v0.9.0

What changed

• Add Visual Job v2 with embedded Visual Plans and deterministic taxonomy routing across all nine modes.
• Add Visual Reviews tied to the actual PNG and render-contract hashes, enforce at most one revision, record checked HTML manifests, and publish approved artifacts atomically.
• Expand CardBench to 20 planning and revision cases, with selective single and tail runs plus honest null scores for unmeasured metrics.
• Add versioned Design Memory, current architecture documentation, refreshed mode guidance, and updated gallery artifacts.
• Preserve Visual Job v1 and direct card.js compatibility while strengthening Studio composition and schema validation.

Validation

• npm run package-skill
• npm test
• npm run smoke
• Full CardBench: 20/20 passing with real rendering, receipts, PNG review, and revision checks
• git diff --check

中文说明

• Visual Job v2 现已内嵌视觉方案（Visual Plan），并通过确定性的分类规则在全部九种模式之间完成路由。
• 视觉评审（Visual Review）现在绑定真实 PNG 与渲染契约的哈希，最多允许修订一次；通过检查的 HTML 会留下清单，审核通过的产物则以原子方式发布。
• CardBench 扩展至 20 个规划与修订案例，开发时可选择单个案例或从失败点继续运行；未测量的指标保持为 null，不会伪装成完整得分。
• 新增版本化的 Design Memory、当前架构文档、更新后的模式指南与图库产物。
• 继续兼容 Visual Job v1 和直接调用 card.js，同时加强 Studio 构图与 schema 校验。

验证结果

• npm run package-skill
• npm test
• npm run smoke
• 完整 CardBench：20/20 通过，覆盖真实渲染、receipt、PNG 审核与修订检查
• git diff --check

v0.8.1

What changed

• Add direct Claude Code marketplace installation alongside the existing Codex and skills paths.
• Keep Claude Code installations on the native plugin updater and include the exact update commands in version notices and documentation.
• Validate Claude marketplace metadata, version sync, and packaged skill paths in the main test suite.
• Refresh the gallery with new Thousand Brains examples.

Validation

• npm test
• npm run smoke
• npm run gallery:check
• Claude Code strict manifest validation and isolated marketplace installation
• GitHub Actions on Node.js 22 and 24 plus runtime smoke

v0.8.0

What changed

• Add the Visual Job production contract for validated single, mixed, and multi-artifact rendering with transactional publication, canonical hashes, and receipts.
• Add Studio composition support with explicit HTML and CSS gates, while keeping direct CLI rendering backward compatible.
• Add strict public schemas and Creative fixtures for comic, infograph, and sketchnote modes.
• Replace the legacy gallery script with production-path gallery jobs, a versioned manifest, and byte-for-byte verification.
• Split article-diagram layout helpers and styles from the renderer, and expand article, poster, and editorial composition coverage.
• Add isolated fresh-context evaluations with semantic assertions and packaged-runtime verification.

Validation

• npm test
• npm run smoke
• npm run gallery:check
• GitHub Actions on Node.js 22 and 24 plus runtime smoke

v0.7.1

What changed

• Automatically upgrade installed card-skill copies in the background after a successful render; the current render stays on its starting version and the next use picks up the update.
• Resolve updates only from the latest stable GitHub Release and pin installation to its exact commit.
• Add per-install render leases and update locks so concurrent renders and multiple installed copies cannot race or suppress each other.
• Verify installed version, content hash, and runtime readiness, with rollback when installation or preparation fails.
• Add opt-outs for all update activity or automatic installation only, plus updater self-tests and packaged-mirror validation.

Validation

• npm test
• npm run smoke
• isolated packaged install: npm ci --ignore-scripts, runtime check, and smoke
• GitHub Actions on Node.js 22 and 24 plus runtime smoke

v0.7.0

Highlights

• Add a poster + reading-notes variant that keeps each highlight paired with its exact personal thought and preserves standalone notes with explicit labels.
• Keep the first card useful by rendering the series title and real note content together, with compact note-specific typography.
• Enforce up to eight cards per batch while keeping the public schema, runtime validation, renderer, and generated plugin mirror aligned.
• Add synthetic selection and full-batch evals, HTML escaping coverage, and end-to-end PNG checks without committing personal reading data.

