Guidelines for AI-Assisted Open Source Projects Entering CNCF Sandbox

[raravena80]

Guidelines for AI-Assisted Sandbox Open Source Projects

Background

The rapid advancement in AI coding tools has enabled community members to develop more quickly than before, creating new opportunities but also raising new considerations for evaluating open source projects.

We're also observing an increase in open source projects that are primarily generated through AI assistance (ChatGPT, GitHub Copilot, Claude, etc.) with minimal human community involvement. These are some of the characteristics that these projects may have:

• Rapid development cycles (weekend/week-long projects)
• Limited contributor diversity (single or very few developers)
• Minimal community engagement or organic adoption
• Documentation and code patterns that suggest heavy AI assistance
• Lack of real-world usage validation beyond proof-of-concept

While AI assistance in development is not inherently bad, projects entering the CNCF ecosystem should demonstrate genuine community value and sustainable development practices.

Suggested Solution

TOC to establish clear guidelines for evaluating AI-assisted projects seeking Sandbox status.

Community

• Meaningful contributor base: Evidence of multiple active contributors beyond the original author(s)
• Organic adoption: Real-world usage by organizations or individuals
• Community engagement: Show active participation in issues, discussions, and feature requests from users outside the core team

Project Maturity Indicators

• Sustained development: Evidence of ongoing development over a reasonable timeframe (suggested minimum: 3-6 months)
• User feedback integration: Demonstrate responsiveness to community needs and iterative improvement based on actual usage
• Technical depth: Projects should show complexity and thoughtfulness.

Transparency Requirements

• Development methodology disclosure: While not requiring an explicit declaration of AI usage, projects should be transparent about their development approach
• Documentation quality: Requires documentation that demonstrates a deep understanding of the problem space

Benefits

• Maintains ecosystem quality
• Encourages community building
• Preserves CNCF's reputation
• Supports genuine innovation

Implementation Considerations

• These guidelines are to complement, not replace, existing Sandbox criteria.
• Focus on outcomes (community, usage, value) rather than development methods.
• Provide clear, objectively evaluated criteria.
• Consider grandfathering existing projects while applying new standards prospectively.

TOC Discussion Points For These Types of Projects

• What specific metrics should we use to evaluate community engagement?
• How do we balance encouraging innovation with maintaining quality standards?
• Should there be different criteria for different types of projects (tools vs. platforms vs. libraries)?
• How do we ensure these guidelines don't inadvertently discourage legitimate innovation?

[TheFoxAtWork]

Thanks @raravena80 for getting this started

Limited contributor diversity (single or very few developers)
Minimal community engagement or organic adoption
These are definitely items that must be addressed within our ecosystem (since their inverse are attributes that drive successful, healthy, and sustainable CNCF projects)

The other items I wanted to mention is code sprawl / spaghetti code review and management. Without constraints on the output, vibe coders can have PRs and projects that span a multitude of languages and technologies, making it exceedingly difficult to conduct a code review or even manage the codebase (dependency and vulnerability management seems like a nightmare).

@angellk can we add this to an upcoming public TOC meeting to weigh these? what would a AI-heavy, vibe coded incubation or graduation project look like? We've done a lot on focusing on outcomes of the criteria, with the points Ricardo raised on this issue, we can follow a similar approach on outcomes, with updates to the guide posts about "how" it could be done. Similar to what "good project governance" looks like.

[TheFoxAtWork]

from @linsun
https://github.com/agentgateway/agentgateway/blob/main/CODE_OF_CONDUCT.md
https://www.linuxfoundation.org/legal/generative-ai