Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
72 lines (45 loc) · 4 KB

CONTRIBUTING.md

File metadata and controls

72 lines (45 loc) · 4 KB

Contributing to AGS

Thanks for your interest in the Agent Governance Spine specification.

AGS is a pattern specification, not a software library. Contributions are most useful in three forms:

1. Implementation reports

If you have built a governance system that implements AGS (or pieces of it) in production, an issue or PR describing what worked, what didn't, and what you'd refine is the highest-value contribution. Anonymized is fine — patterns and incidents matter more than vendor names.

Template: open an issue with title [Implementation] <one-line summary> and include:

  • Stack (policy engine, identity provider, audit log, agent framework)
  • Which of the 10 principles you implemented and which you skipped, and why
  • What broke and how you fixed it
  • What SLAs you measured against the targets in SPEC.md
  • Particularly valued: incident reports — which of the four named failure modes did you hit in real production, and what was the fix?

2. Pattern refinements and additions

If you find a missing principle, an unhandled failure mode, or a refinement to an existing principle, open an issue first to discuss before sending a PR. The spec is intentionally tight — every principle has earned its place. New principles need to be load-bearing, not nice-to-have.

Refinements to existing principles are easier: open a PR with the proposed change to SPEC.md and a one-paragraph rationale. Cite implementations, empirical research, or production incidents where possible.

3. Examples and reference materials

The examples/ directory is open to:

  • More worked policy examples for additional domains (financial-ops approvals, customer-support handoffs, R&D deployments)
  • Identity-attestation patterns in additional ecosystems
  • Audit-log shape examples (commitment anchoring, hash-chain proofs)
  • Privilege-ring examples for specific agent classes
  • Trust-scoring patterns for plugin marketplaces

Keep examples small and concrete. The point is to show the shape; production-grade implementations belong in your own repo.

What we won't accept

  • Vendor advertising — examples that exist primarily to promote a product. Keep examples vendor-neutral; if you need to name a tool (OPA, Cedar, Permit.io, AGT), use it as one of many examples.
  • Speculative principles — additions without an implementation that supports them.
  • Cosmetic edits without rationale.

Style

  • Prose: declarative, no fluff. Match the existing voice in SPEC.md.
  • Diagrams: Mermaid where possible (renders natively in GitHub).
  • Code samples: minimal, runnable, no external dependencies beyond requirements.txt / package.json declarations.
  • Markdown: GitHub-flavored. Wrap at natural sentence boundaries, not at fixed columns.

License agreement

Contributions are accepted under the project's dual license (CC BY 4.0 for prose, MIT for code). By opening a PR, you agree to license your contribution under those terms.

Reporting issues with the spec

If you think a principle is wrong, please say so directly in an issue. Include the principle number, what's wrong, and what you'd change. The spec is not gospel; it's the current best understanding.

Relationship to PDS, ACS, ESF, and CRI

AGS is the fifth specification in the SaaSquach AI Labs catalog:

  • PDS (Progressive Discovery Spine) — single-agent tool discipline
  • ACS (Adversarial Coordination Spine) — multi-agent coordination
  • ESF (External Signal Fabric) — external signal substrate
  • CRI (Composite Risk Index) — composite scoring (private)
  • AGS (Agent Governance Spine) — protocol-layer governance (this spec)

AGS sits across all four — every agent in a PDS / ACS / ESF / CRI system passes through the governance substrate on every action. Cross-cutting contributions are welcome — open an issue on the most-affected repo first.

Contact

Open an issue for anything spec-related. For something that doesn't fit an issue, the author's contact is in the repository About section.