Features/MCP Workflows/Confluence/Confluence design description

Confluence design description

Most people who need your architecture will never open the model. This publishes it where they already read, updates the same page every time, and tells them at the top which revision they are looking at.

ConfluenceWrites · gated

The problem this solves

Published architecture documentation is stale from the day it is written, and nobody can tell. Someone exports a snapshot, it lands in a space beside three older snapshots, and readers pick whichever ranks highest in search. The page carries no provenance, so there is no way to know it describes a design from eighteen months ago.

How it works

The architecture, published where people who never open Dalus will read it. One page that updates in place rather than accumulating snapshots, stamped with the model revision it came from, and carrying a what-changed section on every republication.

01
Find the existing page first
The space is searched for a previous publication and that page is updated as a new version. Two pages describing the same system is the failure that makes published documentation worthless, because nobody knows which is current.
02
Stamp it
Model, branch, revision, date and who ran it, above any content. That stamp is what lets a reader decide whether to trust the page rather than emailing to ask.
03
Write for the non-modeller
No SysML syntax, no element type names, no tool vocabulary. Jargon expanded on first use, and any section only a modeller could follow gets rewritten.
04
Say what the model does not hold
A thin section states that the model records nothing rather than being padded with plausible engineering prose. A visible gap is a finding; invented content propagates into every document that cites the page.

The prompt

You are publishing a system design description from the Dalus model
[model name] into Confluence, for people who will never open Dalus.
The reader is an engineer, a manager or a reviewer in another
discipline. Everything on the page traces back to a model element,
and the page is republishable: running this again updates the same
page rather than leaving a graveyard of snapshots.

CONFIRM FIRST, in one message: which Dalus model and branch; which
Confluence space; whether this is a first publication or an update
to an existing page (and if so, which page); the scope (the whole
model or named subsystems); and whether to publish as one page or a
parent page with a child page per subsystem. Ask anything else you
need in the same message. Then begin.

CHECK FOR AN EXISTING PAGE before writing anything. Search the space
for a previous publication of this model. If one exists, read it,
and update that page as a new version rather than creating a second
one. Two pages describing the same system is the failure mode that
makes published documentation worthless, because nobody knows which
one is current.

READ THE MODEL: parts and their hierarchy, functions and
responsibilities, ports, connections and the flows they carry with
units, interfaces, requirements and their allocations, states and
modes, and any attributes the team uses for ownership or maturity.

STAMP AT THE TOP of the page, before any content: the model name,
the branch, the revision, the date published, and who ran it. A
published architecture page with no provenance is indistinguishable
from a stale one, and this stamp is what lets a reader decide
whether to trust it.

STRUCTURE:
- A short system overview: what the system is, its boundary, and
  its external interfaces. Written for someone who does not know
  the model.
- The architecture view at the top, rendered from the current model
  rather than reused from an older publication.
- One section per subsystem: what it does, its parts, its
  interfaces with what flows across them and in which direction,
  the requirements allocated to it, and its states or modes where
  the model records them.
- An interface summary table across the whole system, since the
  interfaces are what readers from other disciplines come for.
- A requirements index with allocation.

LINK EVERYTHING BACK. Every requirement carries its customer ID as
a link into Dalus. Every part and interface links to its model
element. The purpose of publishing to Confluence is that a reader
who wants more can follow the trail instead of emailing the systems
engineer, so a page with no links back has missed the point.

WRITE FOR THE NON-MODELLER. Expand model jargon on first use. Do
not reproduce SysML syntax, element type names or tool vocabulary
in the body text. If a section can only be understood by someone
who already knows the model, rewrite it.

NEVER INVENT. If the model does not record a subsystem's function,
its interfaces or its owner, the page says the model does not
record it. Do not fill gaps with plausible engineering prose, and
do not soften a thin section by padding it. A visible gap is a
finding the team can act on; invented content is a defect that
propagates outward into every document that cites this page.

ON REPUBLICATION, add a "what changed" section directly under the
stamp: elements added, removed or modified since the previous
publication, requirements whose statements changed, and interfaces
that changed. Compare against the previous page version, and if the
previous version cannot be read, say so rather than presenting the
page as unchanged. This section is the reason people come back to
the page, and without it a republished document is just a fresh
snapshot nobody notices.

BEFORE PUBLISHING, show the user the page structure, the section
list, the stamp, the change summary if this is an update, and any
place where the model was too thin to write a section. Wait for
approval. Publishing to a shared space is visible to everyone in
it, so it is always gated.

AFTER PUBLISHING: report the page URL, what was published, what was
left out and why, and the gaps found in the model along the way as
findings, never as silent edits to the model. Offer, do not
execute: republishing on a schedule before each review, publishing a
stakeholder review pack scoped to one decision if what the team
actually needs is a review rather than a reference, and opening
model tasks for the gaps.

THIS WORKFLOW IS READ-ONLY on the model. It writes only the
Confluence page, and only after approval.

Replace the [bracketed] placeholders with your model and project names.

What you get

One page, updated in place, with a provenance stamp at the top
A what-changed section on every republication
Interface summary table and requirements index with allocation
Every requirement, part and interface linked back into Dalus
Dalus + Confluence

Publishes as one page or a parent with a child per subsystem, into the space you name, and updates the existing version rather than creating a second. Publishing to a shared space is visible to everyone in it, so the structure, the stamp and the change summary are shown for approval first.

Nothing is written until you approve it. The prompt carries the gate: the agent shows the full change set and waits, whether the target is the model or a system it reaches through a connector.

Common questions

Will it create a new page every run?
No. It searches the space for the previous publication and updates that page as a new version. The whole point is one page people can bookmark, not a graveyard of snapshots nobody can rank.
What stops it reading like model output?
It is written for someone who has never opened Dalus: no SysML vocabulary, no element type names, jargon expanded on first use. Elements still link back for anyone who wants the detail, which is the reason to publish into Confluence at all.
What if our model is thin in places?
The page says so. It will not fill a gap with plausible prose, because invented content propagates into everything that cites the page. Those gaps come back to you as findings, not as silent edits to the model.