Validation

• Node.js 22 test suite
• Node.js 24 test suite
• Runtime smoke test
• Packaged runtime smoke test and root/package mirror checks

v0.6.0

Highlights

• Add deterministic cover_motif objects for editorial covers, so the right-side visual can reflect the article tension without sacrificing stable rendering.
• Keep paper-stack as the deliberate legacy fallback while adding drawer, window, lens, path, archive, and layers motifs.
• Require complete custom compositions for in-article and metaphor inputs, with public-schema and runtime validation kept in sync.
• Add a Chromium regression check for the window motif detail borders.

Validation

• Node.js 22 test suite
• Node.js 24 test suite
• Runtime smoke test
• Packaged runtime smoke test and root/package mirror checks

v0.5.1

Highlights

• Add a Codex inline candidate preview flow that externalizes high-cost visual decisions before rendering.
• Add the composition_required contract so selected metaphors that need custom composition cannot silently fall back to the generic scaffold.
• Enforce the contract in the public schema, runtime validator, renderer, and regression suite.
• Add repository-level AGENTS.md guidance for source/package mirror ownership and verification.

Validation

• Node.js 22 test suite
• Node.js 24 test suite
• Packaged runtime smoke test
• Package dry-run and root/package mirror hash checks

card-skill v0.4.1 - WeChat Reading Showcase

Highlights

• Added three public-safe WeChat Reading Showcase assets: annual reading report, overall reading profile, and a personal highlights card set.
• Added a README gallery section with copyable prompts showing the annual, overall, and personal-highlights workflows.
• Kept the plugin mirror synchronized with all Showcase PNGs and the 0.4.1 version fields.

Reliability

• Uses only the requested annual and overall periods plus the annual top-read book for the personal card set.
• Keeps personal highlight and thought labels separate; unpaired thoughts are not force-matched to highlights.
• Excludes low-information personal text from the public Showcase and does not commit raw API responses, IDs, or credentials.

Verification

• npm test
• npm run smoke
• visual inspection of all Showcase card types
• npm pack --dry-run --json
• GitHub CI on Node.js 22 and 24

card-skill v0.4.0 - WeChat Reading integration

Highlights

• Added an explicit WeChat Reading source adapter for turning highlights, thoughts, and personal reading statistics into card-skill visuals.
• Added sourced poster footers and monthly reading-report guidance while keeping WeChat Reading access opt-in.
• Added first-run runtime setup so Codex plugin installs can prepare Playwright and Chromium from the installed skill directory.

Reliability

• Treats all WeChat Reading responses as untrusted content and keeps credentials and private reading data out of rendered output by default.
• Adds exact-count eval assertions, sparse-report handling, package-mirror checks, and excludes prototypes from published packages.
• Fails early with an actionable runtime setup command when Playwright or Chromium is unavailable.

Verification

• npm test
• npm run smoke
• isolated packaged-install render
• npm pack --dry-run --json
• GitHub CI on Node.js 22 and 24

card-skill v0.3.0 - measured formula cards

Highlights

• Redesigned article-diagram around a single Editorial Equation formula-card grammar.
• Added real-font measurement, semantic line grouping, adaptive body-2-1 / body-3-2 selection, and readable density limits.
• Added stable semantic output markers and stricter checks for formula-card balance, overflow, and missing metrics.
• Kept concept-map, process-flow, and boundary-model inputs as legacy-compatible paths.
• Refreshed the README gallery with separate portrait and landscape rows and a new formula-card example.

Reliability

• Rejects mixed legacy/compression payloads instead of silently choosing a renderer.
• Fixes measurement viewport parsing and ensures measured layout is selected before fallback aspect estimation.
• Expands validation for short, dense, mixed-language, arrow, structure-only, and missing-marker cases.

Verification

• npm test
• npm run smoke
• npm pack --dry-run --json
• GitHub CI on Node.js 22 and 24