#24 build(deps): bump @missing-elements/h5p-author from 0.6.0 to 0.9.3

Sourced from @​missing-elements/h5p-author's releases.

author 0.9.3

h5p-author 0.9.3 changes one paragraph of the skill's step 3: a link to the build page that carries the source is only ever one a command printed. The build does not change: every source builds as before, and no problem reads differently.

A link only from a command

Where the package cannot be installed and nothing can be fetched, the skill hands the user a link to https://missing-elements.github.io/h5p/build/ whose #source= carries the source, raw-deflated and base64url, as one of its two one-liners (Node or Python) writes it. A link an owner got from ChatGPT carried 3,167 characters that were no deflate stream, its table broken at byte 82 of 2,375: the chat had typed them itself, and the page built nothing.

The skill now says it outright. A link is what one of the two commands printed there, handed over whole, and where no command ran there is none: it never writes a #source= link itself, from memory or after an example, even when the user asks for one. Where nothing runs, it gives no link and puts the whole source in its reply, in one json block, for the user to paste into the page.

On Haiku with the skill's text and no tools, 12 of 12 replies gave the source as a json block and no link, with 0.9.2's text and with this one, half of them asked for a link to click: the rule costs that position nothing. What it changes in ChatGPT is not measured, and no rule reaches a link copied wrong between a command's output and the reply.

A copy installed with npx skills add takes the change with npx skills update -g; a copy uploaded to the Claude app needs a new upload. https://missing-elements.github.io/h5p/skill/ serves the skill as this release carries it once the site pins 0.9.3.

Full Changelog: https://github.com/missing-elements/h5p-offline-player/compare/author-v0.9.2...author-v0.9.3

author 0.9.2

h5p-author 0.9.2 changes the skill's text and the generated section of every content type's file: the skill now asks the user for the content type, then for the author and licence, before it writes the source (#182, #184, #183). The build does not change: every source builds as before, and no problem reads differently.

The skill asks, in order

The skill names the order before its first step: the content type, chosen with the user; then the author and licence, which they give or skip; then the source; then the build.

• The type. Unless the user named one from the table, the skill proposes one or two that fit, a sentence each, and waits. "A quiz" or "flash cards" says what the user wants, not which type.
• The author and licence. Once the type is chosen, the skill asks for them in a message of its own, only for what the user has not given, and says they may skip. A skipped name is the one the model knows the user by, or "Unknown"; a skipped licence is U (undisclosed). What the user gave stays, and the skill never writes a name made up for them.
• The build. Step 3 now lists the three ways to build and what each needs: here with npm, here from two downloaded files (as in ChatGPT), or in the user's browser through the build page, with no attempt here first, for a chat that can run nothing. The model takes the first that applies and tells the user which.

Every type file names the source's own fields

A source names its own title, language, license and authors beside type and params, and the build refuses a source that leaves out any of them. But every type file's example is type and params alone, and a chat that reads the file copies it: an owner's quiz built in ChatGPT on 0.9.1 had no author and was refused (#182). The Fields section that schema --markdown writes for each of the 28 content types now says so: the example is type and params alone, a source also names the four, all four are required, and the author and licence are the user's to give. The example is labelled rather than filled, because a name written in it would be copied as the author and build. A part other content holds (H5P.AdvancedText, H5P.Image, H5P.Text, H5P.Video) has no such line: it is sub-content.

On Haiku, in a stand-in for a chat with no skill installed (a fetch tool that opens only whole addresses, npm cut off, a scripted user who answers when asked), asked for a short quiz:

• With 0.9.1's text and a Question Set named, 2 of 3 runs asked for the author; the third wrote "Quiz author" and built.
• With the flow, before step 1 said that "a quiz" names no type, 4 of 6 runs chose the Question Set themselves. With the final text all 4 proposed two types and waited, then asked for the author and licence on their own.
• A name given and the licence skipped, a licence given and the name skipped, the author named in the request and the licence skipped, both skipped, both given: 5 of 5 wrote what was given and the default for what was skipped, and the run with the author named in the request asked for the licence alone.

The packages from the final text's runs all pass h5p-verify.

A copy installed with npx skills add takes the change with npx skills update -g; a copy uploaded to the Claude app needs a new upload. https://missing-elements.github.io/h5p/skill/ serves the skill as this release carries it once the site pins 0.9.2.

Full Changelog: https://github.com/missing-elements/h5p-offline-player/compare/author-v0.9.1...author-v0.9.2

author 0.9.1

h5p-author 0.9.1 changes only the skill's text: a chat that has the text and none of its files takes every address from the skill's page (#180, #181). Nothing else changes: every source builds as before, and no problem reads differently.

Every address from the skill's page

A chat given a link to the skill has its text alone, and the text gave it addresses to put together: the brief as a relative link, a type's file as …/h5p/build/types/<Type>.md, a content type's bundle as …/h5p/bundles/<type>.zip. A chat's fetch tool may open only an address it has seen whole, and refuse one put together from a name. The skill now sends such a chat to https://missing-elements.github.io/h5p/skill/missing-elements/missing-elements.github.io#18

On Haiku, in a stand-in for such a chat with npm cut off, asked for a three-question Question Set: with 0.9.0's text neither of two runs could open the brief, the type files or the bundle, and both ended with a build page link. With this text all six opened every one of them, and two of the three given the build tool built the package in the chat at once.

... (truncated)

• See full diff in compare view

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)