@KyrieCheungYep: 推荐一个 PPT 制作 Skill：ultimate-ppt-master-skill，它比较贴近真实工作时候需要做 PPT 的场景，功能和亮点如下： 1、先问清楚需求 很多 AI PPT 工具上来就做页，结果做出来全是空话。 PPT M…
摘要
推荐一个开源PPT制作工具ultimate-ppt-master-skill，它通过先明确需求、整理资料结构、生成可编辑PPTX或Web Deck、配图生成和成稿审阅等流程，贴近真实办公场景，解决AI PPT生成中常见的空话和不可编辑问题。
推荐一个 PPT 制作 Skill：ultimate-ppt-master-skill，它比较贴近真实工作时候需要做 PPT 的场景，功能和亮点如下：
1、先问清楚需求
很多 AI PPT 工具上来就做页，结果做出来全是空话。
PPT Master Skill 会先追问几个关键问题：
给谁看？什么场景？汇报、路演、培训还是方案？
偏正式、科技感、咨询风，还是老板喜欢的“稳一点”？
先把边界问清楚，后面才不容易跑偏
2、整理资料和内容结构
它不只吃一句 prompt。
可以围绕本地资料、项目文档、原始素材去整理信息，先把内容顺一遍，再决定每一页讲什么。
这一步很关键，因为 PPT 最难的地方在于：
这一页到底要表达什么？
上一页和下一页之间有没有逻辑？
老板看到这里会不会问“所以呢”？
3、生成可编辑 PPTX
它的目标不是给你一张死图，也不是只生成一个看起来很炫的网页。
它可以产出可编辑的 PPTX，后面还能继续改文字、换图、调版式。
这点对真实办公很重要。
因为公司里的 PPT 很少是一版过，基本都要被改 3 轮、5 轮，甚至被不同人来回接手
4、也可以先做 Web Deck
有些时候，先做成 Web Deck 更方便。
比如要快速看结构、看视觉方向、看整套表达是不是顺。
等方向定了，再转成 PPTX。
这比一开始就在 PPT 里死磕排版舒服很多
5、配图也纳入流程
如果某一页需要视觉图、概念图、场景图，它可以接 GPT 生图。
这样就不用一边写 PPT，一边满世界找图。
尤其是做产品介绍、AI 方案、战略汇报时，很多图本来就很难在素材网站里找到现成的
6、做完之后还会审阅
这个 Skill 比较有意思的一点，是它把“检查成稿”也放进流程里。
不是生成完就丢给你，而是会渲染出来看：
文字有没有挤在一起？
图片有没有压住标题？
页面是不是空得离谱？
风格有没有前后不一致？
这份东西拿去汇报，会不会一眼像 AI 糊出来的？
7、保留审计记录
它还强调审计留痕。
也就是这份 PPT 为什么这样做、素材从哪来、哪些地方改过，尽量能追得回去。
这对团队协作很有用。
不然很多 AI 生成的东西，第一眼还行，真要改的时候根本不知道从哪下手
如果你刚好要做 PPT，去参加答辩或者汇报，可以试试看这个 Skill！
查看缓存全文
缓存时间: 2026/06/26 12:11
推荐一个 PPT 制作 Skill：ultimate-ppt-master-skill，它比较贴近真实工作时候需要做 PPT 的场景，功能和亮点如下：
1、先问清楚需求 很多 AI PPT 工具上来就做页，结果做出来全是空话。 PPT Master Skill 会先追问几个关键问题： 给谁看？什么场景？汇报、路演、培训还是方案？ 偏正式、科技感、咨询风，还是老板喜欢的“稳一点”？ 先把边界问清楚，后面才不容易跑偏
2、整理资料和内容结构 它不只吃一句 prompt。 可以围绕本地资料、项目文档、原始素材去整理信息，先把内容顺一遍，再决定每一页讲什么。 这一步很关键，因为 PPT 最难的地方在于： 这一页到底要表达什么？ 上一页和下一页之间有没有逻辑？ 老板看到这里会不会问“所以呢”？
3、生成可编辑 PPTX 它的目标不是给你一张死图，也不是只生成一个看起来很炫的网页。 它可以产出可编辑的 PPTX，后面还能继续改文字、换图、调版式。 这点对真实办公很重要。 因为公司里的 PPT 很少是一版过，基本都要被改 3 轮、5 轮，甚至被不同人来回接手
4、也可以先做 Web Deck 有些时候，先做成 Web Deck 更方便。 比如要快速看结构、看视觉方向、看整套表达是不是顺。 等方向定了，再转成 PPTX。 这比一开始就在 PPT 里死磕排版舒服很多
5、配图也纳入流程 如果某一页需要视觉图、概念图、场景图，它可以接 GPT 生图。 这样就不用一边写 PPT，一边满世界找图。 尤其是做产品介绍、AI 方案、战略汇报时，很多图本来就很难在素材网站里找到现成的
6、做完之后还会审阅 这个 Skill 比较有意思的一点，是它把“检查成稿”也放进流程里。 不是生成完就丢给你，而是会渲染出来看： 文字有没有挤在一起？ 图片有没有压住标题？ 页面是不是空得离谱？ 风格有没有前后不一致？ 这份东西拿去汇报，会不会一眼像 AI 糊出来的？
7、保留审计记录 它还强调审计留痕。 也就是这份 PPT 为什么这样做、素材从哪来、哪些地方改过，尽量能追得回去。 这对团队协作很有用。 不然很多 AI 生成的东西，第一眼还行，真要改的时候根本不知道从哪下手
