Governance
How AGF is governed — BDFL model, decision process, contribution expectations, versioning policy, and the planned evolution toward broader stewardship.
AGF needs to answer two questions before any practitioner takes it seriously: who decides? and what does the path beyond v0.x look like? This page answers both. The canonical GOVERNANCE.md on GitHub is the source of truth.
Model
AGF operates under a BDFL (Benevolent Dictator For Life) model for v0.x. Jesse Pike holds sole editorial authority over framework vocabulary, primitive semantics, DECISIONS.md entries, versioning, and contribution acceptance.
This is scope control, not arrogance — governance frameworks need a stable opinionated core before a committee can make meaningful contributions. The evolution path below makes the handoff explicit.
Decision process
All substantive decisions are logged in DECISIONS.md as the public canonical decision record.
How proposals enter:
- File a GitHub issue with your crosswalk, primitive gap, or vocabulary proposal.
- If non-trivial, open a PR with a concrete draft — discussion happens in the PR.
- Jesse reviews, decides, and writes the
DECISIONS.mdentry if accepted. - High-stakes decisions get a multi-model external review round before commit (see recent entries for the pattern).
Jesse has final call on all framework content. No appeal process — that is what BDFL means.
Contribution expectations
| Contribution type | Preferred channel |
|---|---|
| Crosswalk (NIST AI RMF, OWASP Agentic, ISO 42001, EU AI Act, AICM) | GitHub issue with draft table |
| New primitive proposal | GitHub issue — describe the gap, not just the name |
| Vocabulary addition or refinement | PR against docs/shared-vocabulary.md |
| Clarification, typo, or broken link | PR directly |
| Substantive framework change | Issue first; PR only after Jesse signals interest |
Confidence labels are mandatory on substantive claims. Contributions that omit them on novel or unverified assertions will be asked to add them before merge. See Confidence Levels for the full posture.
Versioning policy
AGF uses SemVer-ish versioning. The framework is primarily prose plus schemas, so the increment definitions are adapted:
| Increment | Trigger |
|---|---|
| Major (1.x.x) | Vocabulary or primitive semantics break backward compatibility; requires migration guide |
| Minor (0.x.0) | New primitive, new concept doc, new profile, new crosswalk section |
| Patch (0.x.y) | Clarifications, tone fixes, cross-reference updates, schema non-breaking additions |
v1.0 is earned, not scheduled. Prerequisites:
- External review by practitioners outside the maintainer's network.
- At least two verified crosswalks (NIST AI RMF + one other).
- Outside contributors engaged (not just maintainer + their agents).
- AGF primitives stress-tested against real agentic system designs.
Current: v0.2.0 — framework is usable and stable enough to build on, but evolving.
Evolution
The governance model is expected to evolve at these transitions:
- v0.x (now) — BDFL. Solo or invited-collaborator contributions.
- ~v0.4–v0.5 — Add Core Maintainers (2–3 practitioners with demonstrated crosswalk or domain expertise). Merge rights delegated to them for profile docs and crosswalks.
- ~v1.0+ — Consider Community Working Group for major vocabulary decisions. Maintainer retains veto.
This page updates at each governance transition.
References
DECISIONS.md— all framework decisions, numbered and datedChangelog— version historyRoadmap— what's nextContribute— contributor how-to
For the canonical entry-by-entry text, see the GOVERNANCE.md on GitHub.