War room architecture

Four small TOML files, a Rust generator, and Zola.

Human intent lives in a few small TOML files. Facts about StroggForge come from StroggForge. Rust checks the references and generates the pages. Zola renders them. Git remembers everything else.

How this site gets built

People edit the TOML. Everything after the generator is rebuilt from scratch on every run.

  1. source

    war-room/*.toml

    Projects, plans, platforms and workflow notes. The only part a person writes.

  2. tool

    tools/war-room

    Checks every reference, then writes Markdown and Mermaid into site/content and site/static/generated.

  3. tool

    npm run diagrams

    Renders the Mermaid sources to static SVG, once.

  4. component

    zola build

    The DreamWeave docs shell: routes, search, navigation.

  5. result

    GitHub Pages

    Deployed from main by .github/workflows/war-room.yml.

source

Workflow YAML

Inputs, secrets, jobs and conditions, read straight from .github/workflows and .github/actions.

How the pieces relate
  • war-room/*.toml tools/war-room
  • Workflow YAML read directly tools/war-room
  • tools/war-room npm run diagrams
  • npm run diagrams zola build
  • zola build GitHub Pages

Where Things Live#

InformationHome
Projects, domains, lifecycles, relationships, retired infrastructure, change logwar-room/ecosystem.toml
Releases and campaigns, their requirements and blockerswar-room/plans.toml
Target platforms and compiler policywar-room/toolchains.toml
Workflow operating notes and supply-chain stageswar-room/workflows.toml
Workflow inputs, secrets, jobs, runners, conditionsthe YAML itself, read at build time
Long-form writingsite/pages/
Presentationsite/templates, site/sass, site/static

Anything about another repository, like which foundation a tool depends on or which workflow a project calls, is a reviewed line in ecosystem.toml. StroggForge does not clone other repositories to find out. It would be wrong half the time and slow the other half.

site/content and site/static/generated are rebuilt from nothing on every run and ignored by Git. Editing them is a fine way to lose an afternoon.

What the Generator Refuses#

Duplicate IDs. References to projects, plans, platforms, workflows, jobs or composite steps that do not exist. A plan whose state disagrees with its requirements: blocked without a blocker, or ready with work outstanding. Cycles in prerequisites or successions. Unknown fields and enum values fail when the TOML parses.

That is the whole list, on purpose. Whether a project is really active, or whether a retirement is worth recording, is a judgment call for review, not for a compiler.

Mermaid#

Rust writes .mmd sources from the same records the tables use, so the diagrams cannot drift from the text. npm run diagrams renders them to SVG once, at build time. The browser receives an image and the source next to it; no Mermaid runs in the visitor's tab.

Domains#

Every project belongs to one domain: rust, openmw-lua or web. Maps group by domain before anything else, because the most common misreading of DreamWeave is assuming every repository runs through the Rust supply line. St4sh calling the createRelease helper is CI usage, not membership.

History#

A superseded project keeps its record with a superseded lifecycle and a successor relationship. Infrastructure with no project record, like a replaced builder image, gets a [[retired]] entry saying what replaced it and why. Archaeology is curated. Record a retirement when it explains the current architecture, not because something got old.

Maintenance loops