如果你刚好要做 PPT，去参加答辩或者汇报，可以试试看这个 Skill！
kdnsna/ultimate-ppt-master-skill
Source: https://github.com/kdnsna/ultimate-ppt-master-skill
Ultimate PPT Master - v5.3 Best-Effect Brief Enhancer
Local-first AI presentation production for real office work: first expand short prompts into a best-effect brief, then turn unclear requests, messy source material, or structured briefs into editable PPTX or magazine-style Web Decks with a Guizang-like fixed-style fallback, Visual Brief tags, source confidence, delivery scorecards, official/IP asset boundaries, Codex/GPT no-text visuals, rendered review, and formal delivery audits.
v5.3.0 · English · 中文 README · Docs · v5.3 Release Notes · Agent Skill
Open Web Experience · Benchmark Wall · v5.3.0 Notes · Agent Bridge · Skill Market Distribution
Best Results Prompt
Use this prompt when calling the skill from Codex, ChatGPT, Claude Code, Hermes, OpenClaw, Cursor, or any Agent that can read this repo:
Use $ultimate-ppt-master to expand my short request into a best-effect brief first. If my prompt is extremely thin, use the Guizang-like Magazine Web Deck fixed style by default; if I explicitly need a formal editable deck, use PPTX and keep the same quality checks.
What v5.3 does before production:
| User input | v5.3 behavior |
|---|---|
| Only a topic or one sentence | Create an Auto-expanded brief, then use Guizang-like Magazine Web Deck fixed style unless the user explicitly asks for formal editable PPTX. |
| “Make me a PPT about X” with no sources | Treat it as an extremely thin prompt; default to an 8-page Style A editorial/e-ink Web Deck so the first result is stable and visually strong. |
| Formal report, government/finance/training, or editable file request | Use editable PPTX, Microsoft YaHei, formal-business quality gate, source confidence, official/IP asset plan, and the same best-effect brief record. |
| Source files or a detailed brief | Preserve the user’s intent, write assumptions, and proceed through the normal expectation-fit and quality checks. |
The point is simple: the user should not need to know the perfect prompt. The Agent must first turn a rough instruction into the production brief that would normally lead to the best result.
