systemd 260-rc3, released March 12, ships AGENTS.md — a structured guide covering the project's architecture, coding conventions, development workflow, and contribution requirements written specifically for AI coding agents. It's the first time systemd has formally documented itself for AI readers, treating agents as contributors with their own onboarding path rather than tools that developers happen to use.
The harder-edged change is a new policy requiring explicit AI disclosure on contributed patches. Developers using AI assistance must tag commits with a disclosure trailer modeled on Git's existing 'Co-developed-by' syntax — an auditable record embedded in commit history rather than a checkbox buried in a contributor agreement. That specificity matters: it creates a traceable record of AI involvement at the patch level, not just at the policy level.
Two companion files round out the update. A CLAUDE.md points Anthropic's Claude Code specifically to AGENTS.md as its primary reference, and a new claude-review.yml embeds Claude Code into the pull request review pipeline as a standing reviewer. That last piece is the operationally significant change — it shifts AI from something individual contributors opt into toward a persistent fixture in the project's review infrastructure.
The disclosure tag requirement is the detail most likely to get traction elsewhere. Most projects adding AGENTS.md-style documentation have treated it as optional guidance; mandating a per-commit tag is a different level of commitment. The 'Co-developed-by' trailer it mirrors is already standard in kernel patch workflows, which means the tooling and reviewer expectations are already in place for other upstream projects to follow the same pattern. Whether they do — or whether the requirement draws pushback from contributors who consider it overreach — is now a live question.