课件帮 AIPPT — 一句话生成专业 PPT

接入「课件帮」AI 平台，全流程自动完成 PPT 生成：大纲确认 → 模板匹配 → 智能生成 → 交付链接，无需手动操作任何 API。

适用场景

前提条件

需要环境变量：

KEJIAN365_AUTH_TOKEN=<Your_Token_Here>

密钥获取：https://kejian365.com/oapi-portal

使用方式

直接说：「帮我做一个关于 XX 的 PPT」即可启动。

Execution Checklist

Run through these steps in order. Track completion in context — never repeat a completed step.

[ ] Step 1  Gather requirements from user
[ ] Step 2  Generate outline (local LLM) → show to user → confirm
[ ] Step 3  run list_themes.py → pick theme → tell user
[ ] Step 4  Write params.json + run script (creates task ONCE on success)
[ ] Step 5  Re-run script every 30s to poll until terminal state
[ ] Step 6  Report final result

Hard rules:

• NEVER call POST /skill/task/create directly — always use the script
• NEVER change --work-dir for the same task (causes duplicates)
• NEVER show ppt_id, work_dir, file paths, or raw URLs to the user

Script idempotency:

Step 1 — Gather Requirements

Ask in one message:

好的！请告诉我：

1. PPT 的主题是什么？
2. 有什么特别要求吗？（受众、风格、页数、语言等，没有就跳过）

Extract:

Step 2 — Generate Outline

Generate locally (no API). Then show the outline to the user and wait for confirmation:

已为您生成以下大纲，共 {N} 页：

1. 封面 — {title}
2. 目录 — {title} …（逐页列出）

满意的话我来继续选模板；如需调整请告诉我。

Outline format (OutlineModel):

• pages[].pageNumber sequential from 1, no gaps
• Page 1 = 封面, last = 结束, at least 1 内容
• Add 目录 when ≥ 6 pages; 章节 + its 内容 pages share chapterNumber
• main_title: PPT 主标题（可与 topic 相同或更精炼）
• narrative: 全局叙事策略，指导每页内容基调（可留空 {}）

{
  "main_title": "主标题",
  "narrative": {
    "objective": "受众沟通目的",
    "core_message": "核心信息",
    "mode": "教学式",
    "mode_behavior": "内容推进逻辑",
    "reading_mode": "均衡阅读"
  },
  "pages": [
    { "pageNumber": 1, "pageType": "封面", "title": "主标题", "content": "" },
    { "pageNumber": 2, "pageType": "目录", "title": "目录", "content": "1. 概述\n2. 分析" },
    { "pageNumber": 3, "pageType": "章节", "title": "第一章", "content": "", "chapterNumber": "01" },
    { "pageNumber": 4, "pageType": "内容", "title": "内容页", "content": "...", "chapterNumber": "01" },
    { "pageNumber": 5, "pageType": "结束", "title": "谢谢", "content": "" }
  ]
}

Material (optional): also generate a short research summary to improve content quality and pass it as the material field.

Step 3 — Select Theme

Use the script — never call the API directly (raw JSON response is too large for context):

python kejian365-aippt/scripts/list_themes.py

Output per theme: THEME: theme_id|theme_name|style|scene

Reason over the full list, pick the theme_id that best fits the PPT topic, audience, and tone. Save it for Step 4 themeId.

Tell the user:

已为您选择「{theme_name}」模板（{style} 风格 · {scene} 场景），即将开始生成…

Step 4 — Create Task

Work directory

Use a stable slug per topic — same task = same directory every time:

/mnt/user-data/workspace/{topic-slug}-ppt-tmp

params.json (UTF-8, no BOM, no \uXXXX escaping)

{
  "topic": "主题名称",
  "themeId": "theme_xxx",
  "outline": {
    "main_title": "主标题",
    "narrative": { "objective": "...", "core_message": "...", "mode": "教学式", "mode_behavior": "...", "reading_mode": "均衡阅读" },
    "pages": [ ... ]
  },
  "authToken": "<credential>",
  "requirements": "用户的特别要求",
  "material": "# 研究摘要\n...",
  "themeConfig": {
    "settingPages": "标准",
    "settingLanguage": "中文",
    "settingAudience": "智能决策",
    "contentDepth": "智能生成",
    "illustration": "智能配图",
    "sourceMode": "智能参考"
  },
  "illustrationMode": "standard"
}

Run script

python kejian365-aippt/scripts/create_ppt_task.py \
  --params-file {work-dir}/params.json \
  --work-dir    {work-dir}

On SUBMITTED:, parse 查看链接: from stdout and say:

🎉 PPT 任务已创建成功！ 主题：「{topic}」，模板：「{theme_name}」 生成大约需要 5–15 分钟，可以先点这里预览（生成完成前内容可能为空）。 我会每隔 30 秒自动查询进度，请稍候。

Step 5 — Poll Progress

Re-run every 30s (omit --params-file after first run):

python kejian365-aippt/scripts/create_ppt_task.py \
  --work-dir {work-dir}

When pending, parse 进度: and 查看链接: and say:

正在生成中，已完成 {N}/{total} 页，完成后可在这里查看。

Step 6 — Report Result

Success (DONE:)

Parse 主题: / 页数: / 查看链接: and say:

🎉 「{topic}」的 PPT 已生成完成，共 {N} 页！ 点击这里查看您的 PPT

Provide raw URL only if the user explicitly asks.

Failure (FAILED:)

Lines starting with [INTERNAL] are for the agent only — never show to user. Delete task_state.json silently to retry.

Appendix — API Reference

api-reference.md