Roadmap.

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports.

Restructure the dev-server sidebar (the FILES panel at app.jsx:273–365) so it stops drowning daily work in non-canvas noise:

This plan converts the retro at .ai/logs/system-reviews/design-system-bootstrap-review.md (R1–R6) into sequenced work, plus folds in the multi-DS portion of the old Phase 5 (Tasks 2–7) since multi-...

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports.

Surfaced during /flow:done closeout of feature-docs-site-mdcc-skin on 2026-05-15. The parent feature shipped (commits 78d9d8f + 94b4e77) with 0 hard-blockers in scenario-runner but DS-guard + a11y-...

Validate canvas + fumadocs source before implementing. Token names, fumadocs --fd-* slots, and Berkeley/JetBrains fallback chain are authoritative — don't paraphrase them.

Keep oxc-parser. Work around a Bun 1.3.4+ regression in --compile NAPI native-binding embedding by setting NAPI_RS_NATIVE_LIBRARY_PATH from a Bun-embedded asset path before oxc-parser loads. Per-ta...

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports (DDR-045 paths, DDR-013 module layout, DDR-048 input-router classify table).

Replace the current canvas format — single-file HTML with inline <style> (~50 KB hand-rolled CSS) + <script type="text/babel"> JSX + React 18 UMD + Babel-standalone UMD, reference canvas weighing 7...

Two interlocked workstreams, one phase, because the visual-sanity gate (workstream A) is the same screenshot pass that catches the motion regressions workstream B prevents — splitting them would re...

A first-class export feature surfaced from the canvas UI (not just CLI). One dialog covers seven formats × four scopes:

Lay down the Yjs runtime, Awareness protocol, and persistence layer that Phase 9 needs — and as a side-effect, enable multi-tab collab on a single machine: two browser tabs (or two Claude Code inst...

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports. The four key decisions are baked in below — do not re-litigate them mid-execution.

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports in plugins/flow/.

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports. The 3-stage architecture below is load-bearing — do not collapse stages, do not reintr...

Establish the contribution baseline: documented "how to contribute" path, GitHub repo hygiene (PR / issue templates, branch protection docs, Dependabot), basic CI quality gates beyond version parit...

Productize release-note + release-cut hygiene as a reusable flow capability. Add a new integrations.changelog block to .ai/workflows.config.json (provider + scope + releaseGuide path), implement th...

Stand up a public documentation site under site/ using Fumadocs — a Next.js-based MDX docs framework chosen because its output is AI-readable (clean MDX in repo, indexable via crawlers + LLM tools)...

Reshape the 29 /flow:* commands (26 existing + 3 Phase 3 additions: help, release-changelog, release-cut) so every command except daily-use ones carries a group prefix. Result:

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports.

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports in plugins/design/agents/, plugins/design/commands/, and .design/system/project/.

Refactor the dev-server's shell to a build-pipeline-driven, Bun-runtime-authoritative architecture targeting concrete performance budgets. Concretely: migrate server source to Bun.serve / Bun.file...

Two-part scope:

Make every TSX canvas — UI mock OR DS specimen — a fully self-contained, browser-loadable canvas backed by a small shared canvas-lib (@mdcc/canvas-lib, virtual-module-resolved at the dev-server lev...

Today every .tsx canvas under .design/ renders inside its iframe as a static TSX flow — DCArtboard is a labeled card that participates in normal document flow. Phase 4 turns DesignCanvas into a tra...

Implements DDR-025. The current architecture (DDR-022) treats canvas-lib as project-owned source that gets scaffolded into <designRoot>/_lib/canvas-lib.tsx on first /design:setup-ds. That design ha...

Add a FigJam-equivalent input layer to canvas-lib's DesignCanvas as an opt-in inputMode prop:

Make artboards spatially editable on the infinite canvas:

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports.

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports.

Produce two 1920×1080 H.264 MP4s — site/public/demo.mp4 (primary, ~60s, embedded in the docs landing) and site/public/demo-30s.mp4 (tight cut for the GitHub README, < 10 MB) — by:

Produce one MP4 at site/public/demo.mp4 (~90 s, single cut, < 16 MB post-loudnorm) by composing 16 scenes on top of phase-15.1 infrastructure (nested workspace, <TransitionSeries>, <TerminalFrame>...

Toolbar buttons in canvas chrome: pen, circle, arrow, eraser, color picker. Drawing happens on a transparent SVG layer sized to the canvas world (zoomable with viewport from Phase 4). On stroke com...

Rework annotations-layer.tsx from a self-contained pointer-claiming overlay into a portal-rendered layer that lives inside .dc-world (so the world's CSS zoom/translate moves the strokes natively, w...

Replace the current shell-bound BottomBar composer with a FigJam-grade in-place comment experience: the composer pops up at the click location on the canvas (not in a panel chrome at the bottom), p...

Validate the docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports — this plan refactors load-bearing bootstrap contracts (design-system skill ↔ templ...

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports.

Make collaboration usable across the internet without exposing anyone's laptop, without requiring SaaS, without making contributors learn Yjs / PartyKit / Hocuspocus. Three pieces:

Build on Phase 8's "ambient multiplayer" (Yjs awareness + comments) to add true real-time HTML co-editing for designated edit surfaces. Two peers can change a button's padding simultaneously throug...

Close the gap between flow and design plugins. Flow commands (/flow:plan, /flow:done) become aware of .design/ — /flow:plan automatically detects canvases relevant to a feature and pulls them into...

Add Figma / Webflow-grade direct manipulation: a layers panel (DOM tree of active artboard — collapsible, selectable, drag-to-reorder) and an in-canvas CSS editor (right sidebar that surfaces top-e...

Validate docs and codebase patterns before implementing. Pay attention to existing naming, utils, and imports — the dev-server is Bun-authoritative (DDR-009), the canvas lib is single-source in plu...

---

Add a small library of diagram primitives to the docs site and insert them into the highest-traffic pages. Today the docs are text-heavy markdown tables — accurate but visually flat. Adding SKU-sta...

Today the annotation toolkit ships pen / rect / ellipse / arrow / text (anchored only) / eraser. Real review workflows that the user keeps reaching FigJam for need three things we don't have:

---

---

---

Embed a chat sidebar directly in the canvas chrome that speaks the Agent Client Protocol (ACP). The sidebar can connect to a locally-running Claude Code session (or any ACP-compatible agent) so the...