Contributing to Agent Platform

Thanks for looking at this. The project is a v0.1 draft specification — early enough that a well-argued issue can still change the shape of the standard. This document explains how to contribute usefully at this stage.

Table of Contents

Where the Project Stands

Per the roadmap, the specification itself is still open for comment — nothing below v1.0 is stable. That means:

If you're not sure whether something is "ready" to build against, assume not, and open a discussion first.

Ways to Contribute

Roughly in order of what's most useful today:

  1. Stress-test the spec — try to design an Agent Endpoint for a real business (yours, or a hypothetical one) using the current draft, and report where it breaks down
  2. Argue with a design decision — Skills vs. Tools, the endpoint list, the auth model, anything — with a concrete scenario, not just a preference
  3. Write reference examples — a filled-out agent.company.com manifest for a sample business (retail, SaaS, booking, etc.)
  4. Build a piece of the reference implementation — Core Server, an SDK, the CLI, the Agent Simulator (see Reference Implementation)
  5. Improve the docs — the README, this file, or inline spec comments

Proposing a Spec Change

Spec changes go through discussion before code, not after:

  1. Open an issue describing the problem, not just the proposed fix. What breaks today, for what kind of business, in what scenario?
  2. Propose a change, ideally with a before/after example of what an agent.company.com response looks like under the current spec vs. your proposal
  3. Wait for discussion before implementing — this avoids two people independently building incompatible solutions to the same problem
  4. Once there's rough consensus, the spec section gets updated in a PR that links back to the issue

Small clarifications (typos, ambiguous wording, missing examples) can skip straight to a PR.

Contributing Code

Since there's no reference implementation yet, "contributing code" mostly means starting a piece of one. Before you do:

Style Guide

Commit Messages

Keep the first line short and in the imperative mood:

Add skills/tools distinction to endpoint spec
Fix ambiguous auth example in README

Put the "why," not just the "what," in the body if the change isn't self-explanatory.

Pull Request Process

  1. Fork the repo and branch from main
  2. Keep the PR scoped to one change — spec edits and implementation code should generally be separate PRs
  3. Link the issue or discussion the PR resolves
  4. If the PR changes the spec, update any examples or diagrams elsewhere in the repo that would now be inconsistent
  5. A maintainer will review; for spec changes specifically, expect discussion before merge rather than a quick approval

Reporting Issues

A good issue includes:

"This feels off" is a fine way to start a discussion, but a maintainer will likely ask for a scenario before it turns into a spec change.

Communication

For now, everything routes through this repository:

If a chat/community channel gets stood up, it'll be linked from the main README.

Code of Conduct

Be direct about the work, kind to the person. Disagree with proposals, not with people. Assume good faith from other contributors, and extend the same to maintainers when a PR doesn't get merged right away — spec-first projects move slower than product code by design.