Shrink the artifact. Keep the original in the repo or the object store.
CI logs and screenshot dumps bloat storage. Fileliq can crush a PDF or an image in the queue. It will not magically keep every pixel and hit a 10× ratio. Measure, then keep the source.
PDF artifacts
Compress PDF runs a Ghostscript profile. Photos shrink. Already-linearized or vector-only files may barely move. Always compare input vs output bytes in the queue. Keep the original if the delta is noise.
Text on born-digital PDFs usually stays selectable. Scans may be re-rasterized. Inspect. Password files must be unlocked first.
Do not compress as a substitute for Extract pages. If CI only needs pages 1–3 of a 200-page spec, extract the range, then compress that.
Image artifacts
Compress image keeps the container and exposes quality 40–95. Lower is smaller and uglier. PNG may not shrink like JPEG. If you still need a screenshot of UI, try PNG→WebP after you verified edges — or JPEG if alpha does not matter.
Resize by max width before compress when the screenshot is a 4k capture of a 700 px panel. Upscaling is allowed by the number field and will not invent detail. Do not do it.
CI-shaped constraints
Guest FREE: 50 MB, 5 ops/day, 1 concurrent, ~1 hour retention. A nightly matrix that uploads 40 PNGs will die on quota. Paid plans exist for volume. This UI is not your artifact registry — download and store the result in your own bucket.
HTML→PDF will not fetch the live web. Upload an .html file. External URLs are blocked. Inline assets. That slug may be stubbed on this box.
Do not pipe secrets into a guest convert “because it is easier than gs.” If the PDF is confidential, you are the controller of that upload. Read /privacy. Operator: Giannico Costantiello, VAT 08910530727, Turi (BA).
A pipeline that belongs in git
Keep originals in git-LFS, Release assets, or object storage. Commit the compressed copy only when the review actually needs the smaller file. Record the slug and the quality number in the PR so someone can reproduce.
Raster a PDF to JPEG only when the artifact consumers cannot open PDF. That is a preview path, not an archive path.
- Spec PDF → compress-pdf → attach if bytes actually dropped.
- UI screenshot → resize width → compress image → store.
- Photo dump → JPEG or WebP once → stop.
- Never loop compress. Second pass is generation loss or wasted CPU.
Steps
- Trim first. Extract the page range or resize the screenshot. Do not compress pages you will throw away.
- Run compress. Compress PDF or Compress image. Note quality if you used the slider.
- Compare bytes. If the output is not meaningfully smaller, keep the original. Do not ship a worse file.
- Exfiltrate. Download before guest expiry. Put the file in your artifact store. Fileliq is not your CI cache.
FAQ
Will compress always shrink the file?
No. Compare bytes. Already-optimized PDFs and some PNGs barely move.
Can I script Fileliq from GitHub Actions?
The public site is a UI on https://api.cifradocs.com. Guest TTL and quotas still apply. Do not treat the browser product as a stable CI runner.
Does compress strip EXIF?
Assume metadata may be stripped. This is not an EXIF editor.