The network
This site is one of several that carry one argument between them. Each is a separate repository, a separate domain and a separate release history — and each is built from the same three pieces: a commissioning brief published in full, a static site written from it, and the validate → tag → deploy pipeline that gates every release.
What is shared is machinery, not content. One stylesheet (assets/site.css), one nav component, one pipeline shape, one release discipline — copied deliberately rather than packaged, so any site can diverge without asking permission. What is not shared is a single word of argument: each site is written from its own documents.
The sibling sites
| Site | Owns |
|---|---|
| sgit.ai | The parent project: the zero-knowledge vault layer, the shipped CLI, the demos and the catalogue — and the network index that lists every site here |
| open-source.sgit.ai | The open-source position: the licences, the packages, and the argument for publishing them |
| llms.sgit.ai | Working with language models — what they are used for here and what they are not trusted with |
| standards.sgit.ai | Standards and instruments: the formats, schemas and evidence artefacts |
| graphs.sgit.ai | Graphs as the modelling substrate — the grammar, the depth model and the book |
| wardley-maps.sgit.ai | Mapping: evolution axes, situational awareness, and the maps drawn for this project |
| risks.sgit.ai | Risk: accepted risks with named acceptors and named mitigations |
| nhi.sgit.ai · sentinel.sgit.ai | Non-human identity, and the sentinel that watches it |
| pki.sgit.ai | Key registries for agents |
| sg-compute.sgit.ai | The ephemeral compute platform, measured from its own tree |
The boundary with this site
Now that this site has content, the seams can be named rather than guessed at. Each row below says what the other site owns and what this one does not repeat, because the failure mode for a network of sites is not a dead link, it is two sites making the same argument slightly differently.
| Site | Where the line falls |
|---|---|
| coding.sgit.ai | How code is written here — the five languages, the 31 rules, the measured compliance and the enforcement gap. Nothing about what the code does. |
| sg-compute.sgit.ai | The richest source of the examples quoted here. That site owns what the code does; this one owns how it is written. The broken CI guard belongs to both — that site tells it as a rename story, this one tells it as an enforcement story. Cross-linked, not duplicated. |
| open-source.sgit.ai | /for-agents/ is the shared boundary. That site has the argument — that someone still needs to understand what is underneath, and that code-reading is the scarce asset. This one has the artefact: the conventions that position produces at 217,266 lines. Both directions worth reading. |
| issues-fs.sgit.ai | Shares the Schema__Node and Safe_Str__Graph_Types primitive pattern — the same convention in a different codebase, which is the best evidence available that it generalises rather than being one repository's habit. |
| standards.sgit.ai | SPDX identifiers and licence compliance. A light link: that site owns the instrument, this one owns the CC BY 4.0 / Apache-2.0 distinction as it applies to quoting code. |
| llms.sgit.ai | What models are trusted with. This site does not cover that; it covers what the code looks like because a model writes most of it. |
| graphs.sgit.ai | Little overlap. |
| sgit.ai | The vault product. Its HTML authoring contract is a coding standard in its own right, and is the obvious next language section here if it is ever wanted. |
Two of these need work on the other side and are tracked as T15: this site links out to open-source.sgit.ai and sg-compute.sgit.ai, and neither links back yet.
What every site in the network runs
The same three jobs, in the same order, gated the same way — described in full on this site's engineering page:
validate ──▶ tag-release ──▶ deploy # a failure in any job stops the ones after it; the live site is never # replaced by a release that did not pass its own checks
The differences between sites are the checks each one adds. sg-compute.sgit.ai bans live hostnames and instance ids because it documents running infrastructure. This site bans credential shapes, because a site about how code gets written will quote real sessions and real configuration — and in v0.2.0 it added an AWS-account-id shape, because publishing a redaction list means publishing the values it names unless something stops you. The tripwire is here.
This site also adds one gate the siblings do not have, because its subject demands it: every number and every code example on every page is written in from a published source by a generator, and CI fails the release if one has drifted. A style guide whose examples have drifted from the code is worse than no style guide. How it works →
Licence
Every page of this site, its source documents, llms.txt and llms-full.txt are released under CC BY 4.0. Attribution: Dinis Cruz, with AI co-authorship (Claude, Anthropic). Where this site quotes source code, it quotes it under that code's own licence; those notices are retained and the snippets do not carry this site's licence.