Contribute

AGF is built on community over credit — if this framework helps one organization build a safer agentic system, it has served its purpose. We welcome challenges to claims, evidence from real-world implementations, proposals for new patterns, and corrections to what we got wrong.

The canonical CONTRIBUTING.md on GitHub is the source of truth. This page summarizes the highest-leverage contribution paths.

Where to start

What we welcome

• Challenge & Critique. Where is AGF wrong? What are we overclaiming? What did we miss?
• Evidence & Implementation. Real-world experiences implementing the patterns. What worked, what didn't, what surprised you.
• Standards updates. When AICM, NIST AI RMF, ISO 42001, EU AI Act, or similar issues a new revision, mappings need refresh.
• Pattern proposals. New primitives or composition patterns — accompanied by the problem they solve and the evidence supporting them.
• Diagram improvements. Architecture diagrams are first-class artifacts.

Content conventions

Canonical source

Edit docs/*.md files (the canonical source). The site MDX files (agf-docs/content/docs/*.mdx) are derived. If you submit a PR against an .mdx file, it will be redirected to the canonical source.

Voice

• Humility before authority. AGF synthesizes; it does not decree. Credit the work AGF builds on.
• Rigor before opinion. Every claim either traces to a cited source OR carries an explicit <Confidence level /> label.
• No undefended novelty. Where AGF introduces a new framing, it is marked <Confidence level="informed" /> until adopter evidence promotes it to established, or counter-evidence demotes it to open.

Confidence levels

Use one of three labels on any new claim:

• <Confidence level="established" /> — proven across multiple domains; multiple sources would agree.
• <Confidence level="informed" /> — AGF's synthesis; plausible but unverified at scale.
• <Confidence level="open" /> — flagged but speculative; needs investigation.

See Confidence Levels for the full rationale.

Standards references

When citing existing work:

• Include specific document names, version numbers, and dates (not just organization names).
• Explain how the reference relates to AGF (not just that it exists).
• Prefer primary sources over summaries.

Cross-reference integrity

Framework documents reference each other by primitive number (#1–#19). If your change affects primitive numbering or naming, note which downstream documents need updates.

Review process

All contributions are reviewed by the maintainer. For framework content, the four review dimensions are:

1. Accuracy — Is the claim correct? Is the standards mapping accurate?
2. Voice — Does it maintain AGF's synthesis positioning?
3. Confidence — Are new claims appropriately marked?
4. Cross-references — Does it maintain integrity with adjacent docs?

External reviews land in docs/reviews/; tracked findings land in docs/findings-ledger.md with stable F-IDs that survive across revisions.

Code of conduct

AGF deals with governance and safety; the community that builds it should reflect those values.

• Critique the ideas, not the people.
• Back claims with evidence or clearly mark them as proposals.
• Acknowledge what you don't know.

License

By contributing, you agree your contributions are licensed under the same terms as the project — CC BY 4.0 for documentation; Apache-2.0 / MIT for any code or tooling. See LICENSE for details.

---

The most valuable contributions are usually evidence from a real implementation — even partial, even with caveats. AGF's biggest gap today is the absence of adopter case studies; if you've built on it, your experience is exactly what would close that gap.