Skip to content

Releases

v1.2.1

Choose a tag to compare

@github-actions github-actions released this 19 Aug 13:02
94a6f47

Patch Changes

  • #9 5678733 Thanks @punkadillo! - Point package metadata at the punkadillo GitHub account. The repository,
    bugs, homepage and author URLs still referenced the old handle, which now
    404s, so every link the npm page renders was dead.

v1.2.0

Choose a tag to compare

@github-actions github-actions released this 07 Jun 22:05
7e77f5c

Minor Changes

  • #7 8589fbe Thanks @raveracker! - Coordinator resilience + autonomy, and generated-code comment economy.

    • Crash-recovery & resume. The figma-coordinator now writes a durable per-run checkpoint and can resume a crashed run — restoring the manifest + buildPlan, validating against disk (anti-confabulation), and re-dispatching only the unbuilt components/icons. Adds bounded backoff on API overload (HTTP 529) before the single retry.
    • Token reverse-lookup. New fcc kg:query --used-by <token> mode lists a token's dependent components (with a graceful in-context fallback for older fcc), powering surgical token-rename updates.
    • Autonomy policy (config.autonomy). An opt-in autonomous level resolves mid-run decision gates — unbound values, framework/CSS or library mismatch, removed tokens, ambiguous selection — from a pre-recorded policy instead of blocking, logging every decision to the handover for async review. Token/stack changes prefer an in-place update over a from-scratch rebuild. Defaults to interactive, so existing projects see no behavior change; genuinely unsafe gates (page-selected, recursion cycle, MCP/setup aborts) always stop.
    • Comment economy in generated files. Builders now emit minimal, single-line inline comments capped at 80 characters (no narrative block/banner/restating comments), while concise /** … */ JSDoc on the public component/prop API stays encouraged. Cuts output tokens on every build without sacrificing API docs.

v1.1.0

Choose a tag to compare

@github-actions github-actions released this 07 Jun 14:59
7a83bad

Minor Changes

  • #5 41b4984 Thanks @raveracker! - Push detection, provisioning, and validation upstream so builders assemble against a complete, grounded plan instead of discovering project reality mid-build. Applies to both Claude Code agents and Cursor prompts.

    • Setup-at-init (no command-time failures). The wizard now provisions each enabled track's one-time framework setup (Tailwind entry CSS + @source, vitest.config with the e2e separation already correct — including a projects/.e2e split for co-located e2e — playwright.config, .storybook/) and installs project devDeps; Playwright browsers stay user-run/verify-after. It never edits user-authored config files (greenfield write / brownfield instruct-and-verify). The coordinator verifies setup at pre-flight and aborts cleanly before spawning any specialist rather than failing mid-run.
    • Complexity scoring fix. The coordinator feeds the real (disk- or ledger-based) tokenReuseRatio into the scorer before resolving the tier, eliminating the false reuse penalty that over-tiered components; the fetcher now emits canonical tiers only (trivial|moderate|complex|extreme), never ad-hoc labels.
    • Plan-mode inventory. A cold-start inventory grounds existing assets, real on-disk token names, Figma↔disk token deltas, and house style, passed to builders as buildPlan directives.
    • Dark mode. token-builder emits [data-theme] blocks from manifest dark values on component builds too; the fetcher hardens dark-alias resolution (flag, never fabricate).
    • Docs. story-author emits a lightweight component MDX doc per component.
    • Fetcher reliability. Resume output is disk-validated before use; the manifest root configSnapshot is always populated; the coordinator re-passes configSnapshot to every follow-up spawn.
    • Icons. Vector-before-raster extraction, raster flagged as a fidelity gap, correct {...props}/style spread order, and type-only imports for verbatimModuleSyntax.
    • Config. The wizard wires a knowledgeGraph block (opt-in, default off) and a setup audit block; schema adds writeScope.setupFiles, stories.testUtilsImport, and tests.unit.{excludeE2E,setupFile}. The frozen-paths hook authorizes wizard-provisioned setup files via writeScope.setupFiles.

v1.0.1

Choose a tag to compare

@github-actions github-actions released this 28 May 17:52
2df3c14

Patch Changes

  • Update Github workflow to use node 24

v1.0.0

Choose a tag to compare

@github-actions github-actions released this 28 May 17:37
47e74bc

Major Changes

  • #1 3c73755 Thanks @raveracker! - First release — a drop-in, framework-agnostic Figma→code pipeline scaffold for Claude Code and Cursor.

    npx figma-code-composer copies a multi-agent pipeline into any frontend repo: a Figma file becomes design tokens, framework-native components, icons, Storybook stories, and tests — with a built-in knowledge graph that reuses components across screens instead of duplicating them. Nothing is bundled into your app; the CLI runs on demand.

    Pipeline output

    • Design tokens in your CSS system's native format (CSS vars, Tailwind theme, Panda config, …), with Figma variable names preserved (never resolved to raw values).
    • Framework-native components (React TSX / Vue SFC / Angular standalone / Svelte) with cva-style variants and accessibility baked in.
    • Icons: SVG → component with currentColor / literal fills + barrel re-exports.
    • Storybook stories + unit tests (Vitest / Jest / Karma) and optional Playwright E2E.

    Coverage

    • Frameworks — React (Next · Vite · Remix · Astro · CRA), Vue 3 (Nuxt · Vite · Astro), Angular ≥17 (standalone + signals), Svelte 5 (runes).
    • CSS — Tailwind v4/v3, UnoCSS, vanilla CSS-vars, CSS Modules, Sass/SCSS, vanilla-extract, Panda, styled-components.
    • Design systems — Atomic, Ant Design, Chakra, Hero UI, Mantine, MUI, Radix, shadcn/ui, or none/custom.
    • Methodologies — Atomic Design, Feature-Sliced, Component-Based, Flat/custom.

    Pipeline intelligence

    • Knowledge graph records every built component and reuses it across screens (exact + semantic match) instead of rebuilding duplicates.
    • Complexity routing picks the smallest viable model + skill set per build — saving tokens on easy designs without sacrificing quality on hard ones.
    • Handover summaries let you /clear between runs and re-hydrate from the handover + KG.
    • ~137 bundled skills, auto-pruned to your chosen stack by the wizard.
    • fcc CLI: knowledge-graph query/stage/merge, handover, complexity, and doctor health check.

    Setup & safety

    • One-time /init-figma-compose wizard detects your stack, hard-gates on a reachable Figma MCP, derives the write allowlist, and writes .figma-pipeline/config.json (the single source of truth every agent reads).
    • A config.json-driven write allowlist plus lifecycle hooks keep generation scoped to your configured output directories; .env is hard-blocked, and all Figma-derived strings are treated as data (prompt-injection guard).
    • CLAUDE.md / AGENTS.md get a managed marker block only — your own instructions survive updates.
    • Optional, detect-only (never auto-installed): Graphify knowledge graph and RTK shell-output compression.

    Tooling note: this release supports Claude Code and Cursor. Codex CLI was evaluated and removed — its Figma plugin tools are not available to codex exec, so the pipeline could not run there.