[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7194D3QNC2CB3FZ9ZTEFVGA":3,"skill-same:01M7194D3QNC2CB3FZ9ZTEFVGA":48},{"skillNo":4,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":8,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":14,"categories":15,"keywords":16,"publishedAt":20,"linkOnly":21,"markdownContent":22,"markdownBody":23,"frontMatter":24,"refName":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":31,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":33,"install":34,"resources":47},"01M7194D3QNC2CB3FZ9ZTEFVGA","evolve-crossover",null,"openplaybooks-dev\u002Fconverge","openplaybooks-dev","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F270560357?v=4","01M7191G17AHW5SXN63HJ0BEZV","01M7191G29922BDNWN8CFVG4PD",3,"MIT","https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev\u002Fconverge\u002Fblob\u002Fmain\u002Fexamples\u002Fevolutionary-optimization\u002F.converge\u002Fplaybooks\u002Fskills\u002Fevolve-crossover\u002FSKILL.md",[],[17,18,19],"evolve","evolve crossover","training","2026-11-10 10:12:30",false,"---\nid: evolve-crossover\ntitle: Crossover Training Configurations\n---\n\n# Crossover Training Configurations\n\nProduce a new generation of LLM training configurations by combining selected parents.\n\n## Process\n\n1. Read `selection.json` to get the top-K parent configurations\n2. Generate new configurations using multiple strategies:\n   - **Uniform crossover**: take architecture block from parent A, hyperparameters from parent B, data strategy from the higher-fitness parent\n   - **Interpolation**: blend numerical parameters (average learning rates, interpolate mix ratios, split-the-difference on layer count)\n   - **Novel injection**: 1-2 entirely new configurations exploring underrepresented regions (e.g. if all parents use GQA, try MHA; if all use cosine schedule, try WSD)\n3. Apply mutations to crossover offspring:\n   - Perturb learning rate ±30%\n   - Adjust layer count ±2 (recompute hidden dim to maintain param budget)\n   - Shift data mix ratios by ±0.05-0.1\n   - Swap optimizer or schedule type occasionally\n4. Record provenance: which parents, what crossover strategy, what mutations\n5. Write each new configuration as a separate JSON file\n\n## Outputs\n\n- `candidates\u002Fgen-N\u002Fcandidate-001.json` through `candidate-NNN.json`:\n  ```json\n  {\n    \"id\": \"candidate-001\",\n    \"generation\": 1,\n    \"approach\": \"GQA + code-heavy mix from parent-003, cosine schedule from parent-001\",\n    \"specification\": {\n      \"architecture\": { \"attentionType\": \"GQA\", \"layers\": 34, \"...\" : \"...\" },\n      \"hyperparameters\": { \"learningRate\": 2.5e-4, \"optimizer\": \"AdamW\", \"...\" : \"...\" },\n      \"dataStrategy\": { \"mix\": { \"code\": 0.25, \"webText\": 0.45, \"...\" : \"...\" }, \"...\" : \"...\" }\n    },\n    \"parents\": [\"candidate-003\", \"candidate-001\"],\n    \"crossoverStrategy\": \"uniform\",\n    \"mutations\": [\"took GQA from parent-003\", \"averaged learning rates\", \"added 2 layers\"]\n  }\n  ```\n\n## Quality Criteria\n\n- At least 1-2 configurations use \"novel\" strategy (not derived from parents)\n- Parent-derived configurations clearly trace to their parents\n- Mutations are meaningful (not just noise — e.g. \"perturbed LR from 3e-4 to 2.1e-4\")\n- Architecture choices remain internally consistent after crossover (param count, head divisibility)\n- Data mix ratios still sum to 1.0\n- Population size matches the configured populationSize\n","# Crossover Training Configurations\n\nProduce a new generation of LLM training configurations by combining selected parents.