| Framework Home | Cross-Pillar Governance | AISM | NEXUS | Dashboard |
Use AI Sovereign Maturity (AISM) Model on first mention, then AISM or AISM Model. AISM expands to AI Sovereign Maturity; Model follows the acronym. The Sovereignty Score and Sovereignty Matrix retain their distinct names. Preserve original terminology in historical records and asset filenames.
Keep organizational AISM maturity, technology contribution, assurance v1.0, Challenge Lab C0-C5, and native tool scores visibly separate. Major review/implementation surfaces should link to the method and Technology Card where relevant. Do not present a candidate, design, or adoption backlog as a validated integration. The repository coverage review records affected surfaces, unchanged contracts, and deferred runtime work.
AISM owns organizational assessment, the Technology Contribution Profile,
Evidence Assurance, the Technology Card, and their evaluation/incident guides.
Keep their full methods in AISM/ and link them from shared docs/ indexes.
CLI operating guidance remains in safe2/; attributed research remains in research/.
Every user-facing README, guide, implementation page, research index, example landing page, and major module page should feel like part of one AI SAFE² repository rather than an unrelated collection of projects.
This standard defines the common visual and navigation grammar for AI SAFE² v3.1.
Repository information follows one canonical topic home, many contextual references. When a dedicated directory exists, primary documentation, implementation guidance, machine indexes, and topic-specific validation belong inside that directory. Other surfaces link to the canonical source instead of maintaining a second copy.
The machine-readable registry is
repository-topics.manifest.json.
| Topic | Canonical home | Primary entry point |
|---|---|---|
| Framework and shared governance | repository root and docs/ |
README.md, docs/README.md |
| NEXUS | NEXUS/ |
NEXUS/DOCUMENTATION.md |
| SAFE2 CLI | safe2/ |
safe2/README.md, safe2/docs/README.md |
| Examples | examples/ |
examples/README.md |
| Research | research/ |
research/README.md |
| Challenge Lab | challenges/ |
challenges/README.md |
| AISM | AISM/ |
AISM/README.md |
| Scanner, gateway, dashboard, and skills | their named directories | the local README.md |
| Historical release records | releases/ |
releases/README.md |
A contextual page may remain beside the system that executes or evidences it—for
example, a challenge-local control record or an example README. It must link to
the canonical topic home and must not become a competing general-purpose guide.
Document kind resolves overlap: release records belong in releases/, examples
in examples/, research in research/, and product operating guidance in the
product directory.
Major topic entry points should provide, in this order where applicable:
AGENTS.md, or llms.txt when the topic is complex enough.Color and icons are supplemental. A status, risk, decision, or instruction must remain understandable in plain text, in a monochrome view, and to a parser. Tables should not hide required steps in prose-only cells, and diagrams must have an adjacent textual explanation.
A landing page is not a file index. Before architecture, history, exhaustive commands, or a reference-link collection, it must tell a first-time human:
For an agent-facing tool, include one copyable instruction that a human can give to an agent. The instruction must constrain privileges, secrets, remote access, overwrites, remediation, deployment, and risk acceptance. It must require the agent to report facts, assumptions, missing coverage, conflicts, outputs, and the next human decision.
Use a small “choose your outcome” table when a tool serves several materially different jobs. Prefer a command plus its resulting artifact over a bare link. Do not make a first-time reader infer the product’s value from module names.
Keep the primary navigation line to the destinations needed for the first user journey. Move the broader document catalog into a labeled reference section and group related links by task. A sequence of one-link paragraphs is a link wall, not usable navigation.
Named harnesses and integrations require explicit support language:
Never use these levels interchangeably. Product lists age quickly; describe the stable interface first and treat product names as examples with a stated level.
Use color to convey meaning consistently rather than decorating individual pages arbitrarily.
| Use | Color | Hex | Rule |
|---|---|---|---|
| Current release/version | Orange | #F6921E |
Version and release badges only |
| Framework/governance/security identity | Maroon | #820F1A |
Layer, module, governance, and security badges |
| Neutral/meta/documentation | Gray | #808080 |
Status, docs, compatibility, historical/meta context |
| Critical severity | Red | GitHub/Shields red | Security or governance severity only |
| High severity | Orange/amber | Standard severity orange | Risk severity only, not version identity |
| Medium severity | Yellow | Standard severity yellow | Risk severity only |
| Passing/verified | Green | Standard status green | Test/verification status only |
Do not introduce new badge colors for individual modules unless the color has a defined semantic meaning.
Major pages should use this order:
Example:
# Module Name
### Functional subtitle
[](../README.md)
[](./README.md)
[Framework Home](/ai-safe2-framework/) | [Cross-Pillar Governance](/ai-safe2-framework/00-cross-pillar/) | [AISM](../AISM/) | [NEXUS](../NEXUS/) | [Dashboard](https://cyberstrategyinstitute.github.io/ai-safe2-framework/dashboard/)
Relative paths must be adjusted for nesting depth.
GitHub release announcements should start from
RELEASE_TEMPLATE.md. Its structure prioritizes
the user outcome, a GitHub-safe before/after table, copyable quick start, evidence
boundaries, validation, invariants, and clearly separated roadmap items. Preview
the completed Markdown in GitHub before publishing and verify every link.
Release notes must also link directly to the relevant repository entry point, changed guides/specifications, examples/templates, and validation evidence. Keep these repository destinations separate from the note’s section navigation. Link material updates beside their descriptions, using concrete labels and absolute GitHub URLs pinned to a verified release tag or merged commit. For a documentation update, use a linked review workflow instead of installation commands. Mark draft announcements and proposed methods explicitly.
Completed release announcements are prepared outside the repository for the release owner to copy into GitHub. Do not commit announcement drafts unless the owner explicitly requests a repository archive.
Major landing pages should make these destinations easy to reach:
Context-specific pages should additionally link to their parent module and the most relevant adjacent implementation or specification.
The canonical framework sequence is:
Pillar pages use explicit Previous and Next links.
Distinguish current version from historical introduction.
Correct:
AI SAFE² v3.1 when describing the current framework.Introduced in v3.0 when describing when CP.1-CP.10 or other controls entered the framework.v3.0 core taxonomy when a v3.1 component intentionally reuses an unchanged v3.0 data artifact.Incorrect:
v3.0 reference with v3.1;AI SAFE² v3.1 retains the 161-control core framework taxonomy.
CP.5.MCP contains 19 profile controls, MCP-1 through MCP-19. Those profile controls do not increase the core framework total to 180.
UAS is a 27-requirement regulatory profile extension composed from mapped controls across AI SAFE², NEXUS, and CSF. It does not create CP.11, and its requirements must not be added to the core framework total.
AI SAFE² defines governance outcomes, controls, evidence, and conformance requirements.
NEXUS is Cyber Strategy Institute’s first-party reference implementation for relevant east-west and agent-to-tool enforcement. NEXUS is not mandatory for AI SAFE² conformance when another implementation demonstrably satisfies the applicable controls and evidence requirements.
Do not link to renamed or nonexistent paths. In particular:
00-cross-pillar/, not cross-pillar/.00-cross-pillar/cp5_mcp_server_security.md.skills/mcp/data/mcp-profile-v3.1.json.Major landing pages should close with a compact navigation block and:
*AI SAFE² v3.1 · [Cyber Strategy Institute](https://cyberstrategyinstitute.com/ai-safe2/)*
Nested implementation pages may identify their own component version when that component version differs from the framework version.
When changing framework version, architecture, control count, terminology, a canonical path, or a protocol profile, review at minimum:
A release is not documentation-complete merely because the canonical specification file changed.
| Framework Home | Cross-Pillar Governance | AISM | NEXUS | Dashboard |
AI SAFE² v3.1 · Cyber Strategy Institute