Ready to publish: "A 65-line rules file for AI coding agents has passed Anthropic's official repository on GitHub. Its real argument: the bottleneck is no longer model smarts."
Rachel → approved "A 65-line rules file for AI coding agents has passed Anthropic's official repository on GitHub. Its real argument: the bottleneck is no longer model smarts.": "RSS item with real Type0 fit. The behavioral-discipline inversion is the right angle — not Karpathy's brand, not the star count, but what the star inversion reveals about where AI coding's actual bottleneck now lives. The body is honest about the fourth unseen principle and the self-reported anecdote, which is the right editorial posture for this source base."
Iris → use: "The story is anchored to a concrete, verifiable artifact (a ~65-line rules file at ~17.6 万 GitHub stars, surpassing Anthropic's official anthropics/skills repo in the same month) and a sharp, defensible reframing: AI coding's binding constraint has moved from raw model capability to behavioral discipline under unsupervised execution. The piece names four specific, observable failure modes (silent assumption-filling, over-engineering, scope creep, undefined success criteria) and maps each to a rule a developer can write this afternoon — that gives the reader agency, which is the constructive test. Recommend USE for angle/title/lede work. Constructive direction for the reporter: frame the file as a portable pattern any developer or team can adopt, not as a celebrity endorsement of one repo; the four principles are stack-agnostic and should be presented as a starting template the reader customizes. Preserve legitimate caveats already in the source: the '65% → 94%' lift is developer self-report with no published benchmark behind it, the 65-vs-200-line comparison is anecdotal, and nothing here demonstrates the pattern generalizes across models or stacks. Preserve any source criticism about scope (the rules file does not address security review, deployment, model selection, or eval rigor). Do not extrapolate to business impact on Anthropic, on Dario Amodei's hiring decisions, or on Karpathy's career beyond what the source explicitly states. The Dario/'招回去' (re-hire) hook in the导语 is weakly sourced in this excerpt and should not lead."
Sky → pursuing: "Reporter should hydrate the GitHub repo (andrej-karpathy/skills), retrieve Karpathy's original post or tweet thread, and pull the actual CLAUDE.md text to confirm the four principles. Then draft on the mechanism: AI programming's competitive bottleneck shifting from model capability to constraint/governance design."