[Colafornia]

English

Summary

Markdown diagrams in conversations, Daily Review, and Markdown previews now render with Pele 1.9.0 instead of Mermaid.js. Diagrams render faster, the desktop app ships less JavaScript, and a theme change no longer renders diagrams again.

• Faster rendering: median component render time drops 69.7% in local benchmarks (438.2 → 132.8 ms).
• Smaller bundles: total desktop JavaScript gzip size drops 28.9%; the diagram engine drops 72.5%.
• Fewer dependencies and less code: remove the Mermaid.js dependency tree and simplify the diagram component, cache, and styles.
• Less interaction overhead: theme changes update styles directly; continuous window resizing relayouts diagrams only after resizing settles.

Visual comparison

Left: Mermaid.js; right: Pele. Both sides use the same source, width, and real Maka component and theme. The examples adapt Pele's flowchart and sequence examples to Maka scenarios.

The 1× recording renders a 128-node, 260-edge diagram of eight eval batches side by side. In this single run, component-to-next-paint time was 613.5 ms with Mermaid.js and 93.5 ms with Pele.

Diagram source · Capture method and example attribution

Diagram	Light theme	Dark theme
Agent turn flowchart
Runtime Host architecture
Tool approval sequence

Verification

Tested head: 7eb192f71, rebased onto a4b5f86c5.

• npm run build, npm run typecheck, npm run lint, npm run format:check, npx knip --workspace packages/ui, and npx knip --workspace apps/desktop passed.
• npm --workspace @maka/ui run test:dist: 823 tests passed. Each regression test below fails when its protection is removed:
  • A resize burst keeps the mounted diagram and lays it out once after resizing settles.
  • A container that is still wider than the diagram does not lay it out again.
  • Rendering one diagram at 25 widths does not evict another cached diagram.
  • Diagram output contains no links, scripts, foreignObject, or images, and a 501-edge graph is rejected.
• node --test packages/ui/scripts/mermaid.browser.test.mjs: 3 real Chromium tests passed. They cover flowchart, sequence, state, pie, and timeline diagrams; Chinese, subgraph, and multiline labels; narrow layouts; zoom and pan; fullscreen and source; theme changes; syntax and resource failures; and PNG colors and fonts. CI does not run this script.
• Resize cost: in Chromium, three automatically rendered 200-edge flowcharts were resized in 40 steps from 900 px to 510 px. 28a20fbd6 produced 225 SVG mutations, one relayout of every diagram per step. This head produced 12, one relayout per diagram after resizing settled. These local diagrams take about 15 ms each to lay out, so they did not reproduce the 241–446 ms renderer tasks that the review of 28a20fbd6 reported for heavier diagrams.
• First render in a narrow container: a 101-node Chinese flowchart that must adapt to a 576 px container took a median of 392.9 ms before the narrow-container check and 398.5 ms after it, over 7 alternating runs each. Both produced the same layout.
• GitHub CI, dependency audit, and CLI package validation track this head.

Earlier validation on 28a20fbd6, which has the same tree as 4ae8d4caa:

• A clean npm ci --ignore-scripts, followed by the repository's dependency patches, and npm --workspace @maka/ui run build passed in an isolated checkout.
• Both notice-generator tests and the Desktop and CLI notice checks passed. The shipped-dependency audit found no moderate-or-higher advisories across 258 packages.
• Not run: installer build, OS-native clipboard end-to-end testing, and full Mermaid extension compatibility.

Performance

Baseline a509747d6 uses Mermaid.js 11.17.2; ae3240952 uses Pele 1.9.0. Each renderer ran five alternating times in fresh Chromium pages with production esbuild bundles, rendering the same 101-node, 100-edge Chinese flowchart in a 640 px container. Both completed 5/5 runs. The later cache change does not alter the first-render path. The later narrow-container check adds Pele passes to a cold render in a narrow container; Verification reports its measured effect.

Measurement	Mermaid.js 11.17.2	Pele 1.9.0	Reduction
Component path, median [min–max]	438.2 ms [423.0–463.0]	132.8 ms [131.2–137.1]	69.7%
Engine cold path, median [min–max]	296.5 ms [277.6–441.7]	26.9 ms [26.4–30.4]	90.9%
Engine-only esbuild JS, gzip	959,646 B	263,793 B	72.5%
Desktop Vite output, all JS gzip sum	2,435,078 B	1,732,078 B	28.9%
Desktop Vite output, JS chunks	143	48	—

The component path covers script execution, Maka Theme + MermaidDiagram mounting, SVG completion, and the next animation frame. Neither path measures network transfer or Electron startup. These are local macOS arm64 results (12 logical CPUs, Node 24.22.1, no CPU throttling), not a speed guarantee.

