PPT Agent skill

Q: How do I install PPT Agent in Claude Code?

Run npx skills add sunbigfly/ppt-agent-skills. No separate server is needed: the skill works inside the agent and starts from a request to make a deck, for example "Prepare a 12-slide pitch deck".

Q: What happens at the first step?

The agent interviews you and records the answers in requirements-interview.txt: topic, page count, style, the sub-agent model and whether manual audit is needed. Then you choose a branch: web research or your own materials.

Q: Which formats does the result come in?

A web preview preview.html and two PPTX files. In the raster file slides are pictures for exact looks, in the vector file fonts stay editable. A delivery-manifest.json is written alongside.

A deck from a single sentence: seven stages, a sub-agent per slide, screenshot-based QA and PPTX export.

917 ★ on the repository
Last commit Jun 8, 2026

Skill

sunbigfly/ppt-agent-skillsAuthor:sunbigfly

Install

Works in: Claude Code
Codex
Koda
Coddy

In your terminal, with SkillFoxx CLI

npx skillfoxx add skills/ppt-agent

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

The skill launches sub-agents and Python and Node scripts, renders pages in a browser engine and writes files to `ppt-output/`.
On the research branch the agent goes online and gathers material on the deck topic.

This is third-party code. Review the repository files before installing.

What it does

PPT Agent turns a request such as "a 15-page deck on embodied AI trends, dark tech style" into a finished presentation. The main agent runs the pipeline as a state machine and writes nothing itself: every artifact is made by a separate sub-agent in an isolated context. First comes an interview, then web research or compression of your own materials, then a narrative outline, a global style and page-by-page assembly. For each slide a JSON contract is created first and checked by `planning_validator.py`, and only then is HTML rendered. The finished slide is screenshotted, and the model looks for text overflow and overlaps and rewrites the markup. The result lands in `ppt-output/runs/<RUN_ID>/`: a web preview and two PPTX files, a raster one for exact looks and a vector one where fonts stay editable.

Who it is for. Consultants, founders and teachers who need a structured deck and lack time for manual layout.

Good fit when

A pitch needs a fifteen-page deck built with web data research.
An internal report or document must become slides for management.
An existing deck looks dull and has to be re-laid out in another style.
Slides must reach PowerPoint with editable fonts instead of pictures.
A build was interrupted midway and the work must continue from the last finished slide.

Not a fit when

The agent cannot launch sub-agents with a chosen model, which the skill contract requires.
You need one quick diagram or chart rather than a multi-page deck.
You need only a text outline of a talk without slides: the pipeline always goes through to layout.

Example request

Make a 15-page deck on embodied AI trends in 2026, dark tech style, for an investor pitch.

Limitations

The skill launches many sub-agents and pages, so a run takes noticeable time and spends tokens. The model for sub-agents is named at the interview stage, and a launch without it is forbidden by the contract. The agent must be able to read images, otherwise the visual slide check does not work. GitHub does not recognise the `LICENSE` file automatically; the README states the MIT license. The documentation is in Chinese, with an English version in `README_EN.md`. The last commit was made in June 2026, version 4.1 came out in April.

How to disable. Delete the skill folder from the agent's skills directory. Run results live in `ppt-output/runs/` and are removed separately.

FAQ

How do I install PPT Agent in Claude Code?

Run `npx skills add sunbigfly/ppt-agent-skills`. No separate server is needed: the skill works inside the agent and starts from a request to make a deck, for example "Prepare a 12-slide pitch deck".

What happens at the first step?

The agent interviews you and records the answers in `requirements-interview.txt`: topic, page count, style, the sub-agent model and whether manual audit is needed. Then you choose a branch: web research or your own materials.

How does the skill check slides?

Each slide is screenshotted and the model looks for overflowing text and overlaps. When it finds them, a sub-agent rewrites the DOM and CSS structure instead of nudging margins. Scripts also check the HTML structure.

What if a build breaks off?

The skill needs no state file: it scans the folder and uses existing `outline.txt`, `style.json` and `slide-N.png` files to decide which step to resume from.

Which formats does the result come in?

A web preview `preview.html` and two PPTX files. In the raster file slides are pictures for exact looks, in the vector file fonts stay editable. A `delivery-manifest.json` is written alongside.

Discussion

No questions or reviews yet.

Skills

PPTAgent

A skill that builds presentations from HTML slides with rendering, per-slide visual review, and conversion to an editable PPTX

SkillExtended access5.2K

PPT Master

Editors’ pick

A skill that builds editable PPTX decks from PDF, DOCX, URLs and Markdown with native shapes, charts and animations

SkillExtended access74K

Codex PPT Skill

A skill that turns an article, report or notes into a deck of full-slide images and assembles them into a pptx

SkillExtended access7.7K

luobo-ppt

A skill that builds a deck from a topic or article, or restyles an old one from a reference image, keeping the text and editable slides

SkillExtended access452

PPT Agent Skill

ppt-agent-skill

A Claude Code skill: from one sentence a presentation in HTML and editable PPTX, 26 styles and 18 charts by the author's description

SkillMinimal access178

Humanize PPT

A skill for talks: outlines a deck around the audience's train of thought, picks images and video, and checks finished slides before you present

SkillExtended access992

Comparison with similar tools

Cards that share the main search query, and how they differ.

Tool	Official	Repository stars	Main difference
PPT Agent (this card)	no	917
PPT Master	no	73,972	builds editable PPTX from PDF, DOCX and links with real shapes; PPT Agent builds a deck from a request via an interview, sub-agents and screenshot QA
Guizang PPT Skill	no	29,346	produces a deck as one HTML file with a presenter mode; PPT Agent exports raster and vector PPTX

README in short

The README lists the key ideas: sub-agent isolation, screenshots as a check, recovery without a state file, separating data from layout and dual export. It gives the stage chain from P0 to P5, the artifact chain, the repository layout and four result screenshots.

