Why

The project has seen a steady rise in low-effort, AI-generated pull requests, most of them from people opening their very first contribution. The real cost is not the code itself but the validation burden it pushes onto maintainers: someone has to read, run, and reason about work that the author never qualified. AI_POLICY.md already says AI-assisted PRs must track an accepted issue, but the problem is not specific to disclosed-AI PRs. A first-time drive-by PR imposes the same cost whether or not a tool was involved, and a first contribution is exactly the point where there is no prior trust to lean on, so the burden is highest and the signal is lowest. The policy therefore gates first contributions in particular, asking for an accepted, assigned issue before any code is written, so the design conversation happens where redirection is cheap rather than after the work is already done.

A blanket gate, though, would turn away the very people the project most wants: contributors who already understand the problem domain. So instead of a hard wall the policy offers a trust-based shortcut. If someone is visibly active in the fediverse or in related F/OSS work, their familiarity is already evident, and insisting on a preparatory issue would add friction without adding signal. This exception is deliberately worded as an invitation rather than a credential check, because forcing people to prove themselves would reintroduce the gatekeeping tone we want to avoid, and because whether a given presence is convincing is ultimately a maintainer's judgment call, not a checklist.

The policy also states plainly that a non-conforming PR may be closed without further comment. This is intentional, not careless: the scarce resource being protected is maintainer attention, and writing a bespoke explanation for every slop PR would reintroduce exactly the cost the policy exists to remove. Stating the rule up front is what makes a silent close fair, since anyone who has read the section has already received the explanation a comment would have given. The same paragraph still invites the contributor to reopen once they qualify, so the door stays open.

Finally, because much of this traffic now arrives through coding agents acting on a contributor's behalf, the rule is placed where those agents will actually read it. AGENTS.md (and its CLAUDE.md symlink) gains a short, high-priority note pointing at the new section, so an agent is told to satisfy the gate before it opens a PR. Addressing the problem at its most common source is more effective than catching it only at review time.

How

The guidance lives in a single new First contributions section near the top of CONTRIBUTING.md, so a newcomer meets it before the bug-report and feature-request instructions rather than after them. It is framed throughout as how to make your first contribution land, echoing the welcoming tone of AI_POLICY.md, instead of as a list of prohibitions. The existing carve-out for typo and documentation fixes is preserved, so the lowest-risk, highest-goodwill contributions stay frictionless. The section links to AI_POLICY.md rather than restating it, keeping the accepted-issue requirement defined in one place.

What changed

• CONTRIBUTING.md: adds the First contributions section.
• AGENTS.md: adds an important note in the AI policy compliance section that points first-time-contributor agents to that section.