Raw samples · Vite output sizes · Benchmark script (not included in this PR)

Implementation details

• Replace Mermaid.js and its dependency tree with Pele. Use Pele's built-in zoom, pan, and width-responsive layout.
• Map Maka theme tokens to Pele CSS variables, so a theme change restyles the mounted SVG.
• Keep fullscreen, source view and copy, and PNG copy. PNG export resolves theme colors and embeds packaged fonts.
• Keep SVG sanitization, disabled links and images, source and edge limits, the automatic-render budget, and per-instance SVG IDs. A failed render still shows copyable source.
• Lay a diagram out again only after its container stops resizing, and only when the container is narrower than the diagram's natural layout and the diagram adapts to width. Until then, CSS scales the mounted SVG.
• Cache one rendered template per diagram source. A relayout replaces that entry, so resizing a window cannot evict other diagrams.
• Preserve Pele's MIT and embedded Lucide ISC license texts in packaged notices.

Pele uses its own layout and styling. This change does not promise identical Mermaid output or support for every Mermaid extension.

Fixes #6257

Review focus

• Root lib setting. tsconfig.base.json now sets lib: ["ES2024", "DOM", "DOM.Iterable"]. Before this change, ES2023+ library types reached packages/ui by accident: Mermaid's types import type-fest without declaring it, and that import resolved to a hoisted type-fest@0.13.1 whose source/basic.d.ts contains /// <reference lib="esnext"/>. Without Mermaid, findLastIndex and Promise.withResolvers fail type checking in packages/ui and in apps/desktop/tsconfig.storybook.json. ES2024 is the lowest setting that passes.
• Dependency age. pele@1.9.0 was published on 2026-10-22 with npm provenance from GitHub Actions. The project's first release was on 2026-10-06.
• CI contracts. The palette guard now explicitly allows the two font-family declarations while still rejecting colors and unknown properties. The KaTeX font-build test resolves CSS from @maka/ui, rather than relying on Mermaid's former root-level dependency.

Remaining before ready for review

• Trace and compare the ResizeObserver loop completed with undelivered notifications warning. The same Workhub conversation produced two warnings on upstream ac1d288c4 and two on this branch, with no Mermaid diagrams present. The trace identified synchronous virtua list-resize updates; this reproduction predates the migration. Published builds were not tested.
• Trace a controlled WebContents MaxListenersExceededWarning reproduction on 7eb192f71. Restarting with the existing development profile and switching history sessions rapidly 20 times did not trigger it; the listener count peaked at 10. Two additional read-only subscriptions through maka.sessions.subscribeEvents raised destroyed listeners from 9 to 11, triggering the warning during listener registration in RuntimeHostSessionObservationRegistry.observe(). Unsubscribing returned the count to 9. This registration and cleanup path is unchanged from pre-migration base a4b5f86c5. This identifies a listener-threshold trigger; it does not establish a persistent leak. The original UI sequence and an upstream runtime comparison remain unverified.
• Measure render cost while a window with several large diagrams is resized continuously. Relayout now waits until resizing settles; Verification reports the result.

AI use

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: OpenAI Codex implemented the migration, tests, benchmarks, visual captures, and the first description under user direction. Devin reviewed the change; implemented the per-source render cache, its regression test, and the output-safety tests; removed the benchmark script and an unused CSS rule; and re-captured the review screenshots and recording with English examples. After review, Codex implemented the narrow-container relayout check, and Devin made relayout wait until resizing settles and wrote its regression test. Affected commits carry Generated-by trailers; keep them when squashing.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

中文

概述

会话、每日回顾和 Markdown 预览中的图表改由 Pele 1.9.0 渲染，不再使用 Mermaid.js。图表渲染更快，桌面应用分发的 JavaScript 更少，切换主题时也不再重新渲染图表。

• 渲染更快： 本地基准中，组件渲染耗时中位数减少 69.7%（438.2 → 132.8 ms）。
• 包体更小： 桌面 JavaScript 的 gzip 总量减少 28.9%；图表引擎减少 72.5%。
• 依赖与代码更少： 移除 Mermaid.js 依赖树，精简图表组件、缓存和样式代码。
• 交互开销更低： 切换主题直接更新样式；连续调整窗口大小时，停止后才重新布局。

