The guide goes by what a chat has, not by its name
Merged
AL
3d ago+100+100+012345678901234567890123456789-64-64-01234567890123456789
Closes #80.
The how-to and the build page gave their steps by product: where the skill goes in Claude, ChatGPT and Grok, which of them builds the file, and what each lacks. Chats change within weeks, as Grok's place for skills did between 2026-11-13 and 2026-11-15, and each change left a step sending its reader the wrong way. The pages now go by what a chat has, which a reader can see in their own chat.
What changes
- The guide, step 1 (the skill): three cases in place of three products, with "take the first of these that fits". An agent with a terminal comes before them, with its one command, since it has no settings to look in.
- A place for skills takes
h5p-author.zip, once. A switch for running code, if there is one, goes on. - Plugins and no skills take
h5p-author-chatgpt.zip, the same skill laid out as a plugin. The case says that plugins come in more than one layout, and that a chat which refuses this one takes the zip attached. - Neither, or a refused upload: the zip goes with the message, in each new chat.
- A place for skills takes
- The guide, step 4 (the file): the chat hands over the
.h5pwhere its code runs Node.js and gets the build tool, and the source otherwise. The build archive moves here from step 1, with its own#archiveanchor. Nobody knows beforehand whether a chat's code can download the tool, and the skill's reply offers the archive where it would help. It goes into a project where the chat has projects, and with the message where it has none or the chat does not find it there. - The guide's other sections:
- The lead says the same in one sentence.
- With a local model names no app: where an app runs no code, or its code is not Node.js, the chat hands over the source.
- The help entries say "attached to your message" and "if your chat has projects".
- The chat's link builds nothing said a link is made by running code. The skill gives the build page's plain address and no link, so the entry now says that.
- The build page: "Works with your AI chat", the line under "How it works" and its first and third steps say what a chat can do. The third step links to
#archive. - The skill page and the README describe the two zips by what takes them, "installed once" against "attached to the message in each new chat". The README records the rule and keeps what was seen in each chat, with its date.
What stays
- The titles, descriptions and the guide's heading still say "ChatGPT, Claude or Grok". They are examples of an AI chat and what people search for, and say nothing of any one of them.
- Names that are not claims about a chat: the plugin zip's file name, "Claude Haiku made the sample", and "from Ollama or LM Studio".
- The comments in
vite.config.ts, which are dated records of what was tried.
Tested
- Build:
pnpm install --frozen-lockfileandpnpm build, which runstsc, pass. - Links: every link and anchor of the two built pages resolves, the new
#archiveamong them. - Layout: the changed sections were looked at in Chromium at 1280 px and 390 px. Neither page scrolls sideways.
- Text: in the two built pages no sentence is left that says what a named chat does or lacks.
- Not tested: a project of a chat other than ChatGPT. The step says what to do where the chat does not find the archive there.
Summary by cubic
The guide and build page now describe steps by what a chat has — a place for skills, plugins, projects, or code that runs — rather than by naming Claude, ChatGPT, or Grok.
- Chats change what they offer within weeks, so product-named steps can send readers the wrong way.
- Step 1 leads with the terminal agent's one command, described by what it does with no product named, since it has no settings to look in; the step says the command needs Node.js and falls back to a zip when it does not list the agent. Then goes by what a chat has: the plugin zip is one layout of several, and a chat that refuses it takes the zip attached.
- The build archive instructions move from step 1 to step 4, with its own
#archiveanchor; help entries, the README, and the skill page describe the two zips by what takes them. - The README records what was seen in each chat as a dated observation, not a standing condition.