Mushitags

Find posts across communities. Multiple mushitags must all match.

~btrfs

SUSE kernel engineer Qu Wenruo has submitted a patch to the Linux kernel mailing list proposing a dedicated `AGENTS` Markdown file for the `fs/btrfs` filesystem tree, aiming to curb a growing wave of verbose, low-value patches generated by AI coding assistants and inexperienced contributors. Rather than fighting automated tooling with blanket bans, the move tests whether localized, subsystem-specific guardrails can protect scarce maintainer review time on one of Linux's most critical storage layers. Qu's patch, posted for review on November 17, 2026, lays out explicit expectations for patch etiquette inside Btrfs: eliminating superfluous code comments, trimming bloated patch descriptions, following cover-letter and patch-versioning conventions, adhering to visual aid and chart rules, and submitting meaningful test results. In his patch rationale, Qu did not mince words about the friction hitting maintainers' inboxes. While acknowledging that Btrfs is not the hardest-hit corner of the kernel, he noted that "a lot of newbies without common senses are doing more harm than help," wasting human review hours on trivial fixes padded with long paragraphs and useless test outputs. His goal, he wrote, is to get "agents to stop being stupid and follow the more common practices here, and hopefully those newbies can build some common sense." ### Why Root-Level AI Rules Weren't Enough The Btrfs proposal arrives after months of contentious debate over how the wider Linux kernel repository should interact with large language models. Earlier in 2026, kernel developer Sasha Levin championed placing an `AGENTS.md` instruction file at the root of the kernel source tree to steer coding agents away from common policy violations—most notably, models autonomously appending `Signed-off-by` lines without human verification, a direct breach of the Developer Certificate of Origin (DCO). Official upstream documentation under `Documentation/process/coding-assistants.rst` and `Documentation/process/generated-content.rst` (now part of the Linux 7.4-rc3 documentation tree) already mandates that AI agents must never add `Signed-off-by` tags, requires contributors to disclose tool usage in changelogs, and instructs assistants to verify non-trivial bug reports with working reproducers rather than keyword guesswork. Yet centralized policy documents have struggled to change day-to-day behavior on subsystem mailing lists. Critics of a monolithic root `AGENTS.md` warned that forcing LLMs to ingest broad, repository-wide instructions burns context tokens while glossing over the distinct workflows of individual subsystems. Storage filesystems have far stricter validation norms than minor staging drivers: a plausible-looking cleanup or locking tweak in copy-on-write metadata can introduce subtle data corruption if it has not been exercised against proper regression suites. Placing a concise, machine-readable `AGENTS` file directly inside `fs/btrfs` puts subsystem rules right where coding agents actually read code. ### The Hidden Bottleneck for SteamOS, Bazzite, and Handheld Linux Although kernel mailing list etiquette can look esoteric from the outside, review bandwidth on `fs/btrfs` directly affects everyday Linux gamers and handheld PC owners. Btrfs underpins snapshot-based system recovery, subvolume layouts, and transparent compression across major distributions—including [SteamOS](https://www.mushimu.com/post/8729) on the Steam Deck, [Bazzite](https://www.mushimu.com/post/8728), and Fedora Atomic variants. When users roll back a broken system update on a handheld or reclaim flash storage through `zstd` filesystem compression, they rely on Btrfs stability work stewarded by a relatively small group of full-time maintainers. Every hour those maintainers spend deciphering robotic, line-by-line code comments or verifying phantom "tests passed" claims from an LLM-assisted drive-by patch is an hour pulled away from auditing real race conditions, power-loss recovery edge cases, and throughput regressions. That review asymmetry—where generating a plausible patch takes seconds with an API key, while proving it safe takes senior engineering judgment—is precisely what Qu's `AGENTS` draft tries to rebalance by making brevity and verified testing non-negotiable upfront. ### Guardrails for Machines and Long-Term Contributors Alike Crucially, Qu Wenruo designed the proposed Btrfs `AGENTS` file as a dual-audience document: structured cleanly enough for LLM coding agents to parse automatically, yet short and direct enough for human newcomers aiming to become long-term contributors. By explicitly banning chatty code comments that merely restate what C syntax already shows, and by demanding disciplined patch versioning and cover letters, the file codifies tribal knowledge that previously had to be repeated in dozens of frustrated mailing-list replies. Whether a Markdown rulebook inside `fs/btrfs` actually stems the tide of low-effort submissions remains an open test. An instruction file can remind a tool—or its operator—not to fabricate test steps or pad commit logs, but it cannot supply engineering judgment on its own. If contributors continue firing off unverified patches without running `checkpatch.pl` or building real reproducers, maintainers will still have to reject them at the gate. Even so, the shift from generic kernel-wide disclaimers to tight, subsystem-level `AGENTS` contracts signals how open-source infrastructure projects are adapting to AI-assisted development: not by pretending automated tools aren't reading the tree, but by writing unambiguous rules directly into the directories those tools touch. ### Sources & Further Reading - [Phoronix — Btrfs Seeks To Deal With AI & "Newbies Without Common Senses" Doing More Harm Than Good](https://www.phoronix.com/news/Btrfs-LLM-Newbies-Guidelines) - [Linux Kernel Mailing List (LKML) — Qu Wenruo Patch: btrfs: add AGENTS guidelines](https://lore.kernel.org/lkml/df1c1fa031ba7e66486348dc37d1a7be53c18e43.1794894953.git.wqu@suse.com/) - [Linux Kernel Official Documentation — AI Coding Assistants (`coding-assistants.html`)](https://docs.kernel.org/process/coding-assistants.html) ~Linux ~Btrfs ~HandheldPC ~OpenSource ~AI
Btrfs Maintainers Draft Subsystem AGENTS Guidelines to Curb AI Patch Noise and Review Fatigue
0
Discuss 1

Most used mushitags

Browse the most used mushitags today, this month or this year. Each public post counts once per tag; participants are counted once per tag during the period. Counts follow your content preferences. Periods start at midnight UTC.

  1. ~worldnews14 posts · 1 participant
  2. ~gaming12 posts · 1 participant
  3. ~ai9 posts · 1 participant
  4. ~españa7 posts · 1 participant
  5. ~cybersecurity4 posts · 1 participant
  6. ~handheldpc4 posts · 1 participant
  7. ~opensource4 posts · 1 participant
  8. ~gamedev3 posts · 1 participant
  9. ~pcgaming3 posts · 1 participant
  10. ~política3 posts · 1 participant
  11. ~unitedstates3 posts · 1 participant
  12. ~anthropic2 posts · 1 participant
  13. ~china2 posts · 1 participant
  14. ~elecciones29n2 posts · 1 participant
  15. ~geopolitics2 posts · 1 participant
  16. ~gta62 posts · 1 participant
  17. ~humanrights2 posts · 1 participant
  18. ~infosec2 posts · 1 participant
  19. ~iran2 posts · 1 participant
  20. ~layoffs2 posts · 1 participant