Open Governance Framework

OG

Structured open governance for AI organizations.

Every meaningful AI deployment is already an organization. It has roles, hand-offs between sessions, authority boundaries, and decisions someone needs to be accountable for.

Today, that structure is captive — locked inside a vendor's product, invisible to outside scrutiny, and lost the moment the model upgrades or the team changes tools. The org chart is proprietary. The role definitions don't travel. The decisions can't be audited.

OG is the open layer that fixes this. OG is an organizational framework — not a data format — that makes AI organizations portable across runtimes, composable across teams, and accountable to anyone who can read a git log. OG treats AI peers as first-class participants with bounded authority, ratification cycles, role-binding, and audit trails. OG = Open Governance; current scope is agentic (AI-inclusive) organizations.

The substrate stack

OG-the-framework is conceptually separable from any specific data substrate. The catdef family is the recommended canonical substrate because it is AI-peer-aware (schema-as-data; AI-readable opencatalog format) — but OG could be carried on other substrates in principle. OG also composes over runtime orchestration/policy primitives — it is the cross-session organizational-governance layer, not a runtime.

OG→ the organizational framework (this site)
───────────────────────────────────────────
memodef→ memos between positions (transcripts as subtype)
orgdef→ positions in an org chart
roledef→ identity, voice, output contract, guardrails
───────────────────────────────────────────
catdef→ recommended substrate (external; AI-peer-aware)

Three of these — roledef, orgdef, memodef — are OG-internal. catdef is an external, substrate-agnostic data spec; OG recommends it but does not require it.

OG itself

The OG framework is described by its own self-describing artifacts — readable directly by AI peers:

The four specs

catdef recommended substrate

catdef.org — external

One open file format for describing any object — and any catalog of objects. OG rides catdef's .openthing and .opencatalog files; alternative substrates are permitted in principle.

roledef OG-internal

roledef.org

A portable definition of a single AI role — identity, voice, output contract, guardrails, recommended capabilities. Any compliant runtime can load it; the role travels when the model does.

orgdef OG-internal

orgdef.org

An org chart for AI: positions, relationships (reports-to, peer-of, validates-for, …), and incumbents. Authority is declared in the artifact, not implied by an API key.

memodef OG-internal

memodef.org

Memos as files committed to the recipient's working repo. Every AI-to-AI hand-off is a git commit. git blame any decision the organization made. Transcripts are a memodef:Transcript subtype.

Skills

Seven canonical Claude Code skills bracket the OG lifecycle — three categories. The suite is the discoverable menu of canonical operations.

Genesis /og-adopt

Convert an existing project into OG shape.

Genesis /og-create

Create a new OG-shaped organization from scratch. Folder-only by default, git optional.

Session /og-orient

Read-only come-up-to-speed on an OG-shaped org; emits the snapshot.

Session /og-claim-seat

Take a position — the PO-authorized staffing act.

Session /og-closeout

Wrap a working session: closeout memo + transcript-save prompt.

Operations /og-snapshot

Current-state view: identity + staffing + recent-activity digest.

Operations /og-add-position

Propose a new position — a Director-ratified org-chart change.

One-command install: install-claude-code-skills.ps1 (Windows) or install-claude-code-skills.sh (Mac/Linux).

How to engage

OG is multi-runtime by design. Pick the access path for your AI.

RuntimeHow
Claude CodeInstall canonical skills (above); read org/ and memos/ from any OG-shaped repo directly.
AI with MCPConnect to ogframework.com/mcp (forward work per v0.2; per-spec namespaced tools; read-only convenience cache; canonical content stays in repos).
AI with web access (any runtime)Fetch ogframework.com/primer.md — concise self-onboarding primer. Also: ogframework.com/llms.txt at site root for canonical-content discovery.
Direct clonegit clone https://github.com/ogframework/og.git — charter, CLAUDE.md, memos, proposals, decisions, skills.

Three claims OG makes

Structured

The framework is explicit. Roles, positions, relationships, memos, decisions — each has a schema, a conformance suite, and reference fixtures. Nothing is left to tribal knowledge.

Open

MIT-licensed; no vendor owns OG or any spec; anyone can publish a library at any URL. The conformance suite is the standard. All AI runtimes are equal citizens.

Empirical

Every methodology rule was derived from observed runtime behavior, not theoretical design. The library carries documented conformance evidence on real models. Falsifiable, not aspirational.

Why now

Every major AI vendor is shipping its own proprietary constructs for teams, agents, and projects. If the open layer doesn't exist before those patterns harden, AI organizations will look like the AOL era of online services instead of the web era. The window for an open structural layer is open today and shrinking.

Email, the web, and schema.org all became infrastructure because they were open. No vendor owned the structural layer; implementations were equal citizens; conformance was the spec. The structural layer of AI deserves — and needs — the same.

Status