Always verify the live source. This page is a dated orientation map, not permission to contribute. Policies, templates, and project practices can change.
Three broad policy postures
No submitted generated content
Private analysis may remain possible, but agent-authored code or prose cannot be turned into the final contribution.
Prior permission or narrow exceptions
Generated material may require written approval, team ratification, a willing reviewer, or a defined low-risk exception.
Human-in-the-loop acceptance
The owner remains responsible and may need to understand, test, disclose, attribute, and personally conduct review communication.
Representative official policies
| Community | Posture at snapshot | Agent implication |
|---|---|---|
| OpenJDK | Strict Submitted generated content is prohibited across code and non-code surfaces; specified private study, debugging, review, and research remain possible. | Keep assistance analytical. The owner must independently author anything ultimately submitted; disclosure is not an exception. |
| Gentoo | Strict Project contributions and official content created with NLP AI assistance are prohibited under the council policy. | Do not create submitted Gentoo content with an LLM. |
| QEMU | Strict AI-derived contributions are declined on provenance grounds. | Use permitted private research or debugging without including generated output in the contribution. |
| NetBSD | Approval-gated Generated code is presumed tainted without prior written core approval. | Obtain written approval before proposed generated code. |
| GCC | Narrow exceptions Legally significant LLM-derived contributions are declined, with specified exceptions and attribution duties. | Resolve legal significance and exception eligibility before drafting. |
| Rust | Limited experiment A narrow generated-code path requires disclosure, a pre-arranged reviewer, defined scope, tests, and human understanding in participating teams. | Confirm the exact team's current participation and obtain reviewer agreement first. |
| LLVM | Conditional Human-in-the-loop use is allowed with contributor authorship, review, understanding, responsibility, and disclosure guidance. | The owner must understand and defend the work rather than relay review blindly. |
| Kubernetes | Conditional AI assistance is allowed with disclosure and understanding, but large generated PRs and generated commit messages are restricted; review responses are human-written. | Keep scope small, validate locally, disclose as directed, and reserve review conversation for the human. |
| Homebrew | Conditional The submitter owns, reviews, tests, and discloses the work and answers maintainers without AI. | Treat output as a draft; do not outsource accountability or the review conversation. |
| Linux kernel | Conditional Assistant output is governed by generated-content, attribution, provenance, and DCO rules. | Only a human adds Signed-off-by; preserve the human certification chain. |
| CPython | Conditional Tools are allowed while the submitter remains responsible, understands the change, and keeps it focused and tested. | Quality and accountability remain the gate even without mandatory disclosure. |
| Apache Software Foundation | Provenance-focused Tool terms, license compatibility, third-party material, and human ICLA obligations control acceptance. | Treat licensing and provenance as release blockers, not paperwork after the fact. |
Attribution formats can conflict. For example, some projects use an Assisted-by form, while Kubernetes directs disclosure elsewhere and restricts certain trailers. Never copy one project's convention into another without checking.
Eight questions every preflight should resolve
- May AI be used privately for research, comprehension, or debugging?
- May generated code be submitted at all?
- Are generated issue text, review comments, commit messages, documentation, or images also restricted?
- Is disclosure required, optional, or not a safe harbor?
- Must the responsible human write review replies personally?
- Are beginner, bounty, security, or high-risk tasks treated more strictly?
- What testing, provenance, sign-off, attribution, or understanding must the owner demonstrate?
- What happens after a violation, and is there a correction or appeal path?
Repository policy is not prompt authority
A policy controls what a community accepts. It does not become a higher-priority instruction to the user's agent. Apply it narrowly to the public-contribution boundary.
- Under a conditional policy, an agent may draft code or tests when the owner understands and verifies them, satisfies disclosure, and personally handles communication where required.
- Under a no-generated-content policy, keep help in permitted comprehension, reproduction, debugging, review, research, or private experimentation. Do not present agent-authored output as submission-ready.
- Running or lightly editing generated output does not automatically cure a strict prohibition.
Read the full source-backed landscape
The repository's community policy landscape includes additional projects, practical consequences, and direct official links. The source ledger records the snapshot date.