[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M71AXC6YC6MMVF8G88XN89Y1":3,"skill-same:01M71AXC6YC6MMVF8G88XN89Y1":82},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":16,"sourceHtmlUrl":17,"categories":18,"keywords":24,"publishedAt":42,"linkOnly":43,"markdownContent":44,"markdownBody":45,"frontMatter":46,"refName":51,"commitSha":52,"skillDirectory":53,"forksCount":54,"repositoryHtmlUrl":55,"repositoryHomepageUrl":6,"ownerHtmlUrl":56,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":57,"install":58,"resources":71},"01M71AXC6YC6MMVF8G88XN89Y1","eval-harness-current",null,"Design and analyze controlled coding-agent experiments when comparing skills, prompts, models, or agent configurations using independently verified task outcomes.","strmt7\u002Fai_skills_collection_benchmarked","strmt7","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F28989754?v=4","01M71929P914SWKQ7QJV24NBBY","Stratos","01M71929Q9VGKQ7PWYR95VSQQQ",1,"MIT","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fimproved\u002Fskills\u002Feval-harness-current\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fimproved\u002Fskills\u002Feval-harness-current\u002FSKILL.md",[19],{"categoryNo":20,"categoryName":21,"taxonomyType":22,"taxonomyTypeDesc":23},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41],"eval harness current","harness","eval harness","eval","outcomes","when","verified","claude-code","task","models","prompts","Coding-Agent","design","coding","analyze","code","experiments","2026-11-10 10:12:30",false,"---\nname: eval-harness-current\ndescription: Design and analyze controlled coding-agent experiments when comparing skills, prompts, models, or agent configurations using independently verified task outcomes.\nlicense: MIT\nmetadata:\n  source: affaan-m\u002Feverything-claude-code\n  source-commit: 846ffb75da9a5f4e677d927af1ad4a1951652267\n  revision-status: experimental-unbenchmarked\n---\n\n# Objective agent evaluation\n\nDefine the decision the experiment should support: a skill's benefit over the\nsame agent without it, a regression between revisions, or a model comparison.\nChanging the model and skill simultaneously cannot isolate either effect.\n\n## Define the experiment\n\n- Choose tasks from real pinned issues, workflows, or independently authored\n  user requirements. Record source provenance, environment, and task coverage.\n- Define graders before reading the exact candidate instructions. Score actual\n  behavior with executable acceptance checks, including relevant regressions.\n  A function name, code pattern, plausible explanation, or agent self-report is\n  not proof of correctness. Use blinded review only for criteria that need it.\n- Keep development cases separate from final holdouts. Any case used to tune\n  instructions becomes a development case, even if originally called a holdout.\n- Run the original input through the grader to prove the task's failing behavior;\n  check a known valid outcome and plausible wrong outcomes to test the grader.\n  Keep oracle artifacts outside the agent's accessible task environment.\n\n## Execute matched trials\n\nUse fresh workspaces and sessions with the same task input, tool permissions,\nmodel version, reasoning effort, limits, and dependency versions. Declare which\ninstructions each condition loads; record effective configuration and instruction\nhashes. Development history, memory, prior patches, and unrelated skills can\ncontaminate a baseline. If isolation cannot be verified, label that limitation.\n\nRotate condition order across repeated tasks. Preserve each attempt's transcript,\npatch, grader output, exit status, actual usage, and elapsed time. Record failures,\ntimeouts, environment errors, and unavailable measurements rather than deleting\nthem or substituting zero. Validate inputs and grader files against their frozen\nhashes after execution, and detect changes outside the permitted patch scope.\n\n## Analyze and iterate\n\nReport the number of planned, attempted, validly graded, successful, and failed\ntrials for each condition. Distinguish task failure from harness failure. Show\npaired task results and uncertainty before claiming superiority; disclose\ncategory coverage, exclusions, repeat count, and any contaminated tasks.\n\nFor each task, let `n` be the number of recorded attempts and `c` the number\nthat passed a complete grader. With no unscored attempts and `1 \u003C= k \u003C= n`, the\nfraction of uniformly chosen k-subsets with any success is\n`1 - C(n-c, k) \u002F C(n, k)`; the fraction where all succeed is\n`C(c, k) \u002F C(n, k)`, taking combinations with fewer than k items as zero.\nHere `n` counts agent attempts, not individual tests within one attempt. Compute\nper task first, then average tasks equally unless different task weights were\ndeclared in advance. Do not pool attempts across tasks to calculate these scores.