Built-in skills and their reference docs now load from embedded files rather than inline strings.
What's wrong with this entry?
The built-in skills and their reference documents ship the same content, but it is now read from files embedded in the binary instead of sitting in the JavaScript as giant strings.
- Affected: the design-sync, design-canvas, doc, whiteboard, prototype, data-visualization, plan-artifact, run-skill-generator, artifact-components, artifact-design, artifact-dashboard, artifact-diagramming, artifact-data-table and artifact-pr-review skills, their page and template HTML, their helper scripts, and the bundled API reference set for C#, Go, Java, PHP, Python, Ruby and TypeScript.
- Paths of the form
/$bunfs/root/<name>.mdor.txtappear 164 times in this build and not at all in the previous one; references to that root total 1481. - Skill registrations, the lazy chunk loaders and the per-skill file listings are unchanged, so no skill was removed and nothing loads differently from the model's point of view.
- A small normalizer was added to accept either a raw string or a module namespace from an embedded asset.
- Practical effects: startup no longer parses tens of thousands of lines of prose as JavaScript source, and the prompt text is no longer greppable in the JavaScript bundle.
- Unconditional, with no flag or environment variable involved.
/$bunfs/root/sync-hashes.mjs-5rymp3zc.txt, /$bunfs/root/SKILL-8bdpvcbd.md, "templates/workshop-page.html": a,, references/plugin-eval-quickref.md, /$bunfs/root/color-formula-dc6qvg1m.md, /$bunfs/root/SKILL_COMPOSED-amaxbqpa.md, /$bunfs/root/cli-40vh1c3p.md, /$bunfs/root/SKILL-nrz66j7x.md, /$bunfs/root/platform-availability-nrkk9kb2.md, /$bunfs/root/token-counting-znjwtf00.md, /$bunfs/root/files-api-bmqt0y4f.md, /$bunfs/root/plugin-eval-quickref-2xg66zcm.md
Strings lifted out of the shipped bundle, so the claim above can be checked against them.