[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6P8KWZ0QXG0P22P1X0GDFWD":3,"skill-same:01M6P8KWZ0QXG0P22P1X0GDFWD":80},{"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":43,"linkOnly":44,"markdownContent":45,"markdownBody":46,"frontMatter":47,"refName":60,"commitSha":61,"skillDirectory":62,"forksCount":13,"repositoryHtmlUrl":63,"repositoryHomepageUrl":6,"ownerHtmlUrl":64,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":65,"install":66,"resources":79},"01M6P8KWZ0QXG0P22P1X0GDFWD","lookdev-auto",null,"Automated visual tuning: a vision or video model rates rendered variants in a loop. Render several labeled variants into one artifact, ask the model to rate them and suggest better values, render the suggestions, ask it to pick the best, repeat until good — the model is the eye, you run the loop.","duc-duy-1612\u002FOnline-Delivery-Systems-Web","duc-duy-1612","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F192231384?v=4","01M6NSDKQQDKVTHZ8DZMZ1VYY6","01M6NSDKW8DDVTQHBT1HBDDJPT",0,"MIT","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Flookdev-auto\u002FSKILL.md",[17],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000016","图像与多媒体","SCENARIO","使用场景",[23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42],"lookdev","lookdev auto","auto","community","variants","render-loop","vision-model","visual-eval","you","render","visual","gemini","good","ask","eval","pick","model","loop","Video","automation","2026-11-10 10:12:30",false,"---\nname: lookdev-auto\ndescription: \"Automated visual tuning: a vision or video model rates rendered variants in a loop. Render several labeled variants into one artifact, ask the model to rate them and suggest better values, render the suggestions, ask it to pick the best, repeat until good — the model is the eye, you run the loop.\"\nrisk: safe\nsource: community\nsource_type: community\nsource_repo: connerkward\u002Flookdev-auto-skill\ndate_added: \"2026-06-16\"\nauthor: Conner K Ward\nlicense: MIT\ntags:\n  - visual-eval\n  - vision-model\n  - tuning\n  - automation\n  - render-loop\ntools:\n  - claude-code\n  - antigravity\n  - cursor\n  - gemini-cli\n  - codex-cli\n---\n## When to Use\n\nUse whenever \"looks\u002Ffeels right\" is the success criterion and there's no cheap numeric metric — animation easing\u002Ftiming, zoom\u002Fcamera feel, color grade, layout\u002Fspacing, design params, render\u002Fencoder settings, prompt params. Use the automated counterpart to lookdev when there's no human to sit the loop.\n\n_Source: [connerkward\u002Flookdev-auto-skill](https:\u002F\u002Fgithub.com\u002Fconnerkward\u002Flookdev-auto-skill) (MIT)._\n\n# Visual eval loop — let a vision\u002Fvideo model tune what only an eye can judge\n\nWhen the target is \"does this LOOK\u002FFEEL right\" (not a number you can minimize), a\nvision model (image) or video-understanding model (motion\u002Ftiming) can be the judge in\na tight optimize loop. Worked reference: the `screenstudio-alternative` skill (`iteration.py`)\n(tuned zoom-animation feel via `fal-ai\u002Fvideo-understanding`).\n\n## The loop\n\n1. **Render N labeled variants into ONE artifact.** Vary the parameter(s) across a\n   small spread. **Annotate each variant's params ON the artifact** (burn the label in:\n   \"A · 2.2Hz · ζ0.5\"). Images → a labeled grid\u002Fcontact sheet. Video\u002Fmotion → a\n   labeled *sequence* (label card or burned-in overlay before\u002Fover each clip) so the\n   model can compare temporally.\n2. **One model call, structured output.** Send the single artifact with an explicit\n   rubric (define what \"good\" means — and what \"too much\"\u002F\"too little\" look like).\n   Ask for **per-variant ratings + concrete suggested new values as JSON**:\n   `{\"ratings\":{\"A\":n,...},\"best_so_far\":\"X\",\"suggest\":[[p1,p2],...]}`.\n3. **Coarse → fine.** Round 1 = wide spread to locate the region. Round 2 = render the\n   model's suggestions (+ carry the current best) into one artifact; ask it to **pick\n   the single best**. Usually converges in **2 rounds**.\n4. **Stop when sufficient** — best rates high and suggestions cluster. Apply the winner.\n\n## Token \u002F quality \u002F step reductions (do these)\n\n- **One artifact per round, not one call per variant.** The biggest saver — a 6-variant\n  round is 1 upload + 1 inference, not 6. Montage\u002Fgrid beats a loop of single calls.\n- **Burn params onto the artifact.** The model sees label+result together → no separate\n  \"variant A used X\" context to carry → fewer tokens, fewer mistakes.\n- **Structured JSON out + parse.** No re-asking, no free-text wrangling. Prompt \"return\n  ONLY JSON\"; regex the first `{...