[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-96113":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":9,"languages":9,"totalLinesOfCode":9,"stars":10,"forks":11,"watchers":12,"openIssues":13,"contributorsCount":13,"subscribersCount":13,"size":13,"stars1d":13,"stars7d":13,"stars30d":14,"stars90d":13,"forks30d":13,"starsTrendScore":13,"compositeScore":15,"rankGlobal":9,"rankLanguage":9,"license":16,"archived":17,"fork":17,"defaultBranch":18,"hasWiki":19,"hasPages":17,"topics":20,"createdAt":9,"pushedAt":9,"updatedAt":21,"readmeContent":22,"aiSummary":23,"trendingCount":13,"starSnapshotCount":13,"syncStatus":24,"lastSyncTime":25,"discoverSource":26},96113,"claude-style-patch","andrewroxby\u002Fclaude-style-patch","andrewroxby","A drop-in CLAUDE.md style section that changes how Claude writes prose.",null,126,7,1,0,8,43.51,"Other",false,"main",true,[],"2026-09-20 04:01:32","# Claude Style Patch\n\n[![Mentioned in Awesome Claude Code](https:\u002F\u002Fawesome.re\u002Fmentioned-badge.svg)](https:\u002F\u002Fgithub.com\u002Fhesreallyhim\u002Fawesome-claude-code)\n\nA drop-in style section for `CLAUDE.md` that helps Claude write clearer prose, targeting the worst of recent 'Claudish' in the generations since 4.6. I've been driving it for a long time now with success - this represents the final product battle-hardened against many, many Opus \u002F Fable turns.\n\nDense, gnarled prose was driving me insane when using Claude, so this was built to fix it. I like Anthropic as a company and want them to do well, but I think Claude's communication with humans has been drifting in the wrong direction. The problem showed up in generated documents as much as in chat replies, so the spec covers documents and deliverables and code comments alongside ordinary conversation.\n\nAfter enduring it for a while, I set about diagnosing the exact tics that were bothering me. Then I ran the loop repeatedly — add a specific diagnosis to CLAUDE.md, assess whether prose quality actually changed, tinker again. The spec solidified over many turns of that, across Claude Code and claude.ai and multiple generations of both Opus and Fable.\n\nIt targets several clusters of habits in the default house voice. Claude tends to\nannounce a point before making it, hinge sentences on a colon where the left\nside labels what the right side does, open paragraphs with verbless fragments,\nand stack compressed phrases against each other until a sentence has to be\ndecoded rather than read. Those habits get worse as a conversation gets longer\nand more abstract, which is when clear prose matters most.\n\nThe patch uses a variety of tactics - explicit examples & text exemplars as well as calling out habits more granularly. I've been quite satisfied with the results after daily driving this for quite some time, starting with Opus 4.8. \n\n## Install\n\nAppend it to your global config:\n\n```bash\ncurl -sL https:\u002F\u002Fraw.githubusercontent.com\u002Fandrewroxby\u002Fclaude-style-patch\u002Fmain\u002FSTYLE.md >> ~\u002F.claude\u002FCLAUDE.md\n```\n\nOr scope it to one project by appending to that project's `CLAUDE.md`. Outside\nClaude Code it works as-is pasted into a Claude Project's custom instructions,\nor into any system prompt.\n\n## Caveats\n\nCompliance is good but not total, and it degrades especially on long threads. Fable seems to be better at flawlessly following the spec than Opus, again, especially on longer threads. The opening paragraph tells Claude to re-check the rules when the material turns dense,\nwhich helps and does not fully solve it. Telling Claude directly to 'hew closely to the response style instructions' does, however, seem to work particularly well; eg before creating long prompts to other agents or generating new prose. \n\nThe file is roughly 1,600 words. I've found the context deeply worth it in sessions across both Claude Code and Claude.ai. \n\n## License\n\nCC0. Copy it, fork it, strip it for parts. No attribution needed.\n","这是一个用于优化Claude大模型文本生成风格的轻量级配置补丁，通过向CLAUDE.md注入结构化提示指令，显著改善其 prose 生成质量。核心功能包括抑制冗余句式（如冒号分隔、动词缺失片段、过度压缩短语）、强化逻辑连贯性与可读性，并支持全局或项目级灵活部署。技术特点在于采用诊断驱动的迭代设计，融合显式范例、行为标注与上下文自检机制，在Claude Opus与Fable等版本上实测有效。适用于需高质量技术文档、代码注释、产品文案或人机协作输出的专业场景，尤其适合对语言清晰度要求较高的开发者、技术写作者与AI提示工程师。",2,"2026-09-10 02:30:06","CREATED_QUERY"]