\n\n## Process\n\n1. Read `selection.json` to get the top-K parent configurations\n2. Generate new configurations using multiple strategies:\n   - **Uniform crossover**: take architecture block from parent A, hyperparameters from parent B, data strategy from the higher-fitness parent\n   - **Interpolation**: blend numerical parameters (average learning rates, interpolate mix ratios, split-the-difference on layer count)\n   - **Novel injection**: 1-2 entirely new configurations exploring underrepresented regions (e.g. if all parents use GQA, try MHA; if all use cosine schedule, try WSD)\n3. Apply mutations to crossover offspring:\n   - Perturb learning rate ±30%\n   - Adjust layer count ±2 (recompute hidden dim to maintain param budget)\n   - Shift data mix ratios by ±0.05-0.1\n   - Swap optimizer or schedule type occasionally\n4. Record provenance: which parents, what crossover strategy, what mutations\n5. Write each new configuration as a separate JSON file\n\n## Outputs\n\n- `candidates\u002Fgen-N\u002Fcandidate-001.json` through `candidate-NNN.json`:\n  ```json\n  {\n    \"id\": \"candidate-001\",\n    \"generation\": 1,\n    \"approach\": \"GQA + code-heavy mix from parent-003, cosine schedule from parent-001\",\n    \"specification\": {\n      \"architecture\": { \"attentionType\": \"GQA\", \"layers\": 34, \"...\" : \"...\" },\n      \"hyperparameters\": { \"learningRate\": 2.5e-4, \"optimizer\": \"AdamW\", \"...\" : \"...\" },\n      \"dataStrategy\": { \"mix\": { \"code\": 0.25, \"webText\": 0.45, \"...\" : \"...\" }, \"...\" : \"...\" }\n    },\n    \"parents\": [\"candidate-003\", \"candidate-001\"],\n    \"crossoverStrategy\": \"uniform\",\n    \"mutations\": [\"took GQA from parent-003\", \"averaged learning rates\", \"added 2 layers\"]\n  }\n  ```\n\n## Quality Criteria\n\n- At least 1-2 configurations use \"novel\" strategy (not derived from parents)\n- Parent-derived configurations clearly trace to their parents\n- Mutations are meaningful (not just noise — e.g. \"perturbed LR from 3e-4 to 2.1e-4\")\n- Architecture choices remain internally consistent after crossover (param count, head divisibility)\n- Data mix ratios still sum to 1.0\n- Population size matches the configured populationSize\n",{"id":5,"title":25},"Crossover Training Configurations","main","7d618d8bc059f1a0974706d1d15ec25f73cf1e1d","examples\u002Fevolutionary-optimization\u002F.converge\u002Fplaybooks\u002Fskills\u002Fevolve-crossover",1,"https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev\u002Fconverge","https:\u002F\u002Fplaybooks.to","https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev","2026-06-13 10:11:21",{"source":7,"skillName":5,"commands":35,"prompt":45,"zipAvailable":46},[36,39,42],{"tool":37,"command":38},"npx","npx skills add openplaybooks-dev\u002Fconverge --skill evolve-crossover",{"tool":40,"command":41},"bunx","bunx skills add openplaybooks-dev\u002Fconverge --skill evolve-crossover",{"tool":43,"command":44},"pnpm","pnpm dlx skills add openplaybooks-dev\u002Fconverge --skill evolve-crossover","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev\u002Fconverge\u002Fblob\u002Fmain\u002Fexamples\u002Fevolutionary-optimization\u002F.converge\u002Fplaybooks\u002Fskills\u002Fevolve-crossover\u002FSKILL.md",true,[],[49,77,86,108,133,161],{"skillNo":50,"name":51,"nameZh":6,"description":52,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":8,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":53,"categories":54,"keywords":60,"publishedAt":20},"01M71989MQF3GPZ7P5TX6S93PD","hello-checker","Use this skill when asked to \"verify\", \"check\", or \"validate\" a file's existence and content. It provides a reliable workflow for confirming files meet their spec.","https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev\u002Fconverge\u002Fblob\u002Fmain\u002Ftests\u002Ftest-codex-real\u002F.converge\u002Fskills\u002Fhello-checker\u002FSKILL.md",[55],{"categoryNo":56,"categoryName":57,"taxonomyType":58,"taxonomyTypeDesc":59},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76],"checker","hello checker","hello","it","use","content","validate","spec","verify","check","file","workflow","when","Meet","files","for",{"skillNo":78,"name":79,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":8,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":83,"publishedAt":20},"01M7198823W79B3S15TPV97TCB","ask-skill","https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev\u002Fconverge\u002Fblob\u002Fmain\u002Ftests\u002Ftest-ask-autonomous\u002F.converge\u002Fplaybooks\u002Fdefault\u002Fskills\u002Fask-skill\u002FSKILL.md",[82],{"categoryNo":56,"categoryName":57,"taxonomyType":58,"taxonomyTypeDesc":59},[84,85],"ask skill","ask",{"skillNo":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":8,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":96,"publishedAt":20},"01M71986EJ6VCM8YKGEM2YEDQV","html-review-artifact","Framework-wide guidance for crafting polished human-review artifacts as markdown or HTML","https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev\u002Fconverge\u002Fblob\u002Fmain\u002Fskills\u002Fhtml-review-artifact\u002FSKILL.md",[92,95],{"categoryNo":93,"categoryName":94,"taxonomyType":58,"taxonomyTypeDesc":59},"01JBSEEDCAT000000000000009","前端开发",{"categoryNo":56,"categoryName":57,"taxonomyType":58,"taxonomyTypeDesc":59},[97,98,99,100,101,76,102,103,104,105,106,107],"html","artifact","review","html review artifact","as","framework","wide","guidance","markdown","human","artifacts",{"skillNo":109,"name":110,"nameZh":6,"description":111,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":8,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":112,"categories":113,"keywords":115,"publishedAt":20},"01M719833M2XWXG1VPBW30GW7J","converge-planning","Design a Converge playbook end-to-end. Use when the user says \"plan a project\", \"design a playbook\", \"decompose into tasks\", \"scaffold a Converge workflow\", \"onboard to Converge\", or asks how to write TASK.md \u002F playbook.yml. Two entry points: plan a new playbook or update an existing one. Hand off to `\u002Fconverge-control` after producing the playbook structure.","https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev\u002Fconverge\u002Fblob\u002Fmain\u002Fskills\u002Fconverge-planning\u002FSKILL.md",[114],{"categoryNo":56,"categoryName":57,"taxonomyType":58,"taxonomyTypeDesc":59},[116,117,118,119,72,120,65,121,122,123,124,125,126,127,128,129,130,131,132,73],"converge","converge planning","planning","structure","tasks","decompose","hand","write","plan","how","update","playbook","new","design","how to","end","project",{"skillNo":134,"name":135,"nameZh":6,"description":136,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":8,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":137,"categories":138,"keywords":145,"publishedAt":20},"01M7197XZT1T9GDYAJGVGX1J60","converge-control","Operate a Converge playbook end-to-end — start runs, monitor progress, narrow failures with `--select`, inspect runtime state, validate definitions, surgically reset subtrees, and stop stuck or orphaned runs. Use this skill whenever the user says \"run my playbook\", \"execute this\", \"monitor the run\", \"converge failed\", \"the run is stuck\", \"retry failures\", \"babysit this playbook\", \"what's the status?\", \"show me the gantt\", or asks how to use `converge run` \u002F `--select` \u002F `inspect` \u002F `doctor` \u002F `clean` \u002F `stop`. Also use it for `HTTP 401` \u002F `Invalid API key` failures on the first task — those are environment-not-playbook problems — and for any post-run cleanup work. Hand off to `converge-planning` if the user wants to design or restructure the playbook itself.","https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev\u002Fconverge\u002Fblob\u002Fmain\u002Fskills\u002Fconverge-control\u002FSKILL.md",[139,140],{"categoryNo":56,"categoryName":57,"taxonomyType":58,"taxonomyTypeDesc":59},{"categoryNo":141,"categoryName":142,"taxonomyType":143,"taxonomyTypeDesc":144},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",[146,147,116,148,149,150,151,152,153,154,122,155,156,129,130,157,131,158,159,160],"converge control","control","operate","invalid","show","api","me","work","reset","show me","cleanup","clean","inspect","select","retry",{"skillNo":162,"name":163,"nameZh":6,"description":164,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":8,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":165,"categories":166,"keywords":168,"publishedAt":20},"01M7197SCVN0QTD3CM4M83WYJH","repair-taskts","Fix task.ts definition issues — Seed syntax, missing .build(), empty checks, import errors, TypeScript errors","https:\u002F\u002Fgithub.com\u002Fopenplaybooks-dev\u002Fconverge\u002Fblob\u002Fmain\u002Fpackages\u002Fcore\u002Fsrc\u002Fnavigator\u002Frepair\u002Fskill-templates\u002Frepair-taskts\u002FSKILL.md",[167],{"categoryNo":141,"categoryName":142,"taxonomyType":143,"taxonomyTypeDesc":144},[169,170,171,172,123,173,174,175,176,177,178,179,180,181,182,131,183,116,184,185],"repair taskts","repair","Bash","edit","read","line","at","spawn","code","analysis","number","errors","issues","fix","chain","definition","never"]