skwrit turns a taught process into verified paperwork: drafting with quality gates, an independent review step, a release step, and publishing straight into your fleet's own Nextcloud-based collaboration hub, with a re-download check that confirms the published copy actually matches what was sent.
skwrit is the worker behind the "skwrit matter engine," deployed as an optional role (optional/skwrit) inside SKStacks, the Ansible plus Docker Swarm framework that stands up the rest of the SKWorld fleet. It is new this release cycle, opt-in, and changes nothing for an instance that never deploys it.
A single worker service, one replica, port 8080, deployed with a stop-first update strategy so a bad update rolls back automatically rather than leaving a half-updated pipeline running.
The health endpoint reports back the status of the corpus, the quality gate, and the render assets, not just "process is up."
The engine's code and its "packs" (the taught-process definitions) live in a separate, private repository and ship as a container image built per site. This role deploys that image, it does not build the engine, and it refuses to run without a full @sha256: digest pin.
After a document is published into skhub, skwrit re-downloads its own publish to confirm the copy that landed actually matches what was sent, not just that the upload call returned success.
The architecture is not ornamental. Every piece exists to narrow a specific blast radius.
skwrit sits between three other SKWorld pieces: your fleet's workflow engine (skflow, built on Windmill, which calls the worker directly over the private overlay to drive a draft, review, publish pipeline as a taught process), your fleet's own Nextcloud-based collaboration hub (skhub, where the finished, reviewed document is actually published, readable afterward through the same web, desktop-sync, and mobile clients everyone already uses), and your fleet's own model-serving layer (the composer and reviewer endpoints point at your operator's own inference gateway, skgateway, rather than a hardcoded cloud API).
The name plays on the idea of a writ: a formal, binding, executed document. skwrit is a process that goes from draft to a reviewed, released, published piece of paperwork, not a chatbot that free-writes.
The deploy role, its render tests (a real Ansible test suite covering defaults, every configuration knob, naming, and every refusal path), the secrets-by-reference model, the independent reviewer configuration split, the /health endpoint, and an optional Nextcloud MCP service, also digest-pinned, letting AI tooling interact with the Nextcloud side of the pipeline through a standard tool-calling interface.
Unlike some of its sibling roles (skhub, for example), skwrit has not yet been folded into the framework's own pre-release lab-gate harness (skstack06) that exercises other roles end to end against a disposable live stack before each release. That is a natural next step for this role, but it is not yet true today.
If you see a claim elsewhere that skwrit is already lab-gated, that is wrong for this release. The lab-gate claim belongs to skhub, not skwrit, today.
skwrit's facts live in the actual role, not in a pitch. Start with the README that this page is grounded in, then the framework it deploys into.