}`.\n- **Short representative sample.** Tune on a 3-5s clip \u002F one frame \u002F one component, not\n  the whole asset. Cheaper render, smaller upload, faster inference. Apply the found\n  params to the full render once.\n- **Cap variants at ~5-6.** More doesn't improve the model's discrimination and multiplies\n  render + token cost. Wide-but-sparse round 1, narrow round 2.\n- **Calibration anchors.** Include one deliberately-bad and one safe-default variant as\n  fixed anchors each round — gives the model a reference scale and exposes when its\n  \"best\" is worse than the safe default (catch a bad recommendation early).\n- **Independent rubric, stated up front.** Define \"good\" concretely in the prompt\n  (smooth, subtle settle, not bouncy, not sluggish). Don't ask \"which do you like\" —\n  that lets it echo your framing. A held-out criterion keeps the judge honest\n  (see verify-outputs-rule: the check must be independent of what you tuned).\n- **Reuse renders across rounds.** Carry the round-1 winner's clip into round 2 instead\n  of re-rendering it.\n- **Early-exit.** If round-1 top ≥9\u002F10 and the three suggestions are within a small delta,\n  skip round 2.\n- **Cheapest judge that can see the failure.** Frames-through an image VLM can judge\n  spatial things (layout, color, crop); only reach for a true *video* model when the\n  thing being judged is **temporal** (easing, timing, motion smoothness) — those are\n  invisible in stills.\n\n## When NOT to use it\n\n- A real numeric metric exists and correlates with quality → optimize that directly;\n  don't pay a model per step.\n- The judgment is subjective-to-the-user (their taste, brand) → show them the variants\n  and let them pick; a model's \"best\" isn't their best. (This is why the screen-studio\n  spring auto-tune was dropped — the model's pick didn't match the owner's eye.)\n- One or two variants → just look yourself.\n\n## Caveats (learned)\n\n- The model's pick is an *opinion*, not ground truth — anchor it, and sanity-check the\n  winner against the safe default yourself before committing.\n- Vision\u002Fvideo models perceive gross differences well, fine ones poorly — keep variant\n  spacing perceptible; near-identical variants get noise-rated.\n\n## Example\n\n**User request:**\n\n> Use @lookdev-auto for this task: Automated visual tuning: a vision or video model rates rendered variants in a loop.\n\n## Limitations\n\n- Model ratings are probabilistic aesthetic judgments, not objective truth; keep a human review step for brand-critical or subjective work.\n- Automated rounds can become expensive or slow when renders are heavy or many variants are explored.\n- This skill needs screenshots, frames, or clips that expose the quality difference; it is weak for subtle motion, audio, copy nuance, or user-preference calls.\n","## When to Use\n\nUse whenever \"looks\u002Ffeels right\" is the success criterion and there's no cheap numeric metric — animation easing\u002Ftiming, zoom\u002Fcamera feel, color grade, layout\u002Fspacing, design params, render\u002Fencoder settings, prompt params. Use the automated counterpart to lookdev when there's no human to sit the loop.\n\n_Source: [connerkward\u002Flookdev-auto-skill](https:\u002F\u002Fgithub.com\u002Fconnerkward\u002Flookdev-auto-skill) (MIT)._\n\n# Visual eval loop — let a vision\u002Fvideo model tune what only an eye can judge\n\nWhen the target is \"does this LOOK\u002FFEEL right\" (not a number you can minimize), a\nvision model (image) or video-understanding model (motion\u002Ftiming) can be the judge in\na tight optimize loop. Worked reference: the `screenstudio-alternative` skill (`iteration.py`)\n(tuned zoom-animation feel via `fal-ai\u002Fvideo-understanding`).\n\n## The loop\n\n1. **Render N labeled variants into ONE artifact.** Vary the parameter(s) across a\n   small spread. **Annotate each variant's params ON the artifact** (burn the label in:\n   \"A · 2.2Hz · ζ0.5\"). Images → a labeled grid\u002Fcontact sheet. Video\u002Fmotion → a\n   labeled *sequence* (label card or burned-in overlay before\u002Fover each clip) so the\n   model can compare temporally.\n2. **One model call, structured output.** Send the single artifact with an explicit\n   rubric (define what \"good\" means — and what \"too much\"\u002F\"too little\" look like).\n   Ask for **per-variant ratings + concrete suggested new values as JSON**:\n   `{\"ratings\":{\"A\":n,...},\"best_so_far\":\"X\",\"suggest\":[[p1,p2],...]}`.\n3. **Coarse → fine.