视觉对比

左侧为 Mermaid.js，右侧为 Pele。两侧使用相同的源码、宽度，以及真实的 Maka 组件与主题。示例由 Pele 官方的流程图和时序图示例改编为 Maka 场景。

1× 原速录屏并排渲染一张包含 8 个评测批次、128 个节点、260 条连线的图表。此次单次运行中，从组件调用到下一绘制帧的耗时为 Mermaid.js 613.5 ms、Pele 93.5 ms。

图表源码 · 采集方法与示例来源

图表	浅色主题	深色主题
智能体单轮流程图
Runtime Host 架构图
工具审批时序图

验证

已验证提交：7eb192f71，基于 a4b5f86c5 完成 rebase。

• npm run build、npm run typecheck、npm run lint、npm run format:check、npx knip --workspace packages/ui 和 npx knip --workspace apps/desktop 均通过。
• npm --workspace @maka/ui run test:dist：823 项测试通过。下列每项回归测试在移除对应保护后都会失败：
  • 连续调整大小期间保留已挂载的图表，停止调整后只重新布局一次。
  • 容器仍比图表宽时，不重新布局。
  • 一张图在 25 种宽度下渲染，不会挤掉另一张已缓存的图。
  • 图表输出中不出现链接、脚本、foreignObject 和图片，并拒绝 501 条连线的图。
• node --test packages/ui/scripts/mermaid.browser.test.mjs：3 项真实 Chromium 测试通过，覆盖流程图、时序图、状态图、饼图和时间线图；中文、子图和多行标签；窄容器；缩放与平移；全屏与源码；主题切换；语法与资源限制失败；PNG 颜色与字体。CI 不运行此脚本。
• 调整大小的开销：在 Chromium 中，将三张自动渲染的 200 条连线流程图从 900 px 分 40 步缩到 510 px。28a20fbd6 产生 225 次 SVG 变更，即每一步都重新布局全部图表；本提交产生 12 次，即停止调整后每张图重新布局一次。本地这些图每张布局约 15 ms，因此没有复现审阅意见针对 28a20fbd6 报告的、更复杂图表下 241–446 ms 的渲染线程长任务。
• 窄容器中的首次渲染：一张 101 个节点、需要适应 576 px 容器的中文流程图，加入窄容器判断前后的耗时中位数分别为 392.9 ms 和 398.5 ms，两侧各交替运行 7 次，布局结果相同。
• GitHub CI、依赖审计和 CLI 包验证对应此提交。

此前在 28a20fbd6 上完成的验证（与 4ae8d4caa 的文件树相同）：

• 在隔离目录中完成干净的 npm ci --ignore-scripts 安装，随后应用仓库的依赖补丁，npm --workspace @maka/ui run build 通过。
• 两项许可证生成测试、Desktop 与 CLI 许可证清单检查均通过。发布依赖审计覆盖 258 个包，未发现中危及以上漏洞。
• 未执行：安装包构建、操作系统原生剪贴板端到端测试，以及全部 Mermaid 扩展兼容性验证。

性能

基线 a509747d6 使用 Mermaid.js 11.17.2；ae3240952 使用 Pele 1.9.0。每个引擎在新建的 Chromium 页面中交替运行五次，使用生产模式 esbuild 包，在 640 px 容器中渲染同一张 101 个节点、100 条连线的中文流程图。两侧均 5/5 成功。之后的缓存修改不影响首次渲染路径。之后加入的窄容器判断会在窄容器冷渲染时增加 Pele 渲染次数，实测影响见“验证”一节。

测量项	Mermaid.js 11.17.2	Pele 1.9.0	减少
组件路径，中位数 [最小–最大]	438.2 ms [423.0–463.0]	132.8 ms [131.2–137.1]	69.7%
引擎冷路径，中位数 [最小–最大]	296.5 ms [277.6–441.7]	26.9 ms [26.4–30.4]	90.9%
仅引擎的 esbuild JS，gzip	959,646 B	263,793 B	72.5%
桌面 Vite 产物，全部 JS 的 gzip 合计	2,435,078 B	1,732,078 B	28.9%
桌面 Vite 产物，JS 文件数	143	48	—

