Skip to content

AI_DEVELOPMENT.md

Latest commit

 

History

History
157 lines (117 loc) · 6.23 KB

AI_DEVELOPMENT.md

File metadata and controls

157 lines (117 loc) · 6.23 KB

AI-Driven Development Model

W3C OS uses an AI-first development model: AI agents are the primary developers; humans provide requirements, review, and governance.

How It Works

Human                          AI (Management)                AI (Contributor)
  │                                  │                              │
  ├─ Files Issue ──────────────────► │                              │
  │                                  ├─ Triages, labels, assigns ──►│
  │                                  │                              ├─ Reads issue + codebase
  │                                  │                              ├─ Implements solution
  │                                  │                              ├─ Writes tests
  │                                  │                              ├─ Opens PR ──────────►│
  │                                  ├─ Reviews PR (CI + AI) ◄──────┤                      │
  │  ◄─ Notified for final review ──┤                              │                      │
  ├─ Approves / requests changes ──►│                              │                      │
  │                                  ├─ Merges PR ─────────────────────────────────────────┘

Two Kinds of AI

Management AI

The Management AI is the project maintainer. It runs automatically via GitHub Actions and OpenClaw Lobster workflows.

Responsibilities:

  • Triage: When a new Issue is filed, classify it (bug/feature/task), estimate difficulty, assign labels
  • Assignment: Route ai-ready issues to the next available Contributor AI
  • Code Review: Run CI checks, verify test coverage, check for style/architecture violations
  • Merge Control: Auto-merge PRs that pass all checks and have human approval
  • Release Management: Tag releases, generate changelogs

Trigger: Runs on issues.opened, pull_request.opened, pull_request_review.submitted

Contributor AI

Contributor AIs are the developers. They pick up issues and write code.

Workflow:

  1. Receive an assigned issue (via GitHub Actions dispatch or OpenClaw)
  2. Read the issue description, acceptance criteria, and relevant codebase
  3. Create a branch, implement the solution, write tests
  4. Open a PR with a clear description
  5. Respond to review feedback and iterate

Tools: Cursor Agent, OpenClaw + Lobster, GitHub Copilot Workspace, or any AI coding tool

Funding Model

AI agents consume API tokens (LLM inference). The community funds these tokens through sponsorship.

Where the Money Goes

Sponsors (GitHub Sponsors / Open Collective)
    │
    ▼
Token Pool
    ├── 60%  Contributor AI tokens (coding, PR generation)
    ├── 25%  Management AI tokens (triage, review, merge)
    └── 15%  Infrastructure (CI runners, hosting)

How to Sponsor

Every dollar goes to AI compute. No human salaries. Fully transparent spending.

Sponsorship Tiers

Tier Amount What It Funds
Byte $5/mo ~1 AI-implemented issue per month
Kilobyte $25/mo ~5 AI-implemented issues per month
Megabyte $100/mo ~20 AI-implemented issues per month
Gigabyte $500/mo Sustained AI development capacity

Issue Lifecycle

1. [opened]     Human or AI files an Issue
2. [triaged]    Management AI labels: bug/feature, priority, difficulty, module
3. [ai-ready]   Management AI confirms the issue is well-defined for AI
4. [assigned]   Contributor AI is dispatched
5. [in-progress] AI creates branch and starts coding
6. [pr-opened]  AI opens PR with implementation
7. [review]     Management AI + CI check the PR
8. [approved]   Human maintainer approves
9. [merged]     Management AI merges

Labels

Label Meaning
ai-ready Well-defined, AI can pick up immediately
ai-assigned An AI agent is working on this
ai-pr PR was authored by an AI agent
needs-human Requires human decision (architecture, security, policy)
good first issue Easy task, suitable for new contributor AIs or humans
sponsored Funded by a specific sponsor

For Humans

Your role is critical:

  • File Issues: Describe what you want. The clearer, the better the AI output.
  • Review PRs: AI-generated code needs human judgment for architecture and security.
  • Governance: Decide project direction, approve breaking changes, set priorities.
  • Sponsor: Fund the AI tokens that keep development moving.

You don't need to write code (but you can).

For AI Agents

To contribute as an AI agent:

  1. Look for issues labeled ai-ready
  2. Read ARCHITECTURE.md to understand the codebase
  3. Read the issue's acceptance criteria carefully
  4. Create a branch: ai/<issue-number>-<short-description>
  5. Implement the solution with tests
  6. Open a PR using the PR template
  7. Prefix your PR title with [AI]
  8. Wait for review and iterate

Quality Bar

AI-submitted PRs must:

  • Pass all CI checks (cargo check, clippy, fmt, test)
  • Include tests for new functionality
  • Not introduce new clippy warnings
  • Follow existing code patterns and architecture
  • Have a clear PR description explaining the changes

OpenClaw Integration

W3C OS uses OpenClaw + Lobster for AI workflow orchestration.

OpenClaw provides:

  • MCP Gateway: Connects AI agents to GitHub, CI, and other tools
  • Lobster Pipelines: Deterministic multi-step workflows with approval gates
  • Token Management: Track and allocate API token spending

See .openclaw/ for workflow configurations.

Transparency

All AI activity is public:

  • Every AI-authored PR is labeled ai-pr
  • Token spending reports are published monthly (when sponsorship is active)
  • Management AI decisions are logged in issue comments
  • No hidden AI activity — everything goes through GitHub