Linux kernel developers are weighing an in-tree AGENTS.md file to guide AI coding agents working on the kernel, and a merged convention or an explicit rejection will settle whether the highest-authority OSS project formally accommodates AI-assisted development.
state: seedheat: lowuncertainty: mediumconvergesscott: mediumagents-md linux-kernel coding-agent-workflows
What is this?
Per Phoronix, Linux kernel developers are discussing adding an AGENTS.md file to the kernel tree — a machine-readable instruction file telling AI coding agents (Cursor, Codex, Claude Code, Copilot, etc.) the repo's build/test commands, conventions, and boundaries. AGENTS.md originated at OpenAI in 2025 and is now governed by the Linux Foundation's Agentic AI Foundation, with adoption figures ranging from 20k+ repos (InfoQ, Aug 2025) to 60k+ (project's own claim). The discussion lands on top of an already-merged kernel AI policy (docs.kernel.org/process/coding-assistants): AI agents may not add Signed-off-by tags (only humans certify the DCO), and AI-assisted patches require an Assisted-by attribution tag. The supplied snippets do not show the outcome of the AGENTS.md discussion itself — whether it gets merged or rejected remains open as far as this evidence goes.
Why it matters to Scott
Converges with Scott's canonized agent-instruction-file doctrine — the CLAUDE.md pattern, the fat-AGENTS.md anti-pattern, and the driving ebook's keep-it-slim prescription — at the highest-authority test bed available: whether a kernel-tree AGENTS.md stays slim or accretes is a live test of his bloat claim, and either resolution is dated-receipt material for his published AGENTS.md guidance. It also exercises manners-vs-physics (prompt-file guidance layered over the kernel's deterministic DCO/Assisted-by enforcement), but with no outcome in evidence this is a trackable episode rather than a settled receipt — worth watching to resolution before promoting to high.
ip:concept.claude-md-patternip:concept.fat-agents-md-anti-patternip:source.your-ai-can-code-you-just-don-t-know-how-to-drive-it-ebookip:concept.manners-vs-physicsdev:concept.markdown-native-workflowradar:claude-code-agents-md-supportradar:karpathy-claude-md-rulesradar:memctl-versioned-agent-memoryradar:linux-drivers-llm-code-policyradar:linux-ai-patch-review-overloadradar:concept.open-source-governanceradar:concept.coding-agent-harnesses
queries asked of Scott's wikis
- AGENTS.md CLAUDE.md repo-level agent instruction files
- coding agent harness repo conventions context files
- AI-assisted contributions OSS governance attribution policy
- machine-readable docs for agents vs human README
- Linux Foundation AI standards MCP AAIF governance
- agent guardrails do-not-touch boundaries in codebases
Measured heat
now 0 pts/hpeak 0 pts/hcomments 0/hpeers p14momentum: steady2 platformsage 434h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion
How the heat travelled
pace: p32 vs 1032 stories at the 336h mark (now 434h old) — ahead of addom-local-coding-harness (1.5x), behind agentsec-static-config-auditing (0.8x)
Evidence (2) — ⭐ canonical anchor
Interpretation history
2026-09-26T09:53:32Z
origin walked (opencode/cheap-glm, conf 0.95): anchor hn.story.49854439 -> echo.other.b5b4e7d5b9 by Sasha Levin (
[email protected])
2026-09-26T09:48:57Z
grounded: converges/medium — Converges with Scott's canonized agent-instruction-file doctrine — the CLAUDE.md pattern, the fat-AGENTS.md anti-pattern, and the driving ebook's keep-it-slim p
2026-09-26T09:40:39Z
case created — Concrete, resolvable workflow decision at the top OSS project, reported via Phoronix and absent from the open queue; low traction but a genuine bounded episode.
Decision trace
- 09-26 19:53promote_anchororigin walk conf 0.95
- 09-26 19:48groundConverges with Scott's canonized agent-instruction-file doctrine — the CLAUDE.md pattern, the fat-AGENTS.md anti-pattern, and the driving ebook's keep-it-slim prescription — at the highest-a
- 09-26 19:40createConcrete, resolvable workflow decision at the top OSS project, reported via Phoronix and absent from the open queue; low traction but a genuine bounded episode.