组件路径包含脚本执行、Maka Theme 与 MermaidDiagram 挂载、SVG 完成及下一动画帧。两条路径都不测量网络传输或 Electron 启动。以上是本机 macOS arm64 结果（12 个逻辑核心、Node 24.22.1、无 CPU 降速），不代表性能保证。

原始数据 · Vite 产物大小 · 基准脚本（未包含在本 PR 中）

实现细节

• 用 Pele 替换 Mermaid.js 及其依赖树，复用 Pele 内置的缩放、平移和随宽度调整的布局。
• 将 Maka 主题变量映射到 Pele 的 CSS 变量，切换主题时直接更新已挂载 SVG 的样式。
• 保留全屏、源码查看与复制，以及 PNG 复制。PNG 导出会解析主题颜色，并嵌入随应用分发的字体。
• 保留 SVG 清洗、禁用链接与图片、源码长度与连线数量限制、自动渲染预算，以及实例间 SVG ID 隔离。渲染失败后仍显示可复制的源码。
• 容器停止调整大小后才重新布局图表，且仅当容器比图表的自然布局更窄、图表会随宽度改变布局时才这样做。在此之前，由 CSS 缩放已挂载的 SVG。
• 每份图表源码只缓存一个渲染结果。重新布局会替换该缓存，因此调整窗口大小不会挤掉其他图表的缓存。
• 在随应用分发的第三方声明中保留 Pele 的 MIT 和内嵌 Lucide ISC 许可证全文。

Pele 使用自己的布局与样式。此次迁移不承诺与 Mermaid 输出完全一致，也不承诺兼容全部 Mermaid 扩展。

Fixes #6257

审阅重点

• 根配置中的 lib。 tsconfig.base.json 现在设置了 lib: ["ES2024", "DOM", "DOM.Iterable"]。此前 packages/ui 获得 ES2023 及以上的类型库纯属偶然：Mermaid 的类型文件导入了未声明依赖的 type-fest，该导入解析到提升到根目录的 type-fest@0.13.1，其 source/basic.d.ts 含有 /// <reference lib="esnext"/>。移除 Mermaid 后，findLastIndex 和 Promise.withResolvers 会在 packages/ui 和 apps/desktop/tsconfig.storybook.json 中类型检查失败。ES2024 是能通过检查的最低设置。
• 依赖发布时间。 pele@1.9.0 于 2026-10-22 发布，带有来自 GitHub Actions 的 npm 来源证明。该项目的首个版本发布于 2026-10-06。
• CI 检查约束。 颜色令牌检查现在明确允许两个字体声明，仍拒绝颜色及未知变量。KaTeX 字体打包测试从 @maka/ui 解析 CSS，不再依赖原先由 Mermaid 带入根目录的间接依赖。

标记可审阅前的待办

• 定位并对比 ResizeObserver loop completed with undelivered notifications 警告。同一个 Workhub 会话在上游 ac1d288c4 与本分支各出现两次警告，页面均没有 Mermaid 图表。调用追踪确认触发路径为 virtua 列表尺寸变化时的同步更新；此次复现的问题在迁移前已存在。尚未测试发布版。
• 在 7eb192f71 上定位 WebContents MaxListenersExceededWarning 的受控复现路径。使用原开发数据重启并快速切换历史会话 20 次，均未触发警告，监听器峰值为 10。通过 maka.sessions.subscribeEvents 增加两个只读订阅后，destroyed 监听器从 9 增至 11，在 RuntimeHostSessionObservationRegistry.observe() 注册监听器时触发警告；取消订阅后回到 9。该注册与清理路径相较迁移前基线 a4b5f86c5 没有改动。此结果确认了监听器数量超过默认阈值的触发路径，尚不能据此认定存在持续泄漏。最初触发警告的 UI 操作序列及上游运行时对照仍未验证。
• 测量窗口在显示多张大图时持续调整大小的渲染开销。现在停止调整大小后才重新布局，结果见“验证”一节。

AI 使用

• 生成式工具没有实质性贡献
• 生成式工具有实质性贡献

工具与范围：OpenAI Codex 在用户指导下完成迁移、测试、性能基准、视觉采集及最初的说明。Devin 审阅了改动，实现了按源码缓存渲染结果、对应的回归测试和输出安全测试，删除了基准脚本和一条未使用的 CSS 规则，并用英文示例重新采集了审阅截图与录屏。审阅后，Codex 实现了窄容器下的重新布局判断，Devin 让重新布局等到停止调整大小后再执行，并编写了对应的回归测试。相关提交带有 Generated-by 标记，压缩合并时请保留。

检查清单

• 测试覆盖改动，移除相应实现后会失败
• lint、格式检查、类型检查及受影响测试均在本地通过

此 PR 是否改变行为？

• 是，已在上方概述中说明
• 否

[jackwener]

Review result: blocked

Reviewed exact head 28a20fbd673bb580ec30adb5382c13403c108d6f. The migration delivers a substantial first-render and bundle-size improvement, but I found one user-path performance blocker and one attribution blocker.

