers. Each addition lands with the demanding skin, cross-platform pixels and behavior, idle/active receipts, and agent-facing check errors.
  5. Close the physical platform matrix. Verify macOS external displays, Stage Manager/Spaces/fullscreen/lock, sleep/wake, audio revoke and route recovery, Bluetooth/AirPlay, physical Intel, and MediaRemote at exactly macOS 15.4. Repeat the relevant interaction and performance receipts on the Windows release floor. Unverified hardware behavior stays named.
  6. Define the gallery contract, then package. The gallery can develop its browse, provenance, lineage, capability-badge, and one-click-install contract before packaging, but it does not launch publicly first. Ship a signed/notarized macOS direct-distribution bundle and a Windows installer containing the host, renderer, runtime, CLI, builder, startup registration, updater, repair, and clean uninstall. The macOS MediaRemote route makes Mac App Store distribution out of scope unless that provider strategy changes.
  7. Launch the gallery. Seed it with the attended skin recreations, preserve source visibility and lineage through install/remix/re-publish, and make declared access obvious before install.

Release bars

Alpha

A person who has not cloned the repository can install Weaver, prompt an agent to conjure a Widget, inspect and grant its declared access, keep it running across login/restart, see its real resource bill, install a .weave, remix it, and cleanly remove both the Widget and Weaver on Windows and macOS.

Beta

Every in-scope target in skin-parity-targets.md is indistinguishable side-by-side and functionally equivalent on both OSes, with attended evidence. Target 14 (TranslucentTaskbar) needs an explicit scope ruling because it modifies an OS-owned surface rather than creating a Widget. The packaged update/repair/uninstall paths and the public gallery loop must also pass.

v1

The Alpha and Beta bars hold across the supported hardware/OS matrix; crash recovery, upgrades, provider failure, permission changes, and multi-Widget cost have long-run receipts; the authoring and artifact contracts are stable enough that an agent can build and repair Widgets without repository knowledge.

Deliberate non-goals

GPU text atlases until profiling demands them; a second JSX sugar layer (TSX won); embedded web pages; App Store distribution with the current macOS media route; Linux until it has an owner and measured product demand.