Why Teams Use It
Most AI PPT tools can make something that looks finished. Office teams need something they can trust, edit, audit, and revise, even when the first user request is only “make me a PPT”.
| Need | What Ultimate PPT Master gives you |
|---|---|
| Make short prompts useful | v5.3 adds a Best-Effect Brief Enhancer so the Agent expands vague requests before choosing route, style, page rhythm, and assumptions. |
| Get a strong first version from little input | An extremely thin prompt defaults to Guizang-like Magazine Web Deck fixed style instead of a bland generic PPT or another long interrogation. |
| Make a real PowerPoint when needed | Formal, editable, government, finance, training, and report requests still go to editable PPTX with Microsoft YaHei and formal-business checks. |
| Avoid expectation drift | v5.2 turns the clarity gate into a contract: tags, guided answers, source confidence, reference style, delivery scorecard, and feedback taxonomy are all written into the handoff. |
| Ask the right questions | The agent clarifies audience, scenario, purpose, source status, core message, page count, visual style, asset rules, output format, and compliance boundaries before production when those answers matter. |
| Use better AI visuals | Codex/GPT image generation is treated as a composed visual engine for no-text backgrounds, support scenes, and micro-assets, not element stacking. |
| Keep brands safe | Official/IP marks such as logos, campaign names, cards, QR codes, and partner marks require official-source, user-provided, text-lockup fallback, or replacement status. |
| Keep PowerPoint editable | PPTX output keeps real text, shapes, tables, charts, notes, logos, QR codes, and traceable image sources. |
| Avoid “small PPT” syndrome | Microsoft YaHei defaults, 16:9 spacing guardrails, title/body scale, card-count limits, and layout variety are locked before final assembly. |
| Review before revising | Rendered review writes findings, a safe repair plan, and a user-approved revision-brief.md for the second pass. |
| Stay local-first | The Bridge writes project files on localhost; private source material is not uploaded unless the user explicitly chooses that route. |
Who It Is For
| Audience | Typical work | Why v5.3 fits |
|---|---|---|
| Office teams | Leadership updates, work reports, training decks, project reviews, and sales enablement decks. | Handles vague one-line requests by auto-expanding a best-effect brief before production. |
| Finance, government, and enterprise users | Formal materials that need careful wording, brand boundaries, and traceable source claims. | Keeps official/IP assets documented and blocks fake logo-like placeholders for external release. |
| Consultants and internal strategy teams | Structured narratives, industry scans, solution proposals, and executive summaries. | Builds a single project-brief.json contract before page generation so structure, audience, and style stay aligned. |
| Event and brand teams | Keynotes, campaign decks, culture-tourism decks, product showcases, and public-facing visuals. | Uses Codex/GPT image generation for composed no-text visuals while the business message stays editable. |
Typical Scenarios
| Request | v5.3 default behavior |
|---|---|
| “Make me a PPT.” | Create bestEffectBrief; if it is extremely thin and not explicitly formal/editable, produce the Guizang-like Magazine Web Deck fixed style. |
| “Make this source into a PPT.” | If the source has no audience or purpose, ask for scenario, audience, and desired outcome first; then create editable PPTX by default. |
| “Use a stronger visual style.” | Read the source, name a theme art direction, generate no-text support visuals when useful, then keep titles, numbers, tables, and charts editable. |
| “This is for an external brand or public event.” | Require an official/IP asset plan for logos, campaign marks, cards, QR codes, partner marks, and any generated imagery. |
| “I need to revise it after review.” | Render the deck, write findings, create a safe repair plan, and produce revision-brief.md only after confirmation. |
What You Get
| Deliverable | What is inside | Why it matters |
|---|---|---|
bestEffectBrief | Prompt quality, auto-expanded brief, recommended route, Extreme Thin Prompt Fallback state, fixed style, and assumptions. | Agents no longer use the user’s rough sentence as the production brief. |
| Editable PPTX | Real text boxes, shapes, charts, tables, speaker notes, and editable brand lockups where safe. | The deck can be handed to a PowerPoint user, not only admired as screenshots. |
| Web Deck | A single-file browser presentation for talks, showcases, and magazine-style storytelling. | Useful when visual rhythm and sharing matter more than PowerPoint editing. |
project-brief.json | briefMode, visualBrief, guidedBrief, expectationFit, sourceConfidence, deliveryScorecard, referenceStyle, feedbackLoop, and confirmationBrief in one handoff contract. | The system knows which signals came from the user, which came from tags, which are assumptions, and what to fix if the user is unsatisfied. |
| Visual Brief | Scenario, audience, purpose, content status, visual style, density, asset strategy, output preference, pasted background, links, and extra requirements. | Web users can express a rich PPT request without filling a long form. |
| Guided Brief | Codex-collected answers for scenario, audience, purpose, core message, sources, page count, outline, style, assets, output, must-include, and must-avoid. | Chat users are walked through the missing essentials before serious production. |
| Expectation Fit | Green/yellow/red risk level, source confidence, missing signals, assumptions, delivery scorecard, and readiness for production. | High-risk vague briefs are visible before the deck is generated. |
| Source and asset records | source-map.json, image_sources.json, image_prompts.json, and official/IP fallback notes. | Reviewers can see what was sourced, generated, replaced, or left for authorization. |
| Review package | Rendered previews, findings, repair candidates, and quality status. | Revisions start from observed issues instead of vague “make it better” loops. |
What v5 Changes
v5.0.0 made the product a delivery-defaults system. v5.1.0 added staged intake for unclear requests. v5.2.0 turned those signals into an expectation-fit contract. v5.3.0 adds the layer that was missing for real Agent use: short instructions are auto-expanded into a best-effect brief before route selection, and topic-only prompts get a fixed high-quality Guizang-like Web Deck default.
| v5 default | Practical effect |
|---|---|
| Best-Effect Brief Enhancer | Every Agent run starts by writing bestEffectBrief: prompt quality, auto-expanded brief, assumptions, route, fixed fallback, and user-visible caveats. |
| Extreme Thin Prompt Fallback | Topic-only or one-line prompts default to Guizang-like Magazine Web Deck fixed style unless the user explicitly needs formal editable PPTX. |
| Expectation-fit contract | project-brief.json and quality-report.json now carry sourceConfidence, deliveryScorecard, referenceStyle, feedbackLoop, failureTaxonomy, confirmationBrief, and imageAcceptance. |
| Editable PPTX first | Formal reports, consulting decks, finance/government material, and normal “make PPT” requests go straight to PowerPoint-ready output after the brief is clear enough. |
| Visual Brief Builder | Web users pick diverse tags and paste background material, meeting notes, leadership requirements, links, and special constraints. |
| Codex Guided Intake | If the request is vague, Codex asks by stage: audience and scenario, content source and core message, page structure, visual style, asset policy, output format, and compliance boundaries. |
| Unified brief contract | Web tags and Codex interview answers both flow into project-brief.json, so Bridge, Desktop Worker, audits, and prompts read the same intent. |
| Expectation Fit gate | Green means production-ready; yellow means proceed with caveats; red means clarify or explicitly draft with assumptions. |
| Feedback taxonomy | If the user is unhappy, the next revision classifies the reason first: brief mismatch, source gap, style mismatch, visual-density mismatch, asset/IP boundary, or format mismatch. |
| Codex-first generated visuals | Generated imagery supports the page as no-text layers or reusable micro-assets; body content remains editable. |
| Official/IP asset plan | Deterministic marks are sourced or documented. Fake logo-like placeholders are blocked for external release. |
| Theme art direction | After reading the source, the agent names a subject-fit art concept, such as 山海交汇 烟火同行 for cultural tourism, then carries it through cover/tail pages and title treatment unless the deck is a serious report. |
| Typography and layout system | Microsoft YaHei is the default CJK office font; body text normally stays 18-24px with clear title/body scale and safe margins. |
| Formal-business audit | design_spec.md, spec_lock.md, design-quality-report.md, asset manifests, and PPTX/Web artifacts are checked before delivery. |
v5 Delivery Standard
| Area | Standard |
|---|---|
| Best-effect brief first | Agents must expand short requests before production and record the inferred assumptions in bestEffectBrief. |
| Extremely thin prompts | Default to Guizang-like Magazine Web Deck fixed style unless the user explicitly requests formal editable PPTX. |
| Right questions first | When a missing answer can change the deck, Codex pauses for staged intake instead of silently guessing. |
| Draft escape hatch | If the user says “draft with defaults” or “先做一版草稿”, the system may proceed with assumptions and must record them. |
| PPTX editability | Business content stays editable. Full-slide raster images are reserved for intentional no-text backgrounds, illustrations, or browser-only Web Deck moments. |
| Image generation | Codex/GPT visuals must be composed scenes or support assets with no embedded text, no fake logos, and prompt records. |
| Official assets | Logos, cards, QR codes, campaign IP, and partner marks must be official-source, user-provided, text-lockup fallback, or needs-authorized-replacement. |
| Typography | Microsoft YaHei is the default Chinese office font. Page titles, body text, captions, numbers, and footers must have visible hierarchy. |
| Layout | 16:9 pages use safe margins, card-count limits, readable body scale, and layout variety across summary, process, comparison, data, and closing pages. |
| Handoff | The final package should include the PPTX or Web Deck, rendered review, quality report, source records, expectation caveats, and any asset caveats. |
Product Loop
The production path is intentionally narrow. It should feel like a senior PPT operator turning messy material into a deliverable, not like a user configuring an engine.
