[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M71R3CSHS8MM4HNHBYZMR4JG":3,"skill-same:01M71R3CSHS8MM4HNHBYZMR4JG":75},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":22,"publishedAt":37,"linkOnly":38,"markdownContent":39,"markdownBody":40,"frontMatter":41,"refName":45,"commitSha":46,"skillDirectory":47,"forksCount":13,"repositoryHtmlUrl":48,"repositoryHomepageUrl":6,"ownerHtmlUrl":49,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":50,"install":51,"resources":64},"01M71R3CSHS8MM4HNHBYZMR4JG","karpathy-loop",null,"Iterative prompt optimization — write, measure, cut, repeat with progressive compression","LuisAlbertoMK\u002Fgentleman-agent-gh","LuisAlbertoMK","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F96153499?v=4","01M71QZX0CF9KKGQZ8ZN9BME63","01M71QZX11P5X6Z27HE1GQDF71",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fkarpathy-loop\u002FSKILL.md",[17],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[23,24,25,26,27,28,29,30,31,32,33,34,35,36],"karpathy loop","loop","karpathy","Prompt Optimization","Optimization","tokens","progressive","prompt","with","optimize","cut","write","compression","docs","2026-11-10 13:02:02",false,"---\nname: karpathy-loop\ndescription: \"Iterative prompt optimization — write, measure, cut, repeat with progressive compression\"\ntriggers: \"Karpathy loop, optimize prompt, measure tokens\"\nchangelog: docs\u002Fciclos\u002Fcycle28-20260815.md\ntoken_budget: 3050\n---\n\n## When to Use\nPrompt bloated\u002Fslow\u002Fexpensive · need consistent quality at lower cost · production\u002Fautomation prompts · skill authoring — every skill should pass karpathy-loop.\n\n## STYLE (5 Rules)\n1. **ID+TASK=ENOUGH** — identity+task. 2. **MINIMAL** — no 10+ item lists. 3. **FORMAT=INSTRUCT** — \"Respond ONLY in JSON:{...}\". 4. **CONSTRAINTS=FORMAT** — max X chars \u002F code only. 5. **IMPLICIT CoT** — no \"step by step\"; \"Reason ONLY if ambiguous\".\n\n## LOOP: Write→Measure→Cut→Repeat\n1. **WRITE**: role + task + 1-2 examples + output. Don't optimize yet. 2. **MEASURE**: chars\u002F4 ≈ tokens; score correctness\u002Fconciseness\u002Frobustness. 3. **CUT**: remove redundant→merge→simplify; no output change→cut. 4. **REPEAT**: improves+tokens down→continue; drops→revert; stagnant→new tactic.\n\nExample: Write 680\u002F170tok→6.3 | Cut T1 480\u002F120→7.3 | T2 340\u002F85→7.3 | T3 260\u002F65→5.3 ✗ → REVERT T2. Final 340ch\u002F85tok 7.3\u002F10.\n\n## Compression Levels\n| Level | Target | What to cut |\n|---|---|---|\n| T1 | 20-30% | filler, transitions, \"step by step\" |\n| T2 | 30-50% | merge redundant→bullets, remove context |\n| T3 | 50-70% | template structures, shortcuts, minimal identity |\n\n## Budget\nID+TASK: 20-50 | +example: +100-200 | +constraints: +50-100 | **OPTIMAL: 50-300**\n\n## Decision — Remove?\nOutput changes?→keep · else concision?→remove · else clarity?→keep · else→remove\n\n## STOP\nTokens \u003C50 + works · 3 iterations no improvement · Fits in 1 line. **NEVER**: sacrifice correctness for tokens, or leave edge cases uncovered.\n\n## Anti-Rationalization\n\n| Rationalization | Red Flag | Verification |\n|-----------------|----------|--------------|\n| \"3 iteraciones bastan, corto en T2\" | Stopping at T2 in THOROUGH\u002Fcomplex prompt | Measure T3 — revert only if score drops (e.g., 7.3→5.3), must measure first |\n| \"Tokens ya \u003C300, no medir\" | Skipping MEASURE after CUT | tokens≈chars\u002F4 + correctness\u002Fconciseness\u002Frobustness score each loop |\n| \"Quito constraints para ahorrar\" | Removing FORMAT=INSTRUCT to save tokens | Verify output still matches JSON\u002Fschema without constraint — keep if drift |\n\n## Red Flags\n- Doing work without checking output format → STOP, re-read skill\n- Second occurrence of same rationalization → force RED zone\n\n## Verification\n- Output matches skill ## Output contract + file:line citaton\n- cross-ref-check.ps1 → SKILL.md OK\n## Refs\nlean-context · skill-improver · metricas · code-review-agent · context-watchdog · execution-mode\n\n**Absorbed (2026-09-18)**: `prompt-engineering` archived — its prompt-optimization\u002Ftemplate workflows are covered by this skill.\n\n## Reference\nWorked examples, testing patterns, edge cases, anti-patterns → docs\u002Fskills\u002Fkarpathy-loop\u002Freference.md\n\n","## When to Use\nPrompt bloated\u002Fslow\u002Fexpensive · need consistent quality at lower cost · production\u002Fautomation prompts · skill authoring — every skill should pass karpathy-loop.\n\n## STYLE (5 Rules)\n1. **ID+TASK=ENOUGH** — identity+task. 2. **MINIMAL** — no 10+ item lists. 3. **FORMAT=INSTRUCT** — \"Respond ONLY in JSON:{...}\". 4. **CONSTRAINTS=FORMAT** — max X chars \u002F code only. 5. **IMPLICIT CoT** — no \"step by step\"; \"Reason ONLY if ambiguous\".\n\n## LOOP: Write→Measure→Cut→Repeat\n1. **WRITE**: role + task + 1-2 examples + output. Don't optimize yet. 2. **MEASURE**: chars\u002F4 ≈ tokens; score correctness\u002Fconciseness\u002Frobustness. 3. **CUT**: remove redundant→merge→simplify; no output change→cut. 4. **REPEAT**: improves+tokens down→continue; drops→revert; stagnant→new tactic.\n\nExample: Write 680\u002F170tok→6.3 | Cut T1 480\u002F120→7.3 | T2 340\u002F85→7.3 | T3 260\u002F65→5.3 ✗ → REVERT T2. Final 340ch\u002F85tok 7.3\u002F10.\n\n## Compression Levels\n| Level | Target | What to cut |\n|---|---|---|\n| T1 | 20-30% | filler, transitions, \"step by step\" |\n| T2 | 30-50% | merge redundant→bullets, remove context |\n| T3 | 50-70% | template structures, shortcuts, minimal identity |\n\n## Budget\nID+TASK: 20-50 | +example: +100-200 | +constraints: +50-100 | **OPTIMAL: 50-300**\n\n## Decision — Remove?\nOutput changes?→keep · else concision?→remove · else clarity?→keep · else→remove\n\n## STOP\nTokens \u003C50 + works · 3 iterations no improvement · Fits in 1 line. **NEVER**: sacrifice correctness for tokens, or leave edge cases uncovered.\n\n## Anti-Rationalization\n\n| Rationalization | Red Flag | Verification |\n|-----------------|----------|--------------|\n| \"3 iteraciones bastan, corto en T2\" | Stopping at T2 in THOROUGH\u002Fcomplex prompt | Measure T3 — revert only if score drops (e.g., 7.3→5.3), must measure first |\n| \"Tokens ya \u003C300, no medir\" | Skipping MEASURE after CUT | tokens≈chars\u002F4 + correctness\u002Fconciseness\u002Frobustness score each loop |\n| \"Quito constraints para ahorrar\" | Removing FORMAT=INSTRUCT to save tokens | Verify output still matches JSON\u002Fschema without constraint — keep if drift |\n\n## Red Flags\n- Doing work without checking output format → STOP, re-read skill\n- Second occurrence of same rationalization → force RED zone\n\n## Verification\n- Output matches skill ## Output contract + file:line citaton\n- cross-ref-check.ps1 → SKILL.md OK\n## Refs\nlean-context · skill-improver · metricas · code-review-agent · context-watchdog · execution-mode\n\n**Absorbed (2026-09-18)**: `prompt-engineering` archived — its prompt-optimization\u002Ftemplate workflows are covered by this skill.\n\n## Reference\nWorked examples, testing patterns, edge cases, anti-patterns → docs\u002Fskills\u002Fkarpathy-loop\u002Freference.md\n\n",{"name":5,"triggers":42,"changelog":43,"description":7,"token_budget":44},"Karpathy loop, optimize prompt, measure tokens","docs\u002Fciclos\u002Fcycle28-20260815.md",3050,"main","4af6946b30783e2c8f9182082324d530fc28f838",".agents\u002Fskills\u002Fkarpathy-loop","https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh","https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK","2026-11-10 18:34:27",{"source":8,"skillName":5,"commands":52,"prompt":62,"zipAvailable":63},[53,56,59],{"tool":54,"command":55},"npx","npx skills add LuisAlbertoMK\u002Fgentleman-agent-gh --skill karpathy-loop",{"tool":57,"command":58},"bunx","bunx skills add LuisAlbertoMK\u002Fgentleman-agent-gh --skill karpathy-loop",{"tool":60,"command":61},"pnpm","pnpm dlx skills add LuisAlbertoMK\u002Fgentleman-agent-gh --skill karpathy-loop","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fkarpathy-loop\u002FSKILL.md",true,[65,71],{"path":66,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":69,"sourceHtmlUrl":70},"cut-tactics.md","REFERENCE","参考文档",1101,"https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fkarpathy-loop\u002Fcut-tactics.md",{"path":72,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":73,"sourceHtmlUrl":74},"iteration-examples.md",2366,"https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fkarpathy-loop\u002Fiteration-examples.md",[76,115,138,158,181,207],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":96,"publishedAt":37},"01M71R7K35P668DYEPYGDQP2QA","web-quality-audit","Comprehensive web audit: performance, a11y, SEO, responsive, animation, design tokens.","https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002Fdocs\u002Fciclos\u002Fc28-w1-backup\u002Fweb-quality-audit\u002FSKILL.md",[82,85,88,93],{"categoryNo":83,"categoryName":84,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000015","视觉设计",{"categoryNo":86,"categoryName":87,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000019","营销推广",{"categoryNo":89,"categoryName":90,"taxonomyType":91,"taxonomyTypeDesc":92},"01JBSEEDCAT000000000000032","设计师","PROFESSION","适用职业",{"categoryNo":94,"categoryName":95,"taxonomyType":91,"taxonomyTypeDesc":92},"01JBSEEDCAT000000000000035","市场营销",[97,98,99,100,101,102,103,104,105,106,107,108,28,109,110,111,112,32,113,114],"quality","quality audit","web quality audit","audit","web","Animation","design audit","design","website","ui","responsive","seo","ui audit","Lighthouse","performance","review","a11y","page",{"skillNo":116,"name":117,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":119,"categories":120,"keywords":122,"publishedAt":37},"01M71R7H1RA7CQ568DHRK37PE2","ui-engine","UI system — Grid\u002FFlexbox\u002F@layer\u002F:has(), container queries, compositor-only animation, OKLCH tokens, component patterns","https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002Fdocs\u002Fciclos\u002Fc28-w1-backup\u002Fui-engine\u002FSKILL.md",[121],{"categoryNo":89,"categoryName":90,"taxonomyType":91,"taxonomyTypeDesc":92},[123,124,106,102,125,126,127,114,128,28,129,130,131,104,132,133,134,135,136,137],"ui engine","engine","compositor","css","dark","system","mode","hooks","container","only","query","components","dark mode","grid","queries",{"skillNo":139,"name":140,"nameZh":6,"description":141,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":142,"categories":143,"keywords":146,"publishedAt":37},"01M71R7EPMZFKB833WE4B3GTHC","baseline-ui","Anti-slop UI — layout, typography, responsive, animation, tokens. Use for cleanup or polish.","https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002Fdocs\u002Fciclos\u002Fc28-w1-backup\u002Fbaseline-ui\u002FSKILL.md",[144,145],{"categoryNo":83,"categoryName":84,"taxonomyType":20,"taxonomyTypeDesc":21},{"categoryNo":89,"categoryName":90,"taxonomyType":91,"taxonomyTypeDesc":92},[147,148,106,102,149,150,151,152,28,153,131,104,154,155,156,109,157,133,136,112],"baseline ui","baseline","generic","slop","for","anti-slop","fix","interface","cleanup","design review","use",{"skillNo":159,"name":160,"nameZh":6,"description":161,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":162,"categories":163,"keywords":165,"publishedAt":37},"01M71R7CWK3FBW77Z8WD9GXQFE","skill-creator","Create new AI agent skills following Agent Skills spec -- auto-bootstrap unknown tasks, define structure, register in AGENTS.md","https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002F.archive\u002Fskills\u002Fskill-creator\u002FSKILL.md",[164],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},[166,167,168,169,170,171,172,173,174,175,176,177,178,179,180],"skill creator","creator","create","structure","register","2026","spec","new","ai skill","auto","define","ai agent","unknown","tasks","bootstrap",{"skillNo":182,"name":183,"nameZh":6,"description":184,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":185,"categories":186,"keywords":189,"publishedAt":37},"01M71R78DZM1GA16F1KN8ND926","senior-engineer","Apply senior engineering competencies — system design, trade-off analysis, delegation, mentoring, and tech debt strategy","https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002F.archive\u002Fskills\u002Fsenior-engineer\u002FSKILL.md",[187,188],{"categoryNo":83,"categoryName":84,"taxonomyType":20,"taxonomyTypeDesc":21},{"categoryNo":89,"categoryName":90,"taxonomyType":91,"taxonomyTypeDesc":92},[190,191,192,193,194,195,196,104,197,198,171,199,200,201,128,202,203,204,205,206],"engineer","senior","senior engineer","system design","trade-offs","senior engineering","delegation","senior architect","analysis","debt","architect","tech","strategy","apply","trade","tech debt","engineering",{"skillNo":208,"name":209,"nameZh":6,"description":210,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":211,"categories":212,"keywords":217,"publishedAt":37},"01M71R74WQTXZMG3219RND96CE","prompt-engineering","Design effective prompts using SPEARS framework — scope, principles, examples, assertions, refinements, with ReAct and multi-agent patterns","https:\u002F\u002Fgithub.com\u002FLuisAlbertoMK\u002Fgentleman-agent-gh\u002Fblob\u002Fmain\u002F.archive\u002Fskills\u002Fprompt-engineering\u002FSKILL.md",[213,216],{"categoryNo":214,"categoryName":215,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000009","前端开发",{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},[30,218,206,219,220,104,221,171,222,223,224,225,226,227,228,31,229,230,231,232],"prompt engineering","scope","improve","react","security","patterns","multi-agent","framework","agent patterns","assertions","examples","multi","principles","improve prompt","prompts"]