\n\nFor example, if one task has two passes in two attempts and another has no\npasses in eight, equal-task pass@1 is `(1 + 0) \u002F 2 = 1\u002F2`, not the pooled\nattempt rate `2\u002F10`. When `n \u003C k`, the task's k-result is unavailable. Keep\nunscored attempts visible; report incomplete outcomes or explicit best\u002Fworst\nassignment bounds, not an unconditional pass rate based only on scored runs.\n\nThe subset fractions describe the recorded attempts. Interpreting them as\nfuture-run probabilities requires suitable independent, identically distributed\nsampling and stable conditions; adaptive retries need their own analysis. An\nany-of-k score credits a task if a passing output exists; using that success\nrequires a feasible selection method. Account for the cost of k attempts and\nselection separately. The score is not an all-runs\nreliability measure. All-of-k\ndescribes consistency under the stated setup, not a worst-case guarantee. Do not\nrelabel retry success or a selected best attempt as pass@1. See the original\n[pass@k estimator](https:\u002F\u002Farxiv.org\u002Fhtml\u002F2107.03374v2#S2.SS1) and the\n[agent evaluation discussion](https:\u002F\u002Fsage.cs.princeton.edu\u002Ftrends\u002F#trend-1d).\n\nCompare quality first, then tokens, latency, tool use, and cost for comparable\noutcomes. Do not infer speed gains from host-load noise or efficiency from shorter\nanswers alone. Keep package readiness and harness smoke checks separate from\nagent task performance.\n\nRevise against demonstrated development failures, rerun affected matched cases\nand regressions, then evaluate untouched holdouts. Preserve negative results and\nversion each iteration. If measured gains are absent, report that outcome.\n","# Objective agent evaluation\n\nDefine the decision the experiment should support: a skill's benefit over the\nsame agent without it, a regression between revisions, or a model comparison.\nChanging the model and skill simultaneously cannot isolate either effect.\n\n## Define the experiment\n\n- Choose tasks from real pinned issues, workflows, or independently authored\n  user requirements. Record source provenance, environment, and task coverage.\n- Define graders before reading the exact candidate instructions. Score actual\n  behavior with executable acceptance checks, including relevant regressions.\n  A function name, code pattern, plausible explanation, or agent self-report is\n  not proof of correctness. Use blinded review only for criteria that need it.\n- Keep development cases separate from final holdouts. Any case used to tune\n  instructions becomes a development case, even if originally called a holdout.\n- Run the original input through the grader to prove the task's failing behavior;\n  check a known valid outcome and plausible wrong outcomes to test the grader.\n  Keep oracle artifacts outside the agent's accessible task environment.\n\n## Execute matched trials\n\nUse fresh workspaces and sessions with the same task input, tool permissions,\nmodel version, reasoning effort, limits, and dependency versions. Declare which\ninstructions each condition loads; record effective configuration and instruction\nhashes. Development history, memory, prior patches, and unrelated skills can\ncontaminate a baseline. If isolation cannot be verified, label that limitation.\n\nRotate condition order across repeated tasks. Preserve each attempt's transcript,\npatch, grader output, exit status, actual usage, and elapsed time. Record failures,\ntimeouts, environment errors, and unavailable measurements rather than deleting\nthem or substituting zero. Validate inputs and grader files against their frozen\nhashes after execution, and detect changes outside the permitted patch scope.\n\n## Analyze and iterate\n\nReport the number of planned, attempted, validly graded, successful, and failed\ntrials for each condition. Distinguish task failure from harness failure. Show\npaired task results and uncertainty before claiming superiority; disclose\ncategory coverage, exclusions, repeat count, and any contaminated tasks.\n\nFor each task, let `n` be the number of recorded attempts and `c` the number\nthat passed a complete grader. With no unscored attempts and `1 \u003C= k \u003C= n`, the\nfraction of uniformly chosen k-subsets with any success is\n`1 - C(n-c, k) \u002F C(n, k)`; the fraction where all succeed is\n`C(c, k) \u002F C(n, k)`, taking combinations with fewer than k items as zero.\nHere `n` counts agent attempts, not individual tests within one attempt. Compute\nper task first, then average tasks equally unless different task weights were\ndeclared in advance. Do not pool attempts across tasks to calculate these scores.\n\nFor example, if one task has two passes in two attempts and another has no\npasses in eight, equal-task pass@1 is `(1 + 0) \u002F 2 = 1\u002F2`, not the pooled\nattempt rate `2\u002F10`. When `n \u003C k`, the task's k-result is unavailable. Keep\nunscored attempts visible; report incomplete outcomes or explicit best\u002Fworst\nassignment bounds, not an unconditional pass rate based only on scored runs.\n\nThe subset fractions describe the recorded attempts. Interpreting them as\nfuture-run probabilities requires suitable independent, identically distributed\nsampling and stable conditions; adaptive retries need their own analysis. An\nany-of-k score credits a task if a passing output exists; using that success\nrequires a feasible selection method. Account for the cost of k attempts and\nselection separately. The score is not an all-runs\nreliability measure. All-of-k\ndescribes consistency under the stated setup, not a worst-case guarantee. Do not\nrelabel retry success or a selected best attempt as pass@1. See the original\n[pass@k estimator](https:\u002F\u002Farxiv.org\u002Fhtml\u002F2107.03374v2#S2.SS1) and the\n[agent evaluation discussion](https:\u002F\u002Fsage.cs.princeton.edu\u002Ftrends\u002F#trend-1d).\n\nCompare quality first, then tokens, latency, tool use, and cost for comparable\noutcomes. Do not infer speed gains from host-load noise or efficiency from shorter\nanswers alone. Keep package readiness and harness smoke checks separate from\nagent task performance.\n\nRevise against demonstrated development failures, rerun affected matched cases\nand regressions, then evaluate untouched holdouts. Preserve negative results and\nversion each iteration. If measured gains are absent, report that outcome.\n",{"name":5,"license":15,"metadata":47,"description":7},{"source":48,"source-commit":49,"revision-status":50},"affaan-m\u002Feverything-claude-code","846ffb75da9a5f4e677d927af1ad4a1951652267","experimental-unbenchmarked","main","5c95bcac4646c5aed06bd9aed5235cfc7985aafc","included\u002Fimproved\u002Fskills\u002Feval-harness-current",0,"https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked","https:\u002F\u002Fgithub.com\u002Fstrmt7","2026-11-11 04:18:12",{"source":8,"skillName":5,"commands":59,"prompt":69,"zipAvailable":70},[60,63,66],{"tool":61,"command":62},"npx","npx skills add strmt7\u002Fai_skills_collection_benchmarked --skill eval-harness-current",{"tool":64,"command":65},"bunx","bunx skills add strmt7\u002Fai_skills_collection_benchmarked --skill eval-harness-current",{"tool":67,"command":68},"pnpm","pnpm dlx skills add strmt7\u002Fai_skills_collection_benchmarked --skill eval-harness-current","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fimproved\u002Fskills\u002Feval-harness-current\u002FSKILL.md",true,[72,76],{"path":73,"resourceType":73,"resourceTypeDesc":74,"sizeBytes":75,"sourceHtmlUrl":16},"LICENSE","许可证",1071,{"path":77,"resourceType":78,"resourceTypeDesc":79,"sizeBytes":80,"sourceHtmlUrl":81},"provenance\u002Foriginal-SKILL.md","REFERENCE","参考文档",5515,"https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fimproved\u002Fskills\u002Feval-harness-current\u002Fprovenance\u002Foriginal-SKILL.md",[83,115,142,167,193,219],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":87,"licenseFileUrl":88,"sourceHtmlUrl":89,"categories":90,"keywords":97,"publishedAt":42},"01M71D4H3T6M57DFKN8HENAWRV","webapp-testing","Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.","Apache-2.0","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Ftesting-qa-benchmarking\u002Fofficial-reference\u002Fwebapp-testing\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Ftesting-qa-benchmarking\u002Fofficial-reference\u002Fwebapp-testing\u002FSKILL.md",[91,94],{"categoryNo":92,"categoryName":93,"taxonomyType":22,"taxonomyTypeDesc":23},"01JBSEEDCAT000000000000008","测试与质量",{"categoryNo":95,"categoryName":96,"taxonomyType":22,"taxonomyTypeDesc":23},"01JBSEEDCAT000000000000009","前端开发",[98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114],"webapp","testing","webapp testing","for","applications","local","frontend","debugging","toolkit","logs","browser","with","playwright","web","behavior","ui","screenshots",{"skillNo":116,"name":117,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":87,"licenseFileUrl":119,"sourceHtmlUrl":120,"categories":121,"keywords":123,"publishedAt":42},"01M71D4CQR2A2TB7GASCR5ZBG3","skill-creator","Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Ftesting-qa-benchmarking\u002Fofficial-reference\u002Fskill-creator-anthropics\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Ftesting-qa-benchmarking\u002Fofficial-reference\u002Fskill-creator-anthropics\u002FSKILL.md",[122],{"categoryNo":20,"categoryName":21,"taxonomyType":22,"taxonomyTypeDesc":23},[124,125,126,127,128,129,30,130,131,132,133,134,135,136,137,138,139,140,101,141],"skill