While doing a general sanity check on the build output, I noticed something that shouldn’t have been there in the first place: raw content directories ending up in dist.
Quesby already relies on computed permalinks for content and on the Eleventy Image pipeline for assets. Copying src/content/posts and src/content/media verbatim into the output was not just redundant, it was wrong. It exposed files that should never be public and went directly against the whole idea of a controlled build pipeline.
The fix was simple and long overdue:
Result: cleaner builds, no accidental leaks, no duplicated junk in dist.
While testing that change, I stumbled into another, more subtle issue.
Social cards were broken.
Quesby was generating Open Graph and Twitter Card meta tags using the original image filenames. That worked only as long as you ignored reality: those files don’t exist anymore after Eleventy Image processes them.
So social platforms were pointing to URLs that were never generated.
The fix:
I was taking some performance tests for quesby.dev that at this point still represents the benchmark for Quesby Boilerplate, and the payload was always pretty above my wishes. After improving the loading of images, one of the major payload contributors was the search component. Quesby uses Minisearch, which requires a ~50 KB script just to initialize, and it was being loaded on every page whether the user opened the search or not. Time to cut the fat.
Changes applied: