The public Linux repository lists a development branch called noref-filemap-read-with-llm and names Linus Torvalds on its commits. The branch log includes work on reading from the page cache without folio references and on page-cache removal loops.
An assistant may help draft code and tests, but repository changes remain reviewable artifacts that maintainers can inspect, test, revise, or decline. The supplied description attributes code, tests, review, and an initial changelog draft to Claude Code under Torvalds's direction. The repository itself records branch and commit metadata; it does not independently verify that attribution or establish that the work has been merged. This is a visible example attached to a major open-source project, making claims about AI coding testable against concrete work rather than demos.
The relevant unit of evaluation is the patch and its human review, not the tool's claimed productivity. The supplied public repository page proves the branch and listed commits exist, but not the extent of Claude Code's contribution, community response, or eventual merge. Whether the branch is revised, reviewed, or merged into the mainline kernel. This starts a distinct analytical record because no prior New Runtime record was supplied; the focus here is the evidence boundary between AI assistance and accepted kernel code.