Linux Kernel Developers Weigh Adding AGENTS.md to Guide AI Agents

The Linux kernel continues to be flooded with patches generated by AI/LLM agents, yet until now its source tree has never carried an AGENTS.md file offering instructions tailored to such tools. A patch set posted today aims to fill that gap at last.

The effort comes from longtime kernel developer Sasha Levin, who has previously worked on other AI initiatives for the kernel — including help in selecting patches for back-porting to the stable tree and AI-assisted merge conflict resolution. He now proposes adding an AGENTS.md to the source tree.

The proposed file essentially just links to the kernel's README, from which one can reach documentation on coding assistants and other information relevant to developers.

Why bother

The rationale rests on hands-on experience. In testing, one AI agent working without the file added a "Signed-off-by" tag to a proposed patch — something it should not do without the user explicitly signing off themselves. It also fell back on its own attribution tag instead of the kernel's standardized "Assisted-by" tag.

A second, unnamed agent added no attributions at all without AGENTS.md. But once the file was in place, both agents behaved according to kernel standards, and the second one conformed noticeably better to kernel development best practices.

Not everyone is on board

The proposal is being discussed in a dedicated LKML thread. Even though the file merely links to the README, support is far from unanimous. Some developers object that making AI models consume the entire kernel README and its associated documentation would drive up token consumption. More narrowly targeted, AI-oriented documentation could ease that cost — but where this new proposal lands remains to be seen.

Source: Phoronix