SQLite did something unusual five days ago. They added a file called AGENTS.md to their repository. It’s not for their developers. It’s for yours.

More specifically, it’s for the AI coding agents that are increasingly pointed at open source codebases to suggest improvements, find bugs, or generate pull requests. And the message is blunt: don’t bother submitting code unless you’ve talked to us first.

The file states that “SQLite does not accept pull requests without prior agreement and/or accompanying legal paperwork that places the pull request in the public domain.” The human developers will review a “concise and well-written pull request as a proof-of-concept,” but only before they reimplement the change themselves.

This is worth paying attention to, even if you never touch SQLite’s code.

The Canary in the Coal Mine

SQLite isn’t reacting to a hypothetical problem. The existence of AGENTS.md suggests they’re already seeing agent-generated PRs, or expect to soon. It’s a preemptive strike against a wave of low-context contributions from tools that can read code but don’t understand project norms, legal requirements, or maintainer bandwidth.

SQLite has unusual requirements. They don’t use standard open source licenses. Contributions must be explicitly placed in the public domain. That’s not something an AI agent infers from reading CONTRIBUTING.md, assuming one exists.

The timing matters too. We’re seeing more coding agents launched every month, and more developers using tools like GitHub Copilot Workspace, Cursor, or various Claude-based coding tools to generate entire features or fixes. Some of these tools can autonomously create pull requests. Most don’t understand project-specific legal requirements.

What This Means for Maintainers

If you maintain an open source project, you should probably write an AGENTS.md file too. Even if you welcome agent-generated contributions, you need to set expectations.

The file can specify legal requirements, coding standards, testing expectations, or just ask agents to open an issue before submitting code. It’s a machine-readable way to say “here’s how we work” before you’re drowning in low-quality PRs.

SQLite’s approach is conservative, but it’s their project. They’ve maintained SQLite with a tiny team for over two decades. They don’t want to become a support forum for every agent that thinks it found a performance optimization.

The Bigger Pattern

This isn’t just about SQLite. It’s about the collision between open source culture and autonomous tools that don’t understand context.

Open source works because contributors learn the norms. They lurk, read issues, understand the project’s direction, and then contribute. AI agents don’t do that. They pattern-match against their training data and generate plausible-looking code. Sometimes it’s good. Often it’s not. And it always requires human review.

The problem scales badly. If a thousand developers each point their agent at your project, you’re now reviewing a thousand PRs that may or may not reflect actual understanding of your codebase. That’s not collaboration. That’s spam with extra steps.

Who Should Care

If you maintain an open source project, especially one with legal requirements beyond “Apache 2.0 license, sign the CLA,” you should write an AGENTS.md file now. Don’t wait until you’re dealing with the problem.

If you’re building AI coding tools, you should look for and respect AGENTS.md files. It’s the difference between being helpful and being a nuisance.

If you’re using AI coding agents, remember that they’re tools. They don’t understand project governance, legal requirements, or whether a maintainer actually wants your contribution. Read the docs. Talk to the maintainers. Don’t automate away the human parts of open source.

SQLite’s AGENTS.md file is a small thing. But it’s a signal. Open source projects are starting to set boundaries around AI-generated contributions. That’s healthy. The alternative is maintainer burnout and projects going private or shutting down entirely.

The file is live in the SQLite repository now. It’s short, clear, and probably coming to a project near you.