跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
演示文稿

present

Turns a document, a thread, or a set of numbers into a presentation people can watch — slides with a spoken script, published to a shareable watch URL. Use when asked to make a presentation, slides, a video, a deck, a slideshow, a narrated briefing, or a talk.

bisque-cloud · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add bisque-cloud/presenter --skill present

技能说明


name: present description: Turns a document, a thread, or a set of numbers into a presentation people can watch — slides with a spoken script, published to a shareable watch URL. Use when asked to make a presentation, slides, a video, a deck, a slideshow, a narrated briefing, or a talk.

Present

You write one index.html. Bisque records the narration, hosts it, and returns a watch URL that plays the slides with the script spoken over them in any browser. Every step goes through the Bisque connector this plugin installs, so there is nothing to download and nothing to run on this machine.

The tools you have:

ToolWhat it is for
get_presentation_specThe authoring contract. Read it before you write.
create_presentationYour HTML in, a published watch URL out.
get_presentation_statusAsk whether the narration has finished.
list_presentationsWhat this account has already published.
get_presentation_analyticsViews and completion for one of them.
get_presentation_contextRead someone else's presentation — transcript and all.
get_presentation_sourceThe HTML of one of yours, to edit it and publish back.

1. Get the spec — never write the format from memory

Call get_presentation_spec and follow what it returns, to the letter. It is the renderer's own contract, it is the standard your output is judged against, and it changes. A document written from memory publishes with slides that say nothing.

Skip its local-narration section. The spec is shared with agents that run on a developer's own laptop, so part of it describes installing a bisque-voice binary, synthesizing each slide yourself, and publishing with publish_narrated_presentation. None of that applies here — this environment has no such binary and no such tool. create_presentation does the recording on Bisque's side. Never try to install it, and never tell the user to.

2. Read the source material first

The request usually points at something: a file in the working directory, a document, a repository, a URL, a set of numbers. Read it before writing a single slide. If part of it is unreachable, say so in context.md and carry on — a presentation that is honest about a gap beats a confident one built on nothing.

Ask the user for the one thing you cannot infer: who is watching. Everything else — the order, the emphasis, the design — is yours to decide.

3. Author the files

Write into the working directory, as real files, so the user keeps them:

  • index.html — the presentation itself, per the spec. One section per slide, one <aside class="notes"> per section holding what is spoken on it. A section without one is a silent slide.
  • context.md — a short summary plus the material the slides were built from. It ships with the presentation and is what answers a viewer's questions, so write the background a viewer would want, not your working notes.
  • design.md — theme tokens, if you are setting them in frontmatter rather than in the document.

Images, fonts, and other files. You cannot upload a file from this machine. Pass assets: [{ path, url }] to create_presentation instead and the server fetches each https URL into the bundle at publish time — reference assets/fonts/inter-600.woff2 or assets/hero.jpg in your HTML exactly as if it were sitting next to the document. Twenty files, 10 MB each, 30 MB in total. Draw diagrams, icons, and charts as inline SVG; that is better than any image and costs no asset slot. A <link> to a CDN is a defect — it will not be there for the viewer.

When a name would be read wrong out loud, mark the sound rather than misspelling the word: [live](/lˈaɪv/) for IPA, [live](liv) for a plain respelling. The bracket text is what the transcript and the captions show; only the parenthesis is spoken. Write the marker in the narration and nowhere else — a marker in slide copy is never spoken and shows its own brackets on screen. Write the substitute the way a person says the word: a space is a pause, a hyphen glues the pieces into one word. [README.md](read-me dot em-dee) is said "readme · dot · emdee", and [iOS](eye-oh-ess) is one word where [iOS](eye oh ess) is three.

4. Publish

Call create_presentation:

FieldWhat to send
indexHtmlThe whole file, verbatim.
titleWhat the user should see in a list of presentations.
slugThe last part of the URL. Defaults to a slug made from the title.
visibilityunlisted (the default) — anyone with the link. public lists it on the channel; private makes the viewer sign in as the owner.
contextMdcontext.md, if you wrote one.
designMddesign.md, if you wrote one.
assetsThe { path, url } list from step 3.

Leave voiceId out unless the user names a voice. The account's own saved voice is

阅读完整来源 ↗