[P1] Coalesce width-driven Pele renders instead of rendering every ResizeObserver tick

packages/ui/src/mermaid-diagram.tsx:269-337 copies every observed pixel-width change into component state and then synchronously calls Pele's render(). Every mounted diagram therefore performs layout independently for every width event while the window or a split pane is being dragged. There is no debounce, width bucketing, resize-settle gate, or cross-instance coalescing.

I reproduced this through the built Maka theme, Markdown, and MermaidDiagram components in Chromium using a normal Markdown document with three automatically rendered 200-edge flowcharts. Each diagram was below the 4,000-character source limit, their combined source was below 8,000 characters, and three diagrams is the current automatic-render maximum. Ten viewport changes alternating between 900 and 500 pixels produced 66 SVG child mutations and ten renderer-thread long tasks of 241–446 ms; the resize loop took about 5.75 seconds. Current main produced no SVG mutations and no long tasks during the same resize sequence. Suppressing only this component's ResizeObserver callbacks on the PR head removed the repeated stalls, leaving a single 50 ms task, so the width-driven render path is causal.

Please debounce or bucket width changes and render once after resizing settles while retaining the previous SVG, then add a deterministic browser regression that bounds the render count during a resize burst. This needs to be fixed before approval because ordinary resizing with several valid diagrams repeatedly blocks the renderer thread and contradicts the PR's performance goal.

[P2] Record Devin in the generated-code trailer

The PR body says Devin materially implemented the per-source render cache, its regression test, and the output-safety tests. The sole commit currently carries only Generated-by: Codex. CONTRIBUTING.md requires every affected commit to carry a Generated-by: <tool> trailer for material generated content. Please amend the commit so it retains trailers for both applicable tools.

Verification

• The production-component comparison independently confirmed the intended benefit: initial rendering of the same three 200-edge diagrams improved from about 9.37 seconds on current main to about 1.57 seconds on this head.
• The corresponding minimal component bundle fell from 3,761,952 bytes / 1,040,888 bytes gzip on main to 1,001,300 bytes / 346,175 bytes gzip on this head.
• @maka/core and @maka/ui builds passed, as did the 821-test UI suite, all three real-browser diagram tests, eight Desktop palette/Vite contract tests, Biome, and git diff --check.
• I also checked representative flowchart, sequence, state, pie, timeline, ER, git graph, mindmap, C4, packet, architecture, directive, frontmatter, classDef, and linkStyle syntax, plus malicious script/HTML/link/style inputs.
• Exact-head hosted CI, dependency audit, packaging, and platform checks are green. The PR is currently open, non-draft, and mergeable against current main.

I found no other P0–P2 issues in the reviewed head.

---

Automated review notice: This comment was posted by an automated review agent operated by jackwener. It is not an independent human review and does not replace one.

[Colafornia]

Response to review 5646841614 (head is now 7eb192f71).

Resize relayout — accepted, graded P2 rather than P1. The mechanism is real. At 28a20fbd6, every ResizeObserver width went into layout, and each change re-ran Pele's synchronous render() on every mounted diagram. In Chromium, with three automatically rendered 200-edge flowcharts, a 40-step drag produced 225 SVG mutations. The stall is transient, though: it ends when resizing stops and persists nothing, so I graded it P2. It is fixed in this PR because #6257 requires narrow containers to adapt "without repeated layout changes".

• Fix (7eb192f71): MermaidDiagram applies a width change only after resizing settles (MERMAID_RESIZE_SETTLE_MS, 150 ms). The first measurement is still immediate, and the mounted SVG keeps scaling with CSS until then. Relayout is also skipped when the container is at least as wide as the diagram's natural layout, or when the diagram does not adapt to width, the same as Pele's mount(). The same drag now produces 12 mutations, a single relayout per diagram.
• Not adopted: width bucketing, which would change layout output, and cross-instance coalescing, which adds shared state. Settling each instance already gives the minimum of one render per diagram.
• Regression test: re-lays out once after a resize burst settles and keeps the diagram meanwhile in mermaid-render-cache.test.tsx drives real Pele through the component and fails without the fix. It lives in the test:dist suite rather than mermaid.browser.test.mjs, because CI does not run the browser script.

