[user-github-me]

Website capture skips <img> elements whose images are declared only through srcset, leaving valid image candidates out of the asset catalog. Include images with either src or srcset in the existing discovery pass. Preserve DOM context, image descriptions, optimized-variant deduplication, and <picture> source handling.

The merge with current main retains its validated network transport for GIF metadata, public redirect handling, and refusal of private-address redirects.

Validation

• 102 capture, downloader, network, HTTP/TLS integration, and extraction tests pass. Coverage includes srcset-only images, largest optimized candidates, merged src/srcset references, picture sources, GIF metadata, and blocked redirects.
• Full workspace build, CLI typecheck, lint, formatting, test reachability, comment checks, and normal commit hooks pass.

The published tree matches the tested tree, and all newly published commits have verified GitHub signatures.

[jrusso1020]

Verified the bug is still live on main: the image pass in assetCataloger.ts still selects only img[src], so an <img> declared purely through srcset never reaches the existing srcset split. Widening the selector to img[src], img[srcset] is the minimal fix, and it is safe for srcset-only elements because img.src is the empty string there and add() already drops empty URLs, so no bogus page-URL asset appears. The new happy-dom tests drive the real catalogAssets script (made possible by narrowing the parameter to Pick<Page, "evaluate">, which every caller still satisfies) and cover srcset-only candidates, Next.js variant dedup keeping the largest width, src/srcset merging, and the no-source <picture> img case. Need: yes · Size: ok · Reuse: uses the existing srcset loop and dedup · Simpler: already minimal.

— Rames