User request or source material
-> Best-Effect Brief Enhancer
-> clarity gate / Extreme Thin Prompt Fallback
-> Web Visual Brief tags or Codex Guided Intake
-> project-brief.json with bestEffectBrief + expectationFit
-> official/IP asset plan
-> page roles + recipes + editability/raster policy
-> Codex/GPT no-text visual assets when useful
-> editable PPTX or magazine Web Deck
-> rendered review findings
-> safe repair plan + revision-brief.md
-> formal delivery audits
Core artifacts in plain language:
project-brief.json: the production brief; v5.3 recordsbestEffectBrief,briefMode,visualBrief,guidedBrief, andexpectationFit.design_spec.md: human-readable design contract for visual direction, page roles, typography, assets, expectation risks, and completion caveats.spec_lock.md: machine-readable execution lock for page recipes, visual layers, raster policy, brand assets, expectation contract, and aesthetic checks.storyboard.json: DeckIR page map with page roles, recipe IDs, evidence refs, raster policy, and editability targets.source-map.json: traceable source claims used by the deck.images/image_sources.jsonandimages/image_prompts.json: official/public/generative asset provenance and prompt records.review-findings.json: rendered-review issues with severity, risk level, repair target, and suggested command.repair-plan.json: low-risk repair candidates; default path is dry-run.revision-brief.md: second-generation brief created only after explicit safe apply.quality-report.json: merged delivery, planning, expectation, and rendered-review status.
DeckIR is the page map. Page recipes are the layout instructions for each slide. Raster policy is the rule that decides which visual layers may become pixels and which content must remain editable.
60-second quickstart
git clone https://github.com/kdnsna/ultimate-ppt-master-skill.git
cd ultimate-ppt-master-skill
npm run setup
npm run doctor
npm run bridge
Then open the Web Experience. Choose a recommended Visual Brief combination, add tags, paste background material, connect locally, and deliver the handoff project.
| Need | Best route | Output |
|---|---|---|
| Only a topic or one-line request | Guizang-like fixed Web Deck | Stable 8-page Style A editorial/e-ink HTML deck. |
| Formal report, consulting deck, training deck, editable business material | Editable PPTX | PowerPoint deck with editable text, shapes, charts, tables, notes, and quality checks. |
| Talk, showcase, demo day, editorial presentation | Web Deck | Single-file browser deck with stronger visual rhythm and shareable HTML. |
| Both formal handoff and browser preview | Dual delivery | Separate PPTX and Web projects with shared source and aligned structure. |
| Vague request with high expectations | Codex Guided Intake | Staged questions, confirmed brief, then PPTX/Web production. |
Real Production Flow
| Step | What the agent should do |
|---|---|
| Expand first | Convert the raw request into bestEffectBrief, including route, style, page count, assumptions, and fixed fallback state. |
| Judge clarity | Decide whether the request is production-ready, yellow-risk, or too vague. |
| Ask or tag | Use Visual Brief tags on Web; use Codex Guided Intake in chat when essential context is missing. |
| Confirm brief | Summarize target, audience, content frame, page count, style, assets, output format, and assumptions before production. |
| Read the source | Extract the core message, constraints, figures, and named assets before proposing page structure. |
| Plan pages | Assign every page a role: cover, summary, evidence, process, comparison, data, case, route, call-to-action, or closing. |
| Build assets | Use official/user assets where required; generate no-text visuals only where they improve the message. |
| Assemble | Keep business text, charts, tables, and labels editable; use Microsoft YaHei and a restrained office hierarchy by default. |
| Review | Render the output, inspect layout and asset risks, write findings, then repair only with a safe plan. |
Capability Matrix
| Layer | Release | What it protects |
|---|---|---|
| Best-effect brief and fixed fallback | Release Notes - v5.3.0 | Auto-expanded brief, extreme-thin prompt fallback, Guizang-like Magazine Web Deck fixed style, and PPTX exception path. |
| Guided intake and expectation fit | Release Notes - v5.1.0 | Visual Brief tags, Codex staged questions, unified project-brief.json, and readiness risk before generation. |
| Expectation-fit contract | Release Notes - v5.2.0 | Source confidence, reference-style choices, delivery scorecard, feedback taxonomy, confirmation brief, and image acceptance. |
| Delivery defaults | Release Notes - v5.0.0 | Default PPTX route, one delivery brief, official/IP asset handling, Microsoft YaHei layout scale, and Codex-first image generation. |
| Rendered review and repair brief | v4.3 Rendered Review Loop | Review after rendering, propose low-risk repairs, generate revision-brief.md only after confirmation. |
| AI planning | DeckIR AI Planning Workflow v4.2 | scripts/ai_storyboard.py, storyboard.json, evidence refs, editability targets, and no-key fallback. |
| Simplified Web console | Simplified Web Console v4.1 | Four-step console, one primary next action, grouped previews, lower first-screen complexity. |
| Hybrid-editable generation | Hybrid-Editable Visual Workflow v4.0 | page recipes, no-text generated visual layers, editable PPTX body content, formal raster policy. |
| Public proof surface | Quality Workbench v2.5 | Benchmark Wall, synthetic proof packs, Design Doctor reporting, and release checks. |
Historical release notes: v5.2.0, v5.1.0, v5.0.0, v4.3.0, v4.2.0, v4.1.0, v4.0.0, v3.0.0.
Proofs
The Benchmark Wall keeps public synthetic proof packs visible before anyone installs the project.
| Proof | Link |
|---|---|
| v5.3 best-effect brief enhancer | Release Notes - v5.3.0 |
| v5.2 expectation-fit contract | Release Notes - v5.2.0 |
| v5.1 guided-intake release | Release Notes - v5.1.0 |
| v5 delivery-defaults release | Release Notes - v5.0.0 |
| Rendered review release | Release Notes - v4.3.0 |
| DeckIR AI planning pack | Release Notes - v4.2.0 |
| Hybrid-editable release | Release Notes - v4.0.0 |
| Stable proof matrix | Quality Workbench v2.5 |
| Skill marketplace readiness | Skill Market Distribution |
Use As Agent Skill
Copyable marketplace prompt:
Use $ultimate-ppt-master to expand my short request into a best-effect brief first. If my prompt is extremely thin, use the Guizang-like Magazine Web Deck fixed style by default; if I explicitly need a formal editable deck, use PPTX and keep the same quality checks.
Expert users can install the Skill directly:
bash -lc 'set -e; dir="$HOME/.codex/skills/ultimate-ppt-master"; if [ -d "$dir/.git" ]; then git -C "$dir" pull --ff-only; else git clone https://github.com/kdnsna/ultimate-ppt-master-skill.git "$dir"; fi; cd "$dir"; npm run setup'
Guide: Agent Setup. Local connector: Agent Connect Bridge.