Generated-by trailer — accepted. 4ae8d4caa (formerly 28a20fbd6, same tree) now carries both Generated-by: Codex and Generated-by: Devin, and the fix commit carries both as well.

Verification: @maka/ui build, test:dist 823/823, mermaid.browser.test.mjs 3/3, Biome, knip (packages/ui), git diff --check.

Posted by Devin (AI agent) on behalf of the PR author.

[jackwener]

Re-review result: no findings

Reviewed exact head 7eb192f71bfd9d11c8bfe5c309bffd9b34b4dd56. Both findings from my review of 28a20fbd673bb580ec30adb5382c13403c108d6f are closed, and I found no new P0–P2 issues.

• The resize path now retains the mounted SVG, coalesces observer notifications behind a 150 ms settle interval, and only runs Pele again when the final container is narrower than the diagram's natural layout and the diagram actually adapts to width. In a built production-component Chromium probe with three automatically rendered 200-edge flowcharts, a 40-step width drag produced one three-diagram relayout batch instead of one batch per step. Removing only the settle gate from the generated component made the new regression fail and exposed 17 intermediate SVG replacements.
• The two commits now retain both applicable Generated-by: Codex and Generated-by: Devin trailers, matching the PR's updated AI-use declaration.

The head is directly based on current main (a4b5f86c5256044d063a62045754e4a3411a8426), so the reviewed tree is also the current merge result. @maka/core and @maka/ui builds passed, as did the complete 823-test UI suite, all three real-Chromium Mermaid tests, eight Desktop palette/Vite contract tests, changed-file Biome, and git diff --check. The production probe rendered all three large diagrams initially in about 1.34 seconds. Exact-head hosted checks are successful.

No native Windows clipboard export, packaged Electron restart, or exhaustive compatibility pass over every Mermaid extension was repeated in this incremental review.

---

Automated review notice: This comment was posted by an automated review agent operated by jackwener. It is not an independent human review and does not replace one.

[jackwener]

Approval

Approved exact head 7eb192f71bfd9d11c8bfe5c309bffd9b34b4dd56. I found no unresolved P0–P2 issues.

Both findings from the earlier review are closed. The resize path keeps the mounted SVG visible, coalesces a burst behind a 150 ms settle interval, and only lays out again when the final container is narrower than the diagram's natural layout and the diagram adapts to width. A built production-component Chromium probe with three automatically rendered 200-edge flowcharts reduced a 40-step drag to one three-diagram relayout batch; removing the settle gate made the regression fail with 17 intermediate SVG replacements. Both commits also retain the applicable Generated-by: Codex and Generated-by: Devin trailers.

The head is directly based on current main a4b5f86c5256044d063a62045754e4a3411a8426, so no separate merge delta exists. @maka/core and @maka/ui builds, the complete 823-test UI suite, all three real-Chromium Mermaid tests, eight Desktop palette/Vite contract tests, changed-file Biome, and git diff --check passed. The exact-head required test check and all applicable hosted checks are successful. The PR is open, non-draft, mergeable, and has no review threads.

No merge or contributor-source change was performed.

---

Review notice: This review was prepared by an automated review agent operated by jackwener and is published at the direction of colafornia, who has read these findings and is the human accountable for them.

[me2seeks]

Reviewed exact head 7eb192f. No blocking findings.

Checked the Pele migration, width-settle relayout and per-source cache, instance SVG references, untrusted-output limits, theme handling, PNG export, dependency changes, and packaged attribution. The resize path retains the mounted SVG during a burst and coalesces layout after 150 ms.

Independent local validation in an isolated checkout: core/UI builds; all 823 UI tests; all 3 real Chromium diagram tests; 2 Desktop Vite/KaTeX contract tests; 2 third-party notice tests; git diff --check. Exact-head hosted checks are successful (the Eval job is skipped), the PR is mergeable, and there are no review threads.

Native OS clipboard integration and exhaustive Mermaid extension compatibility were not independently tested.

Automated review prepared and submitted by Codex at me2seeks's explicit direction. This does not replace the independent human review required by CONTRIBUTING.md.