** Round 1 = wide spread to locate the region. Round 2 = render the\n   model's suggestions (+ carry the current best) into one artifact; ask it to **pick\n   the single best**. Usually converges in **2 rounds**.\n4. **Stop when sufficient** — best rates high and suggestions cluster. Apply the winner.\n\n## Token \u002F quality \u002F step reductions (do these)\n\n- **One artifact per round, not one call per variant.** The biggest saver — a 6-variant\n  round is 1 upload + 1 inference, not 6. Montage\u002Fgrid beats a loop of single calls.\n- **Burn params onto the artifact.** The model sees label+result together → no separate\n  \"variant A used X\" context to carry → fewer tokens, fewer mistakes.\n- **Structured JSON out + parse.** No re-asking, no free-text wrangling. Prompt \"return\n  ONLY JSON\"; regex the first `{...}`.\n- **Short representative sample.** Tune on a 3-5s clip \u002F one frame \u002F one component, not\n  the whole asset. Cheaper render, smaller upload, faster inference. Apply the found\n  params to the full render once.\n- **Cap variants at ~5-6.** More doesn't improve the model's discrimination and multiplies\n  render + token cost. Wide-but-sparse round 1, narrow round 2.\n- **Calibration anchors.** Include one deliberately-bad and one safe-default variant as\n  fixed anchors each round — gives the model a reference scale and exposes when its\n  \"best\" is worse than the safe default (catch a bad recommendation early).\n- **Independent rubric, stated up front.** Define \"good\" concretely in the prompt\n  (smooth, subtle settle, not bouncy, not sluggish). Don't ask \"which do you like\" —\n  that lets it echo your framing. A held-out criterion keeps the judge honest\n  (see verify-outputs-rule: the check must be independent of what you tuned).\n- **Reuse renders across rounds.** Carry the round-1 winner's clip into round 2 instead\n  of re-rendering it.\n- **Early-exit.** If round-1 top ≥9\u002F10 and the three suggestions are within a small delta,\n  skip round 2.\n- **Cheapest judge that can see the failure.** Frames-through an image VLM can judge\n  spatial things (layout, color, crop); only reach for a true *video* model when the\n  thing being judged is **temporal** (easing, timing, motion smoothness) — those are\n  invisible in stills.\n\n## When NOT to use it\n\n- A real numeric metric exists and correlates with quality → optimize that directly;\n  don't pay a model per step.\n- The judgment is subjective-to-the-user (their taste, brand) → show them the variants\n  and let them pick; a model's \"best\" isn't their best. (This is why the screen-studio\n  spring auto-tune was dropped — the model's pick didn't match the owner's eye.)\n- One or two variants → just look yourself.\n\n## Caveats (learned)\n\n- The model's pick is an *opinion*, not ground truth — anchor it, and sanity-check the\n  winner against the safe default yourself before committing.\n- Vision\u002Fvideo models perceive gross differences well, fine ones poorly — keep variant\n  spacing perceptible; near-identical variants get noise-rated.\n\n## Example\n\n**User request:**\n\n> Use @lookdev-auto for this task: Automated visual tuning: a vision or video model rates rendered variants in a loop.\n\n## Limitations\n\n- Model ratings are probabilistic aesthetic judgments, not objective truth; keep a human review step for brand-critical or subjective work.\n- Automated rounds can become expensive or slow when renders are heavy or many variants are explored.\n- This skill needs screenshots, frames, or clips that expose the quality difference; it is weak for subtle motion, audio, copy nuance, or user-preference calls.\n",{"name":5,"risk":48,"tags":49,"tools":51,"author":57,"source":26,"license":14,"date_added":58,"description":7,"source_repo":59,"source_type":26},"safe",[30,29,50,42,28],"tuning",[52,53,54,55,56],"claude-code","antigravity","cursor","gemini-cli","codex-cli","Conner K Ward","2026-06-16","connerkward\u002Flookdev-auto-skill","main","f3312ba749561b6159f477dbfc9313c875d40a56",".agents\u002Fskills_archive\u002Flookdev-auto","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612","2026-10-25 23:50:58",{"source":8,"skillName":5,"commands":67,"prompt":77,"zipAvailable":78},[68,71,74],{"tool":69,"command":70},"npx","npx skills add duc-duy-1612\u002FOnline-Delivery-Systems-Web --skill lookdev-auto",{"tool":72,"command":73},"bunx","bunx skills add duc-duy-1612\u002FOnline-Delivery-Systems-Web --skill lookdev-auto",{"tool":75,"command":76},"pnpm","pnpm dlx skills add duc-duy-1612\u002FOnline-Delivery-Systems-Web --skill lookdev-auto","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Flookdev-auto\u002FSKILL.md",true,[],[81,108,131,155,179,203],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":90,"publishedAt":43},"01M6Q0FA5FZ410Z33Z3TFQ4DBA","zero-trust","Implement zero-trust network architecture. Configure identity-based access, micro-segmentation, and continuous verification. Use when implementing modern security architectures.