Every file is an attachment, and nothing is loose
On most sites, media accumulates in two places at once. Some of it lives properly in the library, cataloged and reusable. The rest arrives as loose files uploaded straight to a server by whoever needed them that afternoon, referenced by path, invisible to the system that is supposed to be managing them.
Loose files cause specific problems. Nobody can find them, and nobody knows whether they are still in use. Replacing one means hunting through pages for every reference, because alternative text does not exist and responsive versions were never generated. When a site moves or a path changes, they break silently. The protocol treats that arrangement as a defect rather than a habit. Every asset lands as a first-class attachment in the library, with nothing loose anywhere.
The ingestion gate
Assets pass through a gate before anything is packaged, and the gate refuses rather than warns. Alternative text has to be present, while responsive sources are generated rather than hoped for. Fonts are self-hosted, so a page never waits on a third-party request before text can render. Media budgets are enforced, which means a hero video arriving at twenty megabytes is stopped at the door rather than discovered by a visitor on a phone.
"A report tells somebody a problem exists and relies on them acting. A gate does not let the problem through, so correctness stops depending on anybody’s attention on a busy afternoon."
Referenced by identity rather than by path
An asset is referenced by what it is, not by where it currently sits. The practical consequence is that replacing an image updates every page carrying it, in one action. Swapping a product photograph across forty pages stops being an afternoon of searching. Nothing breaks when a file moves, because no page was ever pointing at a location. That single decision removes most of the maintenance burden media normally creates on a mature site.
Provenance and license travel with the file
Every asset carries where it came from and what may be done with it. Whether a visual was produced in-house, commissioned, generated, or licensed from a library. What the license permits, including whether it can be reproduced or resold. Who supplied it, and when.
Reconstructing that months later is close to impossible, and the moment it matters is usually the moment somebody wants to use an image somewhere new, or a question arrives about rights. Recording it at ingestion costs nothing and answers the question permanently.
Owned media as the preferred source
Ordering matters here, because the easiest file to reach for is nearly always the one everybody else has already used. In-house and commissioned visuals sit at the top of the order, while generated imagery produced for a page’s own subject sits alongside them. Stock libraries are a fallback tier rather than a default, since the same photographs circulate through an entire industry and a page illustrated from a shelf looks indistinguishable from its competitors.
The reasoning behind original imagery as an authority signal is covered in page authority, and generation per page is covered in AI-assisted design.
What sits outside asset work
Media budgets govern images, video and fonts. Code payload is a separate discipline with a separate cause, covered in zero CSS bloat, and the measured result of both is covered in faster page speed.
Where asset work connects
The blueprint that declares where media may appear is covered in composer blueprints, and the document that references it is covered in clean semantic markup.
How the engagement runs
Asset work usually starts with an audit of what is currently loose, which is frequently more than a client expects. Sessions are screen shared and start within days. If replacing one image on your site means finding every page that uses it, the referencing is the problem rather than the workflow.