Repository Map
| Area | Use it for |
|---|---|
SKILL.md and references/ | Agent behavior, guided intake, route selection, asset sourcing, visual generation, and delivery rules. |
apps/web and apps/desktop | Web Experience, Visual Brief Builder, local bridge handoff surface, and desktop worker integration. |
apps/bridge | Local project-brief.json creation, expectation-fit handoff, source parsing, and agent task files. |
scripts/ | Audits, release checks, rendered-review repair, provider setup, and repository maintenance. |
templates/ | Design spec and spec lock references used by formal-business delivery. |
docs/ | User guides, release notes, quality workflows, strategy docs, and Chinese documentation. |
tests/ | Release integrity, audits, worker behavior, bridge behavior, and public-surface guarantees. |
Documentation Map
| Need | Read |
|---|---|
| Try the web front door | Web Experience |
| Connect browser, local files, and Agents | Agent Connect Bridge |
| Install and invoke the Skill | Agent Setup |
| Choose PPTX vs Web Deck vs Desktop | Choosing a Workflow |
| Configure provider keys locally | Model and Provider Setup |
| Review v5.2 expectation contract | Release Notes - v5.2.0 |
| Review v5.1 guided intake | Release Notes - v5.1.0 |
| Review v5 delivery defaults | Release Notes - v5.0.0 |
| Review v4.3 repair flow | Rendered Review Loop v4.3 |
| Apply rendered repair plan | scripts/apply_review_plan.py |
| Understand DeckIR AI planning | DeckIR AI Planning Workflow v4.2 |
| Understand the v4.0 visual contract | Hybrid-Editable Visual Workflow v4.0 |
| Audit visual recipes | scripts/audit_visual_recipes.py |
| Review release maintenance | Release and Maintenance |
| Debug setup or generation issues | Troubleshooting |
Full map: docs/README.md. Chinese map: docs/zh-CN/README.md.
Quality Gates
Run these maintainer checks before publishing a README, version, Skill, Web, or Desktop release:
npm run audit:docs
npm run audit:web-console
npm run audit:presets
npm run audit:quality
npm run audit:market
npm run test:node
npm run test:bridge
npm run test:worker
npm run build:web
npm run build:desktop
git diff --check
The README promise must stay tied to executable checks. If a capability is advertised here, it should have a doc, script, test, audit, or public proof artifact. Desktop binary packaging is not claimed by this README unless a separate packaging command is run and recorded for that release.
相似文章
@yaohui12138: 最近 guizang 、zarazhangrui 花叔等大佬纷纷开源 PPT Skill，我花了一天测试完了7个ppt skill项目，直接把结论分享给大家： AI 生成 PPT 丑的核心原因，不是 AI 能力不行，是你没给它正确的审美约…
作者评测了7个开源PPT Skill项目，指出AI生成PPT丑的核心原因是缺乏正确的审美约束系统，这些项目通过编译审美规则来提升AI生成PPT的设计质量，并分别介绍了各自的特点和适用场景。
@GoSailGlobal: 2026 做 PPT 不再用 PowerPoint。 5 种 AI 驱动的 PPT 风格，覆盖： · founder pitch · McKinsey 咨询风 · 横屏 HTML deck · 图片美学 deck · 全栈营销 slide…
文章对比了 25 种开源 AI PPT 生成技能工具，推荐了 guizang-ppt-skill 等集成 MCP 和 Claude Code 的自动化方案，涵盖咨询风、HTML deck 等多种风格。
@Gorden_Sun: 史上最强原生PPT Skill，更适合中国宝宝 使用我创建的这个Skill，一句话即可生成复杂、豪华、可编辑的PPT文件。试一次，要是生成的效果不让你震惊，你来打我。 几大特色： 1、能生成信息密度高、排版复杂、看起来高大上的PPT，也支…
GordenSun 发布了一个名为 GordenPPTSkill 的开源项目，能让 AI 助手（如 GPT、DeepSeek、Claude 等）一句话生成信息密度高、排版复杂的可编辑 PPT 文件，支持自动更新模板，适合国企和互联网大厂使用。
@aehyok: 最近刚好也在研究PPT。看了一姐的视频，使用Codex +Image2来做PPT，于是顺手找了几个star数比较多的PPT Skill，准备挨个都来试试看，到底那个效果更适合自己呢？ 1、https://github.com/op7418…
介绍了一款名为guizang-ppt-skill的AI Agent技能，用于生成单文件HTML横向翻页PPT、配图和封面，支持Claude Code和Codex等环境。
@aiwarts: https://x.com/aiwarts/status/2070137056144019720
开源了Humanize PPT，一个为演讲设计的HTML PPT技能。它通过AST（Audience, State, Transfer）逻辑重新编排大纲，配合下游渲染技能生成真正可用于演讲的PPT，解决信息分配不均、内容空洞等问题。