Get cloudflare skills from middlecache

[mvvmm]

Summary

Removes the committed copy of Cloudflare Skills from the repo and replaces it with a fetch-at-build-time approach sourcing from middlecache.

• bin/fetch-skills.ts — fetches skills.tar.gz from middlecache and extracts into skills/. Skips if skills/ already exists; --force overrides. --soft mode (used by predev) warns on failure instead of blocking.
• package.json — adds prebuild and predev hooks to run the script automatically
• skills/ — removed from git, added to .gitignore

In dev, fetch runs once and is skipped on subsequent restarts; network failures warn but don't block. In CI, skills/ is always fetched fresh since it isn't committed — fetch failure fails the build.

The well known skills can be verified here: https://11705275.preview.developers.cloudflare.com/.well-known/skills/index.json

[github-actions]

This pull request requires reviews from CODEOWNERS as it changes files that match the following patterns:

Pattern	Owners
*	@cloudflare/pcx-technical-writing
*.ts	@cloudflare/content-engineering, @kodster28
package.json	@cloudflare/content-engineering

[mvvmm]

/bonk please review this PR for any issues

[ask-bonk]

One issue found — filename typo in AGENTS.md.

[ask-bonk]

Posted review on PR #28658. Here's the summary:

One issue found — filename typo in AGENTS.md.

• MEDIUM — AGENTS.md:41 references bin/fetch-skills.js but the actual file is bin/fetch-skills.ts. Posted an inline suggestion to fix.

The rest of the PR looks solid:

• bin/fetch-skills.ts has good error handling with --soft/--force flags, proper streaming to tar, and clean skip-if-exists logic.
• prebuild and predev hooks in package.json are correctly wired — --soft on predev ensures local dev is not blocked by network failures.
• .gitignore correctly excludes skills/ with a descriptive comment.
• AGENTS.md formatting improvements (table alignment, code block fencing) are all correct.
• The ~54k lines of deletions are all the previously-committed skills content being removed as expected.

github run

[github-actions]

Preview URL: https://5f7795df.preview.developers.cloudflare.com
Preview Branch URL: https://get-skills-from-middlecache.preview.developers.cloudflare.com

[elithrar]

this is great @mvvmm - thank you

[cdrubin]

Could you make use of this function perhaps: https://github.com/cloudflare/cloudflare-docs/blob/production/src/util/custom-loaders.ts#L20 for grabbing the file from middlecache?

Also, is there a way to re-use some of the content collection pattern for doing any of this? Or not really useful because all the changes are done before the tarball is generated?

[mvvmm]

Could you make use of this function perhaps: https://github.com/cloudflare/cloudflare-docs/blob/production/src/util/custom-loaders.ts#L20 for grabbing the file from middlecache?

No for a couple of reasons I think. This function saves to .tmp but we need these in /skills/, and the need for the tarball streaming/extraction. This is also "pre-astro" in the build/dev lifecycle.

Also, is there a way to re-use some of the content collection pattern for doing any of this? Or not really useful because all the changes are done before the tarball is generated?

I think the astro-skills is already using the content-collection pattern? See https://github.com/cloudflare/cloudflare-docs/blob/production/src/content.config.ts#L128, but let me know if I misunderstand something here.

[mvvmm]

Used downloadToDotTempIfNotPresent.

[mvvmm]

/bonk mind reviewing this one one more time after my new changes?

[ask-bonk]

UnknownError: "Overloaded"

github run

[ask-bonk]

@mvvmm Bonk workflow failed. Check the logs for details.

View workflow run · To retry, trigger Bonk again.