sqlite AGENTS.md

Background

SQLite is a unique, mission-critical open-source project with an extremely conservative contribution model. Human developers write all code, and external contributions are not accepted in the traditional sense. The rise of powerful AI coding agents has introduced a new dynamic, leading external contributors to point these agents directly at the SQLite codebase to generate patches, bug reports, and fixes. This prompted the project to create formal policies to manage this new type of interaction.

Key Points

• Explicit Policy for Agents: The AGENTS.md file is a dedicated contract for AI agents. It clarifies that SQLite does not accept pull requests without prior agreement, and any accepted code must be legally placed in the public domain. This legal safeguard is crucial.
• A Narrow, Defined Role for AI: The project carves out a very specific, limited space for AI contributions. Agentic code is not accepted, but agentic bug reports with reproducible test cases are. AI can also provide patches purely for documentation or as a proof-of-concept that human developers might review and then rewrite themselves.
• Policy Strengthened Under Pressure: The initial policy contained the qualifier "(currently)" regarding not accepting agentic code. Its removal with the commit message "Strengthen the statement" signals the developers' firm and escalating resolve against AI-authored code, likely in response to the volume of submissions.
• Operational Impact on the Project: The influx of AI-generated reports, of "varying quality," forced a structural change to the project's infrastructure, splitting them into a new SQLite Bug Forum. This indicates a significant volume that disrupted normal processes.
• Human Developer Response: Despite rejecting the AI's code, the project's lead developer, D. Richard Hipp, is actively resolving issues reported via this new channel. This shows a pragmatic approach: utilizing AI for its potential to find bugs while maintaining a strict human-only code authorship policy.

Significance

• A Proactive Stance on AI Governance: SQLite is setting a clear precedent by creating specific, public-facing rules for AI interactions, moving beyond just ignoring or quietly rejecting them.
• Quality Control vs. Noise: The policy is a direct mechanism for quality control. By requiring a reproducible test case for bug reports, they filter for actionable intelligence amidst the noise, leveraging AI's potential for bug detection while ignoring its code generation.
• Highlighting the Unresolved Tension: The situation underscores a fundamental tension in the open-source ecosystem between AI's capability to generate code at scale and traditional models of human-centric software craftsmanship, licensing, and trust. SQLite's model explicitly privileges the latter.
• A Blueprint for Other Projects: This represents a mature, defensive playbook for other critical open-source projects to manage the coming wave of AI-generated contributions: establish clear public policies, strictly gate code acceptance, and define narrow, useful lanes for AI participation like validated bug reporting.