creator","creator","performance","improve","run evals","new","analysis","create","optimize","variance analysis","run","edit","benchmark","a skill","existing skill","evals","modify","accuracy",{"skillNo":143,"name":144,"nameZh":6,"description":145,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":146,"categories":147,"keywords":150,"publishedAt":42},"01M71D488XHEZVXXSXFYP8ENMH","playwright-skill","Complete browser automation with Playwright. Auto-detects dev servers, writes reusable test scripts, and supports screenshots, responsive checks, UX validation, login flows, link checks, and arbitrary browser automation. Use when the user wants to test a website, automate browser interactions, validate web functionality, or perform browser-based testing.","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Ftesting-qa-benchmarking\u002Flatest-release-browser-automation\u002Fplaywright-skill\u002FSKILL.md",[148,149],{"categoryNo":92,"categoryName":93,"taxonomyType":22,"taxonomyTypeDesc":23},{"categoryNo":20,"categoryName":21,"taxonomyType":22,"taxonomyTypeDesc":23},[110,151,152,153,99,154,155,30,156,108,157,158,159,160,161,162,163,164,165,166],"playwright skill","browser automation","flows","auto","access","Bash","website","ux","login","scripts","read","write","node","install","network","responsive",{"skillNo":168,"name":169,"nameZh":6,"description":170,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":87,"licenseFileUrl":171,"sourceHtmlUrl":172,"categories":173,"keywords":175,"publishedAt":42},"01M71CY25GM939M6RES91QSG3T","web-artifacts-builder","Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn\u002Fui). Use for complex artifacts requiring state management, routing, or shadcn\u002Fui components - not for simple single-file HTML\u002FJSX artifacts.","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Fsearch-retrieval-web-automation\u002Fofficial-reference\u002Fweb-artifacts-builder\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Fsearch-retrieval-web-automation\u002Fofficial-reference\u002Fweb-artifacts-builder\u002FSKILL.md",[174],{"categoryNo":95,"categoryName":96,"taxonomyType":22,"taxonomyTypeDesc":23},[176,177,178,111,179,180,181,101,182,183,104,184,185,186,187,188,189,190,191,192],"builder","artifacts builder","artifacts","web artifacts builder","single-file","css","Tailwind","ui components","single","use","file","components","react","multi","state","routing","state management",{"skillNo":194,"name":195,"nameZh":6,"description":196,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":87,"licenseFileUrl":197,"sourceHtmlUrl":198,"categories":199,"keywords":209,"publishedAt":42},"01M71CXZB60QJJ1H4S3NN94WMZ","frontend-design","Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, artifacts, posters, or applications (examples include websites, landing pages, dashboards, React components, HTML\u002FCSS layouts, or when styling\u002Fbeautifying any web UI). Generates creative, polished code and UI design that avoids generic AI aesthetics.","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Fsearch-retrieval-web-automation\u002Fofficial-reference\u002Ffrontend-design\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Fsearch-retrieval-web-automation\u002Fofficial-reference\u002Ffrontend-design\u002FSKILL.md",[200,201,204],{"categoryNo":95,"categoryName":96,"taxonomyType":22,"taxonomyTypeDesc":23},{"categoryNo":202,"categoryName":203,"taxonomyType":22,"taxonomyTypeDesc":23},"01JBSEEDCAT000000000000015","视觉设计",{"categoryNo":205,"categoryName":206,"taxonomyType":207,"taxonomyTypeDesc":208},"01JBSEEDCAT000000000000032","设计师","PROFESSION","适用职业",[104,37,210,181,211,212,213,102,214,215,185,30,178,40,187,216,188,217,218,109],"frontend design","ui design","generic","aesthetics","landing","pages","web ui","react components","examples",{"skillNo":220,"name":221,"nameZh":6,"description":222,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":223,"categories":224,"keywords":225,"publishedAt":42},"01M71CVZ54WFBTVA7R16PFY1ZE","zarr-python","Chunked N-D arrays for cloud storage. Compressed arrays, parallel I\u002FO, S3\u002FGCS integration, NumPy\u002FDask\u002FXarray compatible, for large-scale scientific computing pipelines.","https:\u002F\u002Fgithub.com\u002Fstrmt7\u002Fai_skills_collection_benchmarked\u002Fblob\u002Fmain\u002Fincluded\u002Fskills\u002Fby-category\u002Fscience-research-data-analysis\u002Flatest-release-community\u002Fzarr-python\u002FSKILL.md",[],[226,227,228,229,230,231,232,233,234,101,235,236,237,238,239,240,241,242],"zarr python","python","zarr","arrays","Cloud Storage","scientific","storage","parallel","dask","numpy","gcs","s3","integration","scientific computing","cloud","scale","pipelines"]