README badge

Original SKILL.md

---
name: ppt-agent
description: 专业 PPT 演示文稿全流程 AI 生成助手。模拟顶级 PPT 设计公司的完整工作流（需求调研到资料搜集到大纲策划到策划稿到设计稿），输出高质量 HTML 格式演示文稿。当用户提到制作 PPT、做演示文稿、做 slides、做幻灯片、做汇报材料、做培训课件、做路演 deck、做产品介绍页面时触发此技能。即使用户只说"帮我做个关于 X 的介绍"或"我要给老板汇报 Y"，只要暗示需要结构化的多页演示内容，都应该触发。也适用于用户说"帮我把这篇文档做成 PPT"、"把这个主题做成演示"等需要将内容转化为演示格式的场景。英文场景同样适用："make a presentation about..."、"create slides for..."、"build a pitch deck"、"I need a keynote for..."。隐式意图也应触发："帮我把这个数据可视化一下给老板看"、"我需要一份能拿去路演的东西"、"把这个报告做得好看点能展示"、"beautify my existing PPT"、"redesign these slides"。改善或美化现有 PPT 也属于此技能范畴。
---

# PPT Agent v4.1 — 主控制台合同

## 1. 主 Agent 角色

**只做**：维护计划、调用 harness、管理 subagent 生命周期、校验 Gate、与用户交互。

**不做**：代写任何正式产物；手写 subagent prompt；内联执行任何内容生产；用口头判断替代 validator。

**内容生产全量外包红线**：P2A/P2B/P3/P3.5/P4 的所有正式产物（search.txt、source-brief.txt、outline.txt、style.json、planningN.json、slide-N.html 等）**必须且只能**由对应的 subagent 生成。主 agent 自己写出这些产物内容 = 合同违规。主 agent 唯一允许的"写"行为是通过 harness 生成 prompt 文件和通过 validator 校验产物。

## 2. 全局规则

### 2.1 步骤控制

- **CLI 固定步骤锁（强制）**：必须严格按 Canonical Plan 的主链 `P0 → P1 → (P2A|P2B) → P3 → P3.5 → P4 → P5` 执行；禁止增删改名。
- 分支二选一：进入 P2A 后绝对不可再跑 P2B，反之亦然。
- **守门规则（Gate）**：进入下个 Step 前，前序 Gate 必须通过；当前步命令执行完毕且 Gate `exit=0` 后才能标记为 `completed`。
- 失败时只允许两种动作：`RETRY_CURRENT_STEP` 或 回退 `ROLLBACK→StepID`。**严禁"跳到后续步骤试试看"**。
- `WAIT_USER` / `WAIT_AGENT` 是硬等待点；未收到输入/FINALIZE 前，**禁止执行后续步骤**。
- **人工审计断点**：是否开启、介入哪些节点、可看哪些材料，必须在 Step 0 采访时写入 `requirements-interview.txt`。断点只能挂在既有主链 Step 内，且只允许主 agent 控制；subagent 不得自行向用户发问。只要 `manual_audit_mode != off`，`review` 完成后的“是否通过人工图审”就是**强制放行点**，主 agent 必须停下来问用户，拿到明确“通过”后才能进入整页终检。

### 2.2 Subagent 强制调度（核心约束）

**通用生命周期**：`create(--model SUBAGENT_MODEL) → RUN(prompt路径) → STATUS… → FINALIZE → close`；完成即关，不复用。Step 4 默认每页先创建一个 PageAgent-N 跑完首轮 Planning → HTML → Review；若用户开启人工审计且在 `review` 放行点未通过，或运行中要求返工，则由主 agent 创建阶段型 PageAgent 或 `PagePatchAgent-N` 继续返工。创建时**必须**显式传 `--model SUBAGENT_MODEL`，禁止省略。`SUBAGENT_MODEL` 由用户在 Step 0 采访时指定（详见 3.1.0 及 6.2）。

**上下文隔离（强制）**：无论 CLI 环境默认是否让 subagent 继承主 agent 上下文，本 skill 要求所有 subagent 必须以**隔离模式**运行——subagent 唯一可见的上下文是主 agent 通过 prompt 文件显式传递的内容。如果 CLI 支持隔离参数（如 `--no-context`、沙箱模式等），必须在《Subagent 操作手册》中记录并在调用模板中包含。主 agent 的对话历史、SKILL.md 内容、环境变量等**不应该**泄露给 subagent。

**Subagent 强制调度表（每行 = 一个必须创建的 subagent）**：

| Step | Subagent 类型 | 职责 | 产物 | 主 agent 行为边界 |
|------|--------------|------|------|------------------|
| P2A | ResearchSynth | 联网检索 + 素材整理 | search.txt, search-brief.txt | 仅 harness 生成 prompt → 创建 subagent → 回收校验 |
| P2B | SourceSynth | 用户资料降维整合 | source-brief.txt | 同上 |
| P3 | Outline | 大纲构建（含内部自审闭环） | outline.txt | 同上，禁止介入 subagent 内部自审 |
| P3.5 | Style | 全局风格锁定 | style.json | 同上 |
| P4 | PageAgent-N（每页一个） | 页面规划 + HTML + 审查 | planningN.json, slide-N.html, slide-N.png | 同上，orchestrator 渐进式编排三阶段 |

**红线**：
- 上表中每个 Step 的产物**只允许对应 subagent 生成**，主 agent 内联生产任何产物 = 合同违规
- 即使 subagent 失败，主 agent 也只能重建 subagent 重跑，不能自己"补写"产物

Report a problem