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fzero-trust\u002FSKILL.md",[87],{"categoryNo":88,"categoryName":89,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000011","运维与部署",[91,92,93,94,95,96,83,97,98,99,100,101,26,60,102,103,104,105,106,107],"ZeRO","trust","zero trust","architecture","assessment","verification","segmentation","devops-security","access","configure","only","active","critical","scope","scripts","relevant","implementing",{"skillNo":109,"name":110,"nameZh":6,"description":111,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":112,"categories":113,"keywords":114,"publishedAt":43},"01M6Q0F1YZJW3KX5FBKEVZ8CV4","zcode-delegate","Delegate coding tasks to the Z.AI ZCode CLI only when the user explicitly requests it, while the orchestrator retains review and landing responsibility.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fzcode-delegate\u002FSKILL.md",[],[115,116,117,118,119,120,121,122,123,39,124,125,126,101,127,26,128,129,103,130],"delegate","zcode","zcode delegate","review","cli","orchestration","tasks","run","Bash","node","read","orchestrator","files","desktop app","app","agent-orchestration",{"skillNo":132,"name":133,"nameZh":6,"description":134,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":135,"categories":136,"keywords":137,"publishedAt":43},"01M6Q0ETK41VGNMYEXW0T00AJN","youtube-transcript","Fetch YouTube transcripts through DeepAPI or local fallback tooling and save clean text output.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fyoutube-transcript\u002FSKILL.md",[],[138,139,140,48,141,142,143,144,145,146,147,148,149,26,60,150,151,152,153,154],"transcript","youtube transcript","youtube","github","codex","research","license","https","text","fetch","tooling","clean","local","output","deepapi","youtube transcripts","blob",{"skillNo":156,"name":157,"nameZh":6,"description":158,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":159,"categories":160,"keywords":170,"publishedAt":43},"01M6Q0EPPH2PVMDW40017F8WX0","youtube-transcript-skills","Fetch YouTube video transcripts, search videos\u002Fchannels, browse channels, and extract playlists via the getyoutubetranscript.com API - free tier, no card required.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fyoutube-transcript-skills\u002FSKILL.md",[161,164,165],{"categoryNo":162,"categoryName":163,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000010","后端与接口",{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},{"categoryNo":166,"categoryName":167,"taxonomyType":168,"taxonomyTypeDesc":169},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",[138,139,140,171,48,34,141,142,144,172,173,41,174,147,175,26,60,176,177,178],"youtube transcript skills","search","api","browse","api-integration","free","card","channels",{"skillNo":180,"name":181,"nameZh":6,"description":182,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":183,"categories":184,"keywords":189,"publishedAt":43},"01M6Q0EAB96PX83W5PTE8A0R8D","youtube-notetaker","Turn YouTube talks into local study notes with slides, transcripts, editable annotations, and a markdown-backed viewer.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fyoutube-notetaker\u002FSKILL.md",[185,188],{"categoryNo":186,"categoryName":187,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000003","演示文稿",{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},[140,190,191,119,48,192,193,194,195,144,41,196,197,198,199,60,200,150,201,202],"youtube notetaker","notetaker","workflow","official","notes","code","markdown","slides","annotations","README","study","dair-academy","viewer",{"skillNo":204,"name":205,"nameZh":6,"description":206,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":207,"categories":208,"keywords":210,"publishedAt":43},"01M6Q0E55F0GW5MX48V8HXS8B5","youtube-full","Fetch YouTube transcripts, search videos, browse channels, and extract playlists via TranscriptAPI — no yt-dlp, no Google API key, works from any cloud server.","https:\u002F\u002Fgithub.com\u002Fduc-duy-1612\u002FOnline-Delivery-Systems-Web\u002Fblob\u002Fmain\u002F.agents\u002Fskills_archive\u002Fyoutube-full\u002FSKILL.md",[209],{"categoryNo":162,"categoryName":163,"taxonomyType":20,"taxonomyTypeDesc":21},[211,140,212,213,48,34,172,214,215,173,41,174,147,216,175,26,60,176,217,153],"youtube full","full","